Program Listings

The Jagiellonian-PET (J-PET) collaboration is developing a novel TOF-PET tomography scanner based mainly on the timing of signals instead of their amplitudes for the reconstruction of Lines-of-Response, therefore a very precise time resolution is one of the main challenges of the project. The novelty of the concept lies in employing long strips of polymer scintillators instead of crystals as detectors of the annihilation quanta. The diagnostic chamber consists of plastic scintillator strips readout by pairs of photomultipliers arranged axially around a cylindrical surface. To take advantage of the superior timing properties of plastic scintillators, the signals are sampled in the voltage domain with an accuracy of 20 ps by novel ultra-fast electronics, and the data are collected by the FPGA-based trigger-less data acquisition system. The hit-position and hit-time are reconstructed by the dedicated reconstruction methods based on the compressing sensing theory and a library of synchronized model signals. The solutions are subject to twelve patent applications. So far, a time-of-flight resolution of ~125 ps (sigma) was achieved for a double-strip prototype with 30 cm field-of-view (FOV). It is by more than a factor of two better than the TOF resolution achievable in current TOF-PET modalities and at the same time, the FOV of 30 cm long prototype is significantly larger with respect to typical commercial PET devices. The axial geometry of the J-PET scanner possesses also another advantage. Its diagnostic chamber is free of any electronic devices and magnetic materials, thus it gives unique possibilities of combining J-PET with Computed Tomography or with Magnetic Resonance Imaging, allowing to perform the simultaneous scan of the patient with both methods. A general concept along with the results of the performance tests of the TOF-PET system will be presented.

Current high-resolution scintillators are fabricated using mechanical pixelation technique. However the fabrication cost of finely pitched scintillator arrays together with their lack of flexibility to accommodate new depth of interaction designs has remained a significant issue with mechanical pixelation. Another pitfall of mechanically pixelated scintillators is their relatively large inter-pixel gaps that adversely affects their sensitivity to the incident gamma-ray. The main objective of this project is to fabricate high-spatial resolution and high sensitivity PET detectors with depth of interaction (DOI) capability and single-side readout in a cost-effective manner using laser-induced optical barriers (LIOB) technique. We have created optical barriers with different size and barrier density in LYSO:Ce at various depths up to 20 mm to form pixel-like shapes similar to mechanically pixelated crystals. In order to process 20mm thick crystals we corrected for laser beam defocusing effect and its adverse effect on laser energy density which results in smaller barrier size and reflectivity. The fabrication time for 10x10x1 and 10x10x20 mm3 LYSO crystals (with 1mm pixels) was ~8 and 95 minutes respectively. Our results show that laser-processed crystals with doubled barrier density has an average of 20% improvement in pixel separation especially close the detector edge. Individual pixels in two 1-mm thick crystal stacks can be resolved in the flood image based on their distance from the SiPM array, realizing DOI measurement. An average peak-to-valley ratio of 2.2 was measured for LYSO detector with 1x1x20mm3 pixels. The work presented here is an enabling technology for high-resolution depth-of-interaction PET scanner.

Developing PET detectors with combined time-of-flight (TOF) and depth-of-interaction (DOI) encoding capabilities is a crucial step in the successful implementation of the ultra-high sensitivity EXPLORER scanner, a 2 meter axial length PET scanner capable of imaging the entire patient volume. We previously evaluated 4 x 4 arrays of phosphor-coated LYSO crystals with conventional block detector readout, and showed that both TOF and DOI capabilities were achieved with a suitable phosphor coating. This work extends phosphor-coated crystals to a complete block detector with emphasis on light guide fabrication and developing maximum likelihood (ML)-based DOI estimation. We built a complete block detector, including assembling crystal arrays with Teflon tape (optimal for phosphor-coated crystals), and fabricated custom light guides by cutting 0.13 mm wide slots of varying depth in an acrylic block and filling with ESR film. With the phosphor-coated array, we measured an energy resolution of 13.5% and a timing resolution of 442 ps for the central 5 x 5 crystals as the outer rows could not be resolved with our current light guides. We also evaluated a comparable non-phosphor-coated LYSO array for which nearly all crystals were resolved and the energy resolution was 11.3% for the central crystals. There were no significant differences in timing resolution between the two arrays. Lastly, we developed two ML pulse shape discrimination models to estimate DOI with phosphor-coated crystals. Using simulated events with PMT readout, the ML method provides 35% more precise DOI estimation than the delayed charge integration (DCI) method previously used. Continuing this work, we will optimize the light guide design and investigate alternative fabrication methods as well as further develop the ML model to estimate all three gamma interaction position coordinates in the detector along with the interaction time and energy and apply the model to real detector data.

We are developing a scintillator-based imaging detector using strontium iodide (SrI2) with the objective of evaluating its potential for SPECT imaging applications. The very high light output of SrI2 (>85 photons/keV compared to 38 photons/keV for NaI) and its favorable growth properties make it an attractive option for a SPECT detector, especially when coupled to compact, high-efficiency, spectrally matched photodetectors like silicon photomultipliers (SiPMs). Furthermore, the effective atomic number of SrI2 (Zeff=49) is similar to NaI (Zeff=50), while offering a higher density (4.59 g/cm3 vs. 3.67 g/cm3). A FWHM energy resolution of 5.4% at 122 keV has been demonstrated for a single crystal coupled to a PMT, while single pixel measurements at 662 keV have shown slightly better energy resolution with a SiPM compared to a PMT (3.4% vs. 3.7%). Improvements in energy resolution over NaI would allow for use of narrower energy windows in imaging, reducing scatter. Our initial detector is a 4x4 array of individual SrI2 crystals coupled 1-to-1 to a SensL SiPM pixel array. Characterization measurements to be reported include energy resolution, spatial resolution, uniformity, and detection efficiency.

LSO and LYSO scintillators have become the preferred choice for PET scanners, thanks to their high effective Z, light yield, relatively good energy resolution and good timing properties which allow for coincidence resolving times of the order of 250 ps (FWHM) at the annihilation peak. However, very recently additional inorganic scintillators are becoming available, such as PrLuAG and HRCeGAG which hold a very high potential for PET application, either in single-scintillator arrangements or forming part of phoswich arrays with DOI capability. This work evaluates these scintillators in all relevant aspects for PET, such as energy resolution, coincidence time-resolving time, stopping power, photopeak to Compton ratio, internal activity, internal activity and optical compatibility for phoswich arrangements. As these parameters highly depend on the matching of the emission of the crystal to the photodetector employed, we evaluate these performance parameters with scintillators coupled to standard PMTs as well as to a range of SiPM currently available, from Hamamatsu, SensL, KeteK, Advansis and Excelitas.

The imaging performance of a SPECT system strongly depends on the spatial resolution and the absence of artefacts in the projection images fed to the 3D algorithm used in image reconstruction. These characteristics in turn, depend both on the materials used for the detector and on the algorithm used to calculate the interaction point of the gamma photons with the scintillating crystal. While the materials used strongly affect the cost of the device, the careful choice of the positioning algorithm can give substantial improvements without affecting the device cost. Many solutions have been proposed to find the interaction point of a gamma photon with a scintillating crystal and it is well known that the MLE algorithm is the best performing one. In the current implementation, its only drawback is the need of a complex calibration procedure that is time consuming and computational expensive. Moreover, the calibration must be performed every time that a change is made to the setup of the detector. The need for calibration is given by the need for a set of expected anodic values, function of the scintillation position, to use in the ML estimate. In this work we propose an alternative solution that pre-calculates the expected anodic values using a theoretical model of the scintillation light distribution. This new approach was used to obtain the planar images in a SPECT acquisition made with a gamma camera based on a continuous NaI(Tl) crystal coupled with a 2x2 Hamamatsu HS8500 PSPMT array and fitted with a multi-pinhole (MPH) collimator. A 3D ML-EM pinhole image reconstruction algorithm was applied to the MPH projection data to generate the 3D reconstructed image. A comparison between the 3D reconstructed images from this new method and from our previous approach, based on a least square fit procedure of the anode distribution with a theoretical light distribution, shows an improved spatial resolution and much reduced image artefacts.

Objectives: We have previously reported on depth of interaction (DOI) positioning performance of 2x2 discrete LYSO crystal arrays coupled to digital silicon photomultiplier arrays (PDPC arrays). By optimizing the design of the interface between each neighboring crystal, DOI can be estimated from the overall light distribution collected by the 2x2 array of sensors. A significant challenge to the design is controlling how light is shared between neighboring crystals. We have previously used crystals with lapped surfaces to improve light sharing. In this work we investigate using etched LYSO crystals to improve light sharing. Methods: We investigated the DOI postioning performance for two 2x2 LYSO discete cystal arrays constructed from crystals that either had a finely roughened surface finish or a coarsely roughened surface finish. The LYSO crystals were each 4x4x19 mm3. High index of refraction (i.e., 1.704n) melt mount resin and custom shaped mirror film material were used to control how light was shared between neighboring crystals. The detector unit was calibrated using an electronically collimated 511 keV photon flux stepped along the long axis of the crystal. DOI event positioning was done using maximum likelihood. Results: The crystal array fabricated from the crystals with the finely etched surfaces had better DOI positioning performance than the array constructed with the coarsely etched surfaces. It was also determined that using the signals from the two adjacent crystals to the ‘hit’ crystal provided better DOI estimation than using all four signals. Conclusions: Customizing the surface properties to optimize light sharing can improve DOI positioning for detector designs that utilize light sharing to estimate DOI positioning. The advantage of the etching process used in this work is that it is highly reproducible and it also allows for custom patterns to be etched onto crystal surfaces. .

Objectives: We investigate the imaging characteristics of a low cost, discrete crystal BGO detector module for dedicated PET brain imaging studies. The low cost design is enabled by using BGO crystals and position sensitive sparse sensor (PS3) arrays. PS3 arrays are low cost because they significantly reduce the number of silicon photomultiplier (SiPM) sensor elements needed to decode the crystal of interaction from a large surface area PET detector. Methods: We investigated the performance of a 14x14 crystal array readout by a 16 element PS3 array. The crystal array consisted of 2x2x10 mm3 discrete BGO crystals. Two arrays were fabricated. For one array, the BGO crystals were polished on five sides and one 2x2 mm2 side lapped with 1000 grit material. For the second crystal array, five of the BGO sides were lapped and one 2x2mm2 side polished. Mirror film material was placed between the discrete crystal elements to optically isolate them and help with light collection efficiency. The crystal arrays were coupled to the PS3 array via a custom light guide to help channel the light toward the SiPM sensor elements. For the BGO array consisting of polished crystals a thin diffusion film was placed between the crystal array and the light guide to broaden the light spread to the sensor array. The PS3 array consisted of sixteen 3x3 mm2 SiPM sensor elements covering a 28x28 mm2 detector area. Crystal decoding characteristics, enery resolution and timing resolution were measured for each of the detector modules. Results: Simulation results indicate outstanding performance using a 16 sensor MPPC PS3 array coupled to a 14x14 discrete crystal BGO detector. Experimental results will be provided at the meeting. Conclusions: Using SiPM based PS3 arrays is a cost effective method to fabricate high spatial resolution PET detectors for human brain imaging. This design methodology can also be applied to other organ specific PET imaging systems (e.g., breast or extremities).

The OpenPET data acquisition system for nuclear imaging has recently become available. This system is modular in design with a VME crate and a backplane that can currently support up to eight detector digitizer boards (DBs). In the current implementation, each DB has 16 analog to digital converter (ADC) channels operating at 40 MSPS. The initial firmware is released with the OpenPET system that supports capture of data in singles-mode only using ï¿½oscilloscope modeï¿½, in which the entire digitized waveform is sent to the host PC via the QuickUSB 2.0 interface module. Use of oscilloscope mode creates a large data packet size (~2.1 kB/event for 32 ADC samples/channel), which limits the singles event rate to ~18 kcps for 32 ADC samples/channel/event. In this work, we present our initial experience with the OpenPET system for acquiring data from 16 LYSO/SiPM based detector modules used in a recently constructed PET/MRI imaging system. Data acquired from OpenPET are compared with traditional nuclear instrument module (NIM) based acquisitions. We also describe our modifications to the OpenPET firmware which reduced the data packet size to 140 bytes/event and allowing imaging at up to 280 kcps. The results of the comparison of the OpenPET and NIM acquired data showed that the detector flood image quality is similar and that the average energy resolution is slightly better with NIM hardware (12.5% vs. 12.0%). The coincidence resolving time for OpenPET acquisitions was measured to be approximately 80 ns full width at half maximum, limited by the current lack of firmware to support both the hardware fast timing trigger and a global reference timing clock. Despite these limitations, the OpenPET system was successfully used to acquire PET imaging data from all 16 detectors in our PET insert system, allowing imaging of both phantoms and mice.

In this work we present the initial performance evaluation testing of a mini gamma camera based on a lead parallel hole (hexagonal) collimator and a monolithic NaI(Tl) crystal coupled to a Hamamatsu Photonics H8500 position-sensitive photomultiplier tube (PS-PMT). Uniformity was evaluated using a square flood source filled with ^99mTc-water solution. High non-uniformities at the detector edges, possibly due to internal light reflections in the scintillation crystal, were reduced by implementing an energy window ?E=[110,170] keV. The flood source-image was used to correct for non-uniformities in subsequent acquisitions. Energy resolution was computed for different regions of the flood-source image with an average value of 14.6% for 140.5 keV photons. Linear distortion was obtained using a collimated ^99mTc point-source moved along the detector field-of-view to form a square grid. System resolution was evaluated with a line-source, which consisted of a glass capillary tube (1.1 mm inner diameter, 7 mm long) containing 4.5 MBq of ^99mTc-water solution, as a function of source-collimator distance. Profiles across the line-source images were taken from which the FWHM and FWTM were extracted to quantify system spatial resolution. Profiles showed long tails which could be explained by photon septal penetration. For the line-source placed in contact with the collimator, a system spatial resolution of 2.8 mm was obtained. Future work will include distortion corrections, measurements of detector sensitivity, intrinsic resolution, the use of local energy windows and the implementation of position estimation using B-spline parameterization.

We are developing a 3D positioning positron emission tomography (PET) detector module made up of dense stacks of 0.5 mm x 0.5 mm x 1.0 mm LYSO scintillation crystal elements coupled to position sensitive solid state photomultipliers (PS-SSPMs.) In this work, crystal arrays with four different types of outer face reflectors were tested to investigate the effect of outer reflectors on crystal identification near the edges of PS-SSPMs. To quantify the quality of the flood maps, the minimum (QI1) and the average (QI2) of the distances between adjacent crystals divided by the standard deviation of each peak near edges in the flood histograms were calculated. We found that crystal arrays with both specular reflectors and diffuse reflectors attached to the edges and the top-face of crystal array showed the best crystal separation (QI1 = 3.0601, QI2 = 4.5238), and crystal arrays with only diffuse reflectors attached to the edges and the top-face of crystal array also showed comparable results (QI1 =2.6897 , QI2 = 4.4549.) The quality of flood maps was poor when only specular reflectors were attached to the top face of the crystal arrays and diffuse reflectors or diffuse reflectors and specular reflectors on the edges (QI1 =1.6706, QI2 = 3.7501 and QI1 =1.7945, QI2 = 4.1964, respectively.) This work significantly improved identification of 0.5 mm crystals close to the edges of PS-SSPMs.

In positron emission tomography (PET), the length of a scintillator should be long enough to achieve high-sensitivity. The spatial resolutions achieved with the long scintillators, however, degrade with increasing radial offset from the center of field of view. This is called parallax errors and can be minimized by measuring depth of interaction (DOI). In this study, a high spatial resolution DOI encoding PET detector was constructed using a digital SiPM (dSiPM, DPC-3200-22-44, Philips) with dual layer pixelated scintillator arrays shifted from each other by one half a scintillator pitch in both X and Y directions.
The DOI encoding PET detector was designed by using 4 x 4 pixels of dSiPM coupled with dual-layer scintillator blocks consisting of a 12×12 LYSO array on a 13×13 LYSO array and a glass plate. The pixel size of each LYSO was 0.9 x 0.9 x 6 mm³. The glass plate was inserted between bottom scintillator layer and the surface of dSiPM for minimization of overlapping of each scintillator position in the 2D flood histogram. A Monte Carlo simulation was also conducted to optimize the dual-layer scintillator blocks with different thicknesses of the glass plate ranging from 0.5 to 3 mm with 0.5 mm steps. The experientially measured 2D flood histogram was acquired with a 1 mm thick glass plate. The simulation results showed that 1 mm thick glass plate was optimal to distinguish layers without overlapping of flood position and that the light loss rate of this scintillator block was ~25%. The average energy resolution and the average peak-to-valley ratio were 16.4% and 4.5, respectively. DOI in the fine spatial resolution scintillator structure can be discriminated by flood position with the offset scintillator arrangement. This study demonstrated that dual-layer scintillator blocks coupled with dSiPM having low electrical noise could provide DOI information along with sub-millimeter spatial resolution capability.

We designed a depth-encoding PET detector using the bottom & side readout method with the cross-arranged scintillator array. To evaluate the characteristics of the novel detector module, we used the DETECT2000 simulation tool performing the optical photon transport in the crystal array. The simulated results were validated by the experiments. The detector module is made up of N x M cross-arranged crystal array. The cross-arranged crystal array is composed of N/3 x M sub-arrays consisting of 3 x M crystals. The second row of the sub-array is perpendicularly arranged to the first and third rows. The crystals are optically coupled to crystals in other rows, however, the surfaces between crystals in the same row are treated as the reflector. A 6 x 5 crystal array consisting of two sub-arrays was considered to demonstrate the proof of concept. The two SiPM arrays are coupled to the bottom and one side of the crystal array, respectively. The x-y position is determined by the bottom SiPM array and the side SiPM array gives the depth information. All pixels were distinguished and about 2 mm DOI resolution was achieved in the experiments as well as the simulations. Keywords: cross-arranged scintillator array, bottom & side readout, DETECT2000, Depth of Interaction

Positron Emission Tomography (PET) is based on the detection of annihilation photons through their interactions with scintillator crystals. Traditionally, detection of scintillation light is achieved with multi-anode photomultipliers. Due to mismatch between the base area of photodetectors and scintillator arrays crystals, sharing of light among the detector pixels is necessary. In our detector head the sharing of light is implemented via a light guide placed in between the BGO array and the PMT. However, the PMT pixels at the edge of the detector still cannot share light in all directions. Consequently, resolving the BGO crystals placed on the edges and corners of the photodetector is difficult. To address this issue, with the goal of designing a high spatial resolution preclinical PET, we examined the performance of different detector pixel binning configurations for a BGO crystal array, coupled to photomultipliers through a continuous 1mm light guide (glass slab). Simulations were performed by a Monte Carlo based software (GATE) and validated by measurements on a similar set up. Three different pixel binning configurations were investigated: one in which the detector area was divided into 3x3mm pixels, a second that was created with 6x6mm pixels, and the last configuration included 3x3mm pixels placed at the edges of the array and 6x6mm PMT pixels everywhere else. Flood maps of the crystal array as well as the average line profiles (through rows and columns) were obtained from simulations and measurements. Preliminary results from this study indicate an advantage of using smaller pixel area like our 3x3mm sensors over the other two configurations. However, in this case the cost of PMT pixels increases for smaller base areas. Therefore, using our last configuration might be a feasible solution since edge BGO crystals were better resolved using this configuration versus the 6x6mm pixel configuration.

Tracking systems are used in many medical scenarios for the localization of the patient or of devices, for example in intra-operative nuclear imaging. The optical infrared tracking system commonly used for this purpose is outside-in and requires free line of sight between a camera and the tracked object, a distance that can reach several meters. To alleviate issues due to this requirement, we present an inside-out tracking approach where a video camera is attached to a hand-held mini gamma camera to provide combined activity and tracking information to reconstruct 3D SPECT-like images in a compact and flexible setup suitable for interventions. The radioactivity distribution is presented as an augmented reality overlay of the image of the video camera. We show results on phantoms mimicking common nuclear medicine procedures, such as the sentinel lymph node biopsy in thyroid and breast cancer. Distances between reconstructed hot spots differ less than 3% from the ground truth distances. We also show a first clinical images acquired on a breast cancer patient that show a 'correct' visualization substantiated by the physician's diagnosis. The tracking performance is comparable to the standard outside-in approach, while easing the interventional procedure in terms of a much easier and shorter setup as well as a substantial reduction of the line of sight requirements. This work significantly simplifies and reduces costs in a multitude of 3D nuclear image guided procedures.

As our knowledge, in spite of intensive reserch effort being spent, none of groups reported truly simultaneous SPECT/MRI imaging with sub 500µm millimeter resolution, which is critical for preclinical application. In this report, we are going to present first 500µm millimeter resolution simultaneous SPECT/MRI imaging, by using MRC-SPECT-I inserted in Siemens 3T Trio-I scanner. We will study strong magnetic distortion charge collection process of semiconductor detector, explore potential correction models and evaluate its impact over SPECT imaging. To explore application value of MRC-SPECT-I system, we carried on neural stem cells(NSC) tracking studies in mice brain. Due to high sensitivity of MRC-SPECT-I, our preliminary study showed that MRC-SPECT-I without MRI could track as few as 400NSCs tagged with MSN-111In. We will carry similar study using simultaneous SPECT/MRI.

We build up a prototype system of Inverted Compound Eye (ICE) Camera for Second Generation of MRI Compatible SPECT Imaging, which aims at experimentally evaluating the capability of the ICE system with a large number of independent micro-camera-elements closely packed in a dense 2-D array. This work depends on the simulation result which shows the ICE system has 1.5% peak geometry sensitivity while the resolution could achieve sub millimeter in side 1cm FOV. The system consist a CdTe hybrid pixel detector that offers an active area of 2.56 cm × 2.56 cm, divided into 256× 256 square pixels with a pixel size of 0.1 mm 0.1 mm. The CdTe detector is coupled to a dedicated collimation aperture that has an array (say 8× 8) of micro pinhole. 408 parameters are used to module the system using analytical an model, incorporating a comprehensive system model that includes the effect of DOI, and the photon penetration through the complex pinhole geometry using the SPECT simulation package that we had previously developed. By means of minimize the error of projections’ weighted center between the simulation and experiment results, we get the system geometry information. The fitting result shows that the mean square error between the experiment and simulated projections is around 0.030 mm.

The feasibility of the VP-PET approach has been established, and we are now extending this principle to a more modular and flexible design by using flat panel geometry. Such geometry can be easily positioned against any region of interest to be imaged with high resolution while maintaining whole body imaging capability and can be particularly beneficial in tumor characterization in early cancer diagnosis. A flat panel PET insert system is currently being developed to meet some organ specific imaging applications with the image quality equivalent to some dedicated PET scanners. This insert works in conjunction with a whole body PET scanner to provide whole body images with enhanced sensitivity and spatial resolution in the area close to the insert. The completed panel detector consists of 32 modular detectors arranged in an 8 x 4 array. Each detector module consists of a 4 x 4 silicon photo-multiplier (Hamamatsu S12641-PA-50(x)) array coupled to a 16 x 16 LYSO crystal array (1.0 x 1.0 x 3 mm³) via a custom light guide. The panel has an active sensing area of approximately 13.4 x 6.6 cm². These detector modules demonstrate good performance in terms of crystal separation, energy (10.19+/-0.68%) and timing resolution (0.95 +/- 0.08 ns). The flat panel detector has been integrated into a Siemens Biograph 40 whole body PET-CT scanner with modified readout electronics and coincidence logic to acquire coincidence events. An image reconstruction framework previously developed to model arbitrary geometry is used for joint image reconstruction using both scanner and insert data. System characteristics and improvement in image resolution will be presented based on phantom studies.

Time-of-flight (ToF) capabilities with depth-of-interaction (DOI) PET detector are known to improve image quality while reducing spatial resolution degradation in the peripheral area of PET systems. In this research, we present results on a single-end detector with TOF-DOI capabilities made up of arrays of discrete scintillating crystal elements and read out with digital silicon photomultiplier (dSiPM). Two scintillator array with 3 × 3 and 4 × 4 elements were stacked such that the 3 × 3 array was offset by one-crystal size in both the x and y direction to form a 20 mm thick stacked offset detector. Each crystal element was 3.2 mm × 3.87 mm × 10 mm, polished and wrapped with ESR reflector. The dSiPM sensor was maintained at around 15 °C for all experiments. The flood map, with Na-22 source, showed distinct crystal peak with average peak to valley (P/V) ratio of 30.8 and 18.8 for the top and bottom layer respectively. The average Gaussian fitted energy resolutions of the 511 keV peaks were 14.8% for the top layer and 23.1% FWHM for the bottom layer. With a 3 mm × 3 mm × 5 mm LYSO reference detector, a coincidence timing resolution of 241 ps FHWM was attained. This value is expected to improve with better skew correction and energy gating. From the well distinct flood histogram, we can infer that there is room for smaller pixelated crystals to be implemented. This simple single-ended DOI-TOF detector configuration may be an alternative to thick monolithic scintillators or dual-ended configurations in achieving DOI and ToF capabilities simultaneously.

Recently we presented a monolithic scintillator detector based on a 32 mm × 32 mm × 22 mm LYSO:Ce crystal and a digital photon counter (DPC) array, which achieved an (x,y) spatial resolution of < 2 mm FWHM, a DOI resolution of ~5 mm FWHM, and a CRT of ~225 ps. Thus, monolithic detectors appear to be competitive with current state-of-the-art clinical TOF-PET detectors based on pixelated crystals matrices readout by either photomultiplier tubes (PMTs) or (digital or analog) silicon photomultipliers (SiPMs). As a proof-of-concept of the 3D imaging performance of a complete ring of monolithic scintillator detectors, we built a setup that emulates a full clinical TOF-PET scanner using only two detector modules. The modules are made of 2 x 2 detectors, each detector consisting of a 32 mm x 32 mm x 22 mm LYSO crystal and a DPC array. The setup comprises two rotating arms that each carry a detector module, and a central rotating phantom table. The fully automated setup can be used to sequentially acquire all possible lines of response (LORs) of a complete detector ring, using a step-and-shoot acquisition approach. A comprehensive characterization of the spatial- and time resolution of the emulated ring of monolithic scintillator detectors will be presented. Specifically, the reconstructed spatial resolution will be quantified using point sources at different radii, up to 150 mm. Moreover, a Derenzo resolution phantom will be scanned to determine the imaging capability of the emulated TOF-PET system.

The first all-digital clinic Position Emission Tomography (PET) with an flexible scalable architecture and modularized components has been implemented by Digital-PET Technology (DPET) and Huazhong University of Science and Technology (HUST). It has a very flexible and extensible architecture for easily setting up different scale PET systems with different number of detectors to meet different applications. It consists of several modularized components such as the Plug-and-Play (PnP) PET detectors, the scalable global clock matrix, the auto calibration subsystem, the flexible smart gantry, and the others. All of these modularized components/modules with standard net interfaces are all connected by Ethernet and joined up into two net chains for monitor/control and data transmitting.

Spatial resolution of conventional SPECT system is poor (approximately 10 mm). Recently, high-spatial resolution SPECT systems employing solid-state detectors are developed but expensive. We are developing brain SPECT system with excellent spatial resolution less than 5 mm like PET system. In this study we developed a full-digital gamma camera with position-sensitive photomultiplier tube (PMT) to achieve high spatial resolution and constructed SPECT system using it. In addition, we evaluated the performance of the developed gamma camera and SPECT system. The gamma camera for brain SPECT consists of NaI plate scintillator of 25 cm ×15 cm coupled to a 5×3 array of a 2-inch position-sensitive PMTs with 8×8 multiple anodes (Hamamatsu H8500) and full analog-to-digital converter. Detection position was calculated by Anger logic software. A parallel-hole Pb collimator with 0.8-mm holes and 50-mm length was made for high-resolution imaging. We measured detector intrinsic resolution and sensitivity with Tc-99m solution, and evaluated the feasibility of high resolution SPECT imaging by physical experiments using multiple line-source phantom and 3-dimensional brain phantom. SPECT data were reconstructed by filtered back-projection method with Chang attenuation correction. From results of experiments, intrinsic spatial resolution of the detector was approximately 1.5 mm. The detector sensitivity was low, approximately 10.7% of that of conventional SPECT Toshiba GCA7200A with LEHR collimator. The phantom experiments showed that our system could provide SPECT image with spatial resolution of approximately 3.5 mm at 150-mm radius of rotation for brain imaging. In conclusion, this study suggested that our system with the developed high-resolution gamma camera could provide SPECT image with spatial resolution less than 5 mm like PET system. In near future the reconstruction software to improve S/N of SPECT image will be studied to overcome low sensitivity.

The challenge of neurodegenerative disease has spurred the development of various positron emitting radiotracers for PET imaging applications. The necessity of being able to effectively extract information from these images has shown that PET, a high sensitivity modality, is in need of further refinement of its spatial resolution if animal models are to be used. Our work focuses on the development and implementation of Hybrid Pixel-Waveform (HPWF) capable CdTe detectors for use in PET imaging applications. Preliminary results have indicated that the system is potentially capable of sub-300 micron resolution from a reconstructed point source image. We will verify the system's performance by imaging Zr-89 filled resolution phantom and identify the viability of the system for imaging transgenic mice with neurodegenerative diseases by imaging a mouse brain phantom.

We have developed a prototype positron emission tomography (PET) scanner intended for breast imaging. The scanner uses 24 high resolution depth of interaction encoding detectors arranged in a single ring. It has a bore diameter of 198mm. The detector consists of an RMD, Inc. avalanche photodiode (APD) and a Hamamatsu C12 position sensitive photo-multiplier tube (PSPMT) coupled to either end of a 14×14 element array of unpolished 1.5mm pitch, 20mm thick lutetium yttrium orthosilicate (LYSO) scintillation crystals separated by Toray diffuse reflector. Cardinal electronics suite (CES), a high performance data acquisition system designed for PET by Siemens is used for data acquisition.The 24 detectors are assigned to 6 coincidence groups each having four detectors.We modified CES to provide four bits of DOI information which is used to bin coincident events to four DOI bins. Operational firmware of the scanner is fully tested for a single detector and is currently undergoing scale-up for the full ring. Full width at half maximum (FWHM) energy resolution obtained for the single detector is 23.8%. For illuminations at the center of DOI bins, the probability of correctly assigning the DOI bin is 0.74.

Multiplexing many SiPMs to a single readout channel is an attractive option to reduce the readout complexity of high perfromance time of flight (TOF) PET systems. However, the additional dark counts and shaping from each SiPM cause significant baseline fluctuations in the output waveform, degrading timing measurements using a leading edge threshold. This work proposes a simple analog filtering network to reduce the baseline fluctuations in highly multiplexed SiPM readouts. With 16 SiPMs multiplexed, the FWHM coincident timing resolution for single 3mm x 3mm x 20mm LYSO crystals was improved from 401 +/- 4ps to 248 +/- 5ps. With 4 SiPMs multiplexed, using a 20mm length, 2 layer DOI array of LYSO crystals the mean time resolution was improved from 277 +/- 6ps to 217 +/-4ps using a ADCMP572 comparator for timing pickoff. All experiments were performed at room temperature with no active temperature regulation. This results show a promising technique for the construction of high performance multiplexed TOF PET readout systems with simple analog leading edge timing pickoff.

We have developed a SPECT gamma photon ray–tracing method based on the SolidWorksTM stereolithography (.STL) file format that models penetration, scatter, and fluorescence in the design work flow[1]. With this powerful tool, now we are able to model various kinds of novel SPECT systems including those with moving collimators. In this work, the configuration of a rotating-slit aperture in combination with a 1D or 2D detector is revisited, and the method of virtual pinhole synthesis is discussed. Our preliminary simulation results indicate an increasing advantage with decreasing aperture size due to the relatively larger sensitivity of slit apertures compared to real pinhole apertures.

Positron emission tomography (PET) is a valuable clinical tool for cancer staging and monitoring response to treatment, but it is limited by its relatively poor resolution and high noise level. In this study, we investigated the impact of using anatomical information, either during or post reconstruction, for PET imaging of sub-cm lung nodules. The XCAT phantom with four inserted lung lesions (diameter: 6-12 mm, contrast: 4:1) was used to simulate twenty 3-minute [18F]-FDG PET thorax scans on a Siemens mCT. Maximum a posteriori reconstruction with an anatomical prior was compared to ordered-subsets expectation-maximization with resolution recovery (with and without the use of time-of-flight (TOF) information). Furthermore, an anatomy-guided partial volume correction (PVC) technique was applied to each of these reconstructions. Adding TOF information improved both the convergence and the contrast-to-noise ratio (CNR). The anatomical prior further improved the CNR at the cost of a systematic bias. The template-based PVC method improved the quantitative accuracy for all reconstructions, within a clinically acceptable number of iterations. By combining the anatomical prior with the anatomy-guided PVC technique, quantitatively accurate, high-contrast PET imaging could be achieved even for the smallest lung nodule (6 mm), provided that the nodule was visible on CT and accurate respiratory motion correction has been performed.

The majority of all modern PET cameras use LSO and LYSO crystals which contain a small proportion of the radioactive isotope of Lutetium. The aim of this study is to obtain attenuation maps for PET/MR scanners based on the background activity of the LSO scintillator. In particular, attenuation maps for the TX/RX coils in the PET Field of View have to be provided. To ensure that an attenuation map is possible to be reconstructed the stability of the scanner was checked in temperature and background activity. First image reconstructions of the Head coil obtained with LSO background prove that the method is feasible as long as the overall scan time is sufficiently long.

The sensitivity of a PET scanner is a function of the solid angle of coverage from the scintillation material to the object being scanned and the stopping power of the scintillation material being used. PET scanners also operate with an energy window centered on the 511 keV annihilation photons, wide enough to account for the energy resolution of the scanner’s scintillator. The energy window reduces the amount of object scatter accepted, but also eliminates any detector scatter events that do not experience object scatter. By adding lower energy windows, an unscattered, full energy deposition event can be put in coincidence with a lower energy event, creating a high angle scatter line of response, or a detector scattered event that represents an object unscattered line of response. Using a normalization specifically calculated for the coincidences with the new, lower energy windows and scatter correction of the high angle object scattered events, the data is reconstructed separately from the simultaneously collected PET data acquired with the traditional energy windows. Initial images show that the detector scattered events can be reconstructed with similar results to the images reconstructed from the standard energy window.

PET/CT is an important tool for clinical studies and pre-clinical research by combing functional and anatomical registration images. To achieve high quality fused results, geometrical aligning of PET and CT scanner is a critical procedure. Common methods are using positron source phantoms imaged both by PET and CT scanner. In this paper, a novel PET/CT geometrical alignment method was developed. Firstly, a multi-tungsten-alloy-ball phantom without position source was designed and imaged by CT and the PET scanner using coincidence event of Lutetium-176 intrinsic radiation included in LYSO. Secondly, the centroids of the spheres are derived from the reconstructed images. Lastly, the rotation matrix and the translation vector were calculated by least-square fitting the centroid data. Based on an animal PET/CT system (InliView3000) developed in our lab, simulation and experimental studies were carried out, and they proved the feasibility of the proposed method. The simulation study validates that 2h data of LYSO background image has an average location error of 0.26 mm for the sphere centers. Experimental studies show that the average geometrical alignment error between the Lu image and the CT image is 0.07 mm. The transformation matrix was employed in a 9-point 18F-FDG source phantom experiment which shows good fusion results. The proposed method is absense of positron sources in the calibration procedure and reduces the impact of positron range and non-colinearity effects. Thus, it is expected to get higher spatial resolution than the positron source measurement. More quantitative comparison between the proposed method and the traditional methods is being studied. The phantom is easily to be fabricated, low-cost and reusable, which makes the PET/CT geometrical alignment more convenient. Moreover, without dealing with the position sources, it reduces the radiation dose to the operators. We conclude that the proposed method is gainful to replace the positron source based methods.

The simultaneous acquisition of measured µ-maps and MR images on hybrid MR-PET systems may significantly improve the validation and development of: (1) new MR-based µ-map algorithms and, (2) template approaches for coil and implant attenuation correction. For several MR-PET systems, however, use of conventional transmission schemes is not possible as the patient table and relatively small diameter scanner bore significantly restrict radioactive source motion and limit tomographic sampling. As an alternative we propose using a fixed source geometry in combination with patient table motion to acquire full tomographic sampling and reconstruct continuously valued µ-maps. We present an implementation of a fixed torus coincidence transmission strategy using ¹⁸F for the Siemens Biograph MR-PET scanner and assess quantification of this approach in phantom and nonhuman primate scans. We utilized a Bayesian reconstruction algorithm which directly generates µ-map estimates from multiple bed positions. For scatter estimation we developed a maximum likelihood expectation maximization scheme based on the comparison of counts in the tails of blank and transmission sinograms; explicitly accounting for residual errors in the time related corrections of these projection sets. Acquisitions with five table positions (65 mm steps) of a uniformly filled water cylinder produced µ-maps with a deviation of <3% in axial uniformity over the entire field of view, with absolute quantification calculated at 0.098 cm^-1. A simultaneous MR-PET transmission study of a rhesus macaque showed a clearly delineated mandible in the reconstructed µ-map, overlaying exactly the expected signal void in the registered T1 MR volume. Preliminary experimental results suggest that an ¹⁸F fixed torus transmission scheme specific for a hybrid MR-PET system can produce highly quantitative µ-maps in a range of imaging scenarios.

A rigid motion correction technique is applied to both PET and CT data from the Siemens mCT PET/CT scanner. Motion was applied to a Hoffman brain phantom during both the CT and PET scans, as well as between them. The motion data, which was tracked by an external motion tracker, was used to correct the motion corrupted tomographic data during reconstruction. The motion corrected CT reconstruction was used for attenuation correction in the PET reconstruction. The subsequent reconstructions are compared to those without motion to show the effectiveness of the motion correction.

In positron emission tomography (PET), patient respiratory motion can cause misalignment between the estimated attenuation map and the PET data affecting the quantitative accuracy of the attenuated corrected PET images. In PET/MR attenuation correction is a challenging task due the poor MR signal from important tissue types in common MR sequences leading to misclassification of tissues during attenuation correction. This thesis evaluates the effect of motion induced uncertainties caused during attenuation correction and compares MR attenuation map (MRAC) and CT attenuation map (CTAC). The whole investigation was based on an anthropomorphic numerical phantom. Different approaches for attenuation correction have been investigated; using either breath-hold attenuation images or respiratory-gated attenuation images that are phase-matched with respiratory-gated PET images. Lung and liver lesions displacement, SUVmean and SUVmax were evaluated. The results of this study show that respiratory motion induced artifacts can cause underestimation of lesion uptake and overestimation of lesion volume and displacement. Reconstruction of PET images with gated attenuation map has better recovery on uptake value, lesion volume and lesion position to the actual value, comparing to using static or averaged attenuation map. Reconstructed images with CTAC have a better recovery on uptake value comparing to using MRAC. This study indicates that the accuracy of attenuation map has a similar effect as motion in image quality. Both attenuation and motion have to be considered for accurate PET quantification.

We investigated the performance of a post reconstruction dual respiratory and cardiac (R&C) motion compensation method for improved image quality of 4D cardiac gated small animal myocardial perfusion (MP) SPECT images. A normal mouse was injected with ~8 mCi of Tc-99m sestamibi, anesthetized, fitted with ECG leads for cardiac gating signal acquisition, and placed on top of a pressure gauge bellow for respiratory motion measurements. A 2-hour list-mode dataset was acquired using a MILab small animal SPECT system fitted with a multi-pinhole collimator with 0.4 mm resolution in 5-minute sections. They were subsequently sorted for different acquisition times and reconstructed using a vendor provided OS-EM algorithm with simultaneous 6 respiratory and 8 cardiac equal-time gates over each motion cycle. Using a group-wise B-spline non-rigid image-based registration method, the deformation fields of the respiratory motion (RM) were estimated and applied to each cardiac phase for RM correction. Then, the RM compensated cardiac gated SPECT images were similarly used to estimate cardiac motion (CM) and later transformed to a reference frame and summed. Finally, the reference frame was inverse-transformed using the estimated CM to each of the 8 cardiac frames. The cardiac gated images with dual R&C motion compensation were compared to those without correction but with post-smoothing filter. The results showed the dual R&C motion compensation significantly reduced image noise level. At the same time, the image resolution was improved by 10% to 40% depending on the different acquisition times when compared with that obtained without motion compensation at the same image noise level. We conclude that dual R&C motion compensation provides significant reduction of noise level in 4D cardiac gated small animal MP SPECT images with minimum degradation of resolution. The improved image quality can be traded for reduction of acquisition time or radiation dose to the animal.

In hybrid brain imaging, the integration of CT for the purpose of providing attenuation correction has proven to provide substantial benefit to emission image quality. Unfortunately due to the sequential nature of the acquisitions this benefit may be reduced or even nullified if the patient fails to remain motion-less after the transmission image is acquired. In particular, the misalignment of the CT from the PET can have a negative effect on the perceived activity or quantification by erroneously altering count levels in the emission image post-attenuation correction. Current methods to mitigate the misalignment include head constraints, co-registration of the transmission and emission data retrospectively, and re-acquiring the transmission scans. Most reports show the negative effects a misaligned attenuation map on perceived activity through simulation with simple motions, but little has been done to explore the more realistic complex motions that often occur in clinical settings. In this abstract we show that misalignment of the CT from complex transformation for CT-based attenuation correction can negatively impact distributions and count levels more than simple motions with greater magnitude of displacement. Through motion simulations, we investigate the extent that complex motion can have on attenuation correction.

A new reconstruction method is explored using Bayesian inference for Poisson Statistics for emission tomography. The Gamma density function is chosen as the natural choice for the activity distribution at each voxel, being the conjugate-prior of Poisson distribution. The update equations of the shape and rate parameters for Gamma distribution are derived and tested on a simple 2D example using Metropolis-Hastings algorithm. The results show promise with quick convergence within ~20 iterations and stable noise properties with iteration. A 3D algorithm and comparison with OSEM is underway.

For a quantitative analysis in positron emission tomography (PET) or single-photon emission computed tomography (SPECT), attenuation correction is mandatory. A CT-scan or transmission scan are very common to determine the attenuation µ-map, but in the case of a PET/MR hybrid system it is difficult to associate one of these scans. Many techniques have been developed in order to get the attenuation µ-map. Some methods are based on a template- or atlas techniques, other methods are based on a segmentation technique such as Dixon or UTE (Ultrashort Echo Time). Another method using the emission sinogram data have been developed to compute the attenuation µ-map. An iterative MLAA (Maximum-Likelihood reconstruction of Attenuation and Activity) algorithm has already been proposed to find the estimated emission image for the PET/MR imaging. In this context, we proposed to drive the MLAA algorithm thanks to a scale factor aj and by using the UTE µ-map and the T1-weighted (T1w) MR image in order to converge to a solution. We proposed a method to determine these aj factors in the attenuation update. Results show that the optimized MLAA algorithm improved the estimated emission image compared to OSEM/UTE, OSEM/DIXON and a standard MLAA (aj=1). By using this optimized algorithm we improved the uniformity of the reconstructed image. Compared to the PET image reconstructed with an aligned CT image, the error in different circular regions of interest (ROI) shows a relative error inferior to 2% in all the region of the brain using the optimized values of aj in the MLAA algorithm. We proved that an optimization of the aj is a promising technique to improve the PET image, but further validation is required on a larger number of patients.

We propose two regularizations techniques for a bent-ray (BR) tracing algorithm to reconstruct the speed of sound maps of breast tissues in an Ultrasound Computed Tomography (USCT) system. When high frequencies are employed, the use of BR is a good approximation to describe the propagation of the front of the pressure wave. The quantitative accuracy of the images reconstructed with the BR algorithm was evaluated without any kind of regularization, and with two regularization methods. The regularizations were based on some available a priori information, namely the known higher and lower values of the speed of sound expected in the breast tissues, and the maps of the internal structures obtained from the standard reflection ultrasound (US) imaging. The use of the proposed regularizations in the implemented algorithm improves the convergence and quality of the resulting images, although further improvements are still possible. These methods will help obtaining quantitative US images in a reasonable amount of time, expanding the possibilities and applications of this technique.

When multiple pinholes are used to acquire SPECT projection data, there may be regions on the detectors where data from two or more pinholes overlap. Such "multiplexing" of projection data can cause artifacts and/or increased noise in reconstructed SPECT images, depending on the exact acquisition geometry. We previously described a modified MLEM algorithm that alternates between (i) estimating the separated projection data that would be obtained through each individual pinhole (i.e., demultiplexed data), based on the current MLEM iteration's image, and (ii) reconstructing the SPECT image by MLEM, but assuming that the demultiplexed data estimates are the actual measured data. The modified MLEM algorithm was first tested with data simulated for a stationary 39-pinhole collimator tube, and more recently with Tc-99m phantom data acquired from three vials of different sizes on a 6-pinhole, triple-detector SPECT camera. The iterative algorithm asymptotically eliminated artifacts obtained when reconstructing larger objects that yielded multiplexed projection data. Reconstructions of non-multiplexed 6-pinhole data from the small vial (2.3-cm diam. × 4.5-cm long) were free of artifacts, whether using standard MLEM or the new demultiplexing algorithm. Images of a medium-size vial (3-cm diam. × 6-cm long) and a large vial (4-cm diam. × 7 cm long) -- which showed, respectively, moderate and severe multiplexing artifacts with standard MLEM -- were greatly improved by using the iterative demultiplexing approach. When simulating SPECT data with multiplexing corresponding to ~42% of the non-zero detector pixels, the new method reduced the structural noise (artifacts) by ~360%, with only a modest increase in stochastic noise (8.3% with respect to images reconstructed from ideal, but unrealistic non-multiplexed projection data of the same number of counts). The new algorithm provides a practical and useful approach to correct for data multiplexing in multiple-pinhole SPECT.

Proton computed tomography (pCT) represents a promising tool for proton and ion therapy treatment planning as it provides the relative stopping power (RSP) values required to accurately place the distal edge of the beam at the desired position within the patient. These RSP values have previously been reconstructed using iterative projection methods with filtered back projection (FBP) used to define the initial iterate x₀ and determine the object boundary used in the most likely path (MLP) calculations that generate the system matrix A. The purpose of this work was to develop methods for improving the accuracy of A and x₀ by incorporating hull detection into the calculations that generate them. Using the convex hull to improve the calculation of entry and exit points of each proton's MLP results in a more accurate system matrix A. By intersecting a morphology dilation of the convex hull with a median filtered, filtered back projection (FBP) reconstruction, an accurate initial iterate x₀ was generated. Starting from an initial iterate x₀, reconstructed images were generated with the diagonally-relaxed orthogonal projections (DROP) method using a Geant4 simulated proton CT scan of the CTP404 phantom. An analysis of the reconstructed images showed less than a 1\% discrepancy between reconstructed and reference RSP values in each of the 8 material inserts of the phantom. Using hull detection to improve path approximations and refine the initial iterate thus resulted in highly accurate reconstructed RSP values. This approach represents a promising development in pCT as this level of accuracy is an improvement compared to x-ray-based treatment planning for proton therapy.

Gated computed tomography (CT), or four dimensional CT (4DCT), is important not only for the estimation of respiratory motion to reduce motion blurring artifacts in applications such as radiotheraphy planning, but also for the estimation of motion pattern under certain diseases. Because of the additional CT scans acquired in the gated CT, the radiation dose is always a concern. In this project, we develop an iterative reconstruction algorithm to jointly estimate all the gated CT images together to reduce the dose while keeping the image quality. Specifically, we combine 4D non-local and motion-based low-rank regularizations to take the advantage of the redundant motion information in the gated CT images. Non-local regularization is a robust and edge preserving method that computes the patch-based weighted summation.While calculating non-local regularization, we can make a pseudo motion vector that includes spatio-temporal positions with minimum distance. Even though the motion vector is not perfect, we can generate a pseudo registration matrix that is low-rank. We demonstrate that the proposed method improves the image quality using computer simulations.

Obtaining accurate left ventricular (LV) myocardial activity estimates and segmentation is important for quantitative and functional analysis of cardiac PET and SPECT images. However, factors such as finite spatial resolution and noise can significantly degrade the accuracy of these estimates. We propose a model-based reconstruction method that can accurately estimate LV activity and segmentation based on models of the imaging system and myocardial geometry. The LV activity was modeled by a set of segments with homogeneous uptake. The endocardial and epicardial surfaces were modeled using a set of rays emanating from the long axis in fixed directions and the endocardial distance and wall thickness for reach ray. Surfaces were defined by cubic spline interpolation between these points and the resulting voxelized solid, and combined with the activity model to produce an activity image. The image was then fitted to the measured image or data based on models of the imaging system to estimate the segmental activities and geometric parameters. As a preliminary demonstration, we used a simplified 2D short-axis slice generated from 8 rays, assuming homogeneous activity concentrations in the background (5), blood pool (15) and myocardium (95). We investigated two estimation methods: one that maximized the projection-domain likelihood using a realistic model of the SPECT image formation process, and another that modeled the imaging process as the convolution with a Gaussian kernel plus Gaussian noise and used a least squares fitting method. A total of 50 noise realizations were used. Results demonstrated the feasibility of the method with dice coefficients > 0.95 for segmentation and myocardial activity estimation error < 1%. The projection-domain estimation method generally performed better due to more accurate imaging model.

We consider X-ray coherent scatter ima¬ging, where the goal is to reconstruct momentum transfer profiles (spectral distributions) at each spatial location from multiplexed measurements of scatter. Each material is characterized by a unique momentum transfer profile (MTP) which can be used to discriminate between different materials. We propose an iterative image reconstruction algorithm based on a Poisson noise model that can account for photon-limited measurements. To improve image quality, previous approaches use edge-preserving regularizers to promote piecewise constancy of the image in the spatial domain while treating each spectral bin separately. Instead, we propose spectrally grouped regularization that promotes piecewise constant images along the spatial directions but also promotes smoothness of MTPs across neighboring spatial bins. We demonstrate that this group regularization results in a considerable improvement of both spectral and spatial image quality. We pursue an optimization transfer approach where convex decompositions are used to lift the problem such that all hyper-voxels can be updated in parallel and in closed-form. The group penalty introduces a challenge since it is not directly amenable to these decompositions. We use the alternating directions method of multipliers (ADMM) to replace the original problem with an equivalent sequence of sub-problems that are amenable to convex decompositions, leading to a highly parallel algorithm. We demonstrate the performance on real data.

Data acquired through the PET system tend to be very noisy, partly due to low radiation doses. In this paper, a new reconstruction strategy based on a combination of the MLEM and total variation (TV) is presented. The method uses a regularizing term, providing noise reduction and data reconstruction from low number of counts. Additional an adaptive regularization parameter is embedded within the method. Experimental results, for the performance evaluation using PET simulated data from a small mouse phantom, shows the efficiency of the MLEM-TV reconstruction method proposed in comparison with those obtained by MLEM.

PET image reconstruction is an ill-posed problem. Maximum /a posteriori/ (MAP) estimation with a local smoothness prior is frequently used to reduce noise. However, smoothing across different organs can introduce bias and reduce the contrast of objects. To address this problem, we propose to model organ boundaries explicitly using level set functions and to avoid smoothing across these edges. We incorporate a statistical shape prior into the MAP reconstruction to assist automatic segmentation of organs with low contrast. The overall image prior models both the smoothness of PET tracer uptake within each organ and the similarity between the reconstructed organ boundaries and the prior shape. One level set function is used to represent the boundary of each region in reconstructed images. When multiple organ boundaries are close to each other, the level set functions evolve to reduce the feature difference within each region while considering the competing forces from neighboring regions. We conducted computer simulations to evaluate the performance of the proposed method. A whole-body mouse phantom with multiple organs, including lung, myocardium, liver, kidneys, spleen, and tumors, was constructed. Statistical shape priors of these organs were derived from a set of real mouse FDG PET images. PET data were generated and reconstructed using the proposed method and the conventional MAP reconstruction followed by multi-phase level set segmentation. Compared with the conventional method, the proposed method was able to simultaneously segment all organs, and obtained accurate region definition even for neighboring organs with similar activities.

Precise geometric calibration of SPECT cameras is essential for accurate image reconstruction. Directly generating parameters from SPECT projections is challenging due to the need to accurately resolve small point sources. These uncertainties can propagate into the geometric characterization but still yield a self-consistent set of parameters that permit image reconstruction. We show the feasibility of this method for pinhole SPECT. We present results for a simulation study and for characterization of single-pinhole system using projection acquired from a triple-isotope calibration phantom. We simulate a single-pinhole system with 36 projections at 10-degree intervals along a circular trajectory of 45mm radius. The calibration phantom was made of 6 sources (each 1.5mm diameter, 3mm long). The phantom was positioned obliquely with respect to the system axis. For the physical experiment, a nanoSPECT (Mediso) small-animal scanner was used to create a camera configuration that was similar to the simulation study. The 6-source calibration phantom was created using 3 isotopes: one pair each of Tc99m, In111, and Tl-201. Projections of the calibration phantom were created using a true system. From these projections, we obtained the geometric parameters of the system and thereby created a measured system matrix. Projection data for a digital cardiac phantom were generated with the true system matrix and then reconstructed with both the true and measured system matrix. Separate projections were acquired with the small-animal scanner using energy windows corresponding to the three isotopes. Geometric parameters were found and used to create a measured system matrix. The point source calibration phantom was then reconstructed. These studies show that geometric characterization of pinhole SPECT systems is possible using a 6-point calibration phantom. Use of multi-isotopes for point sources facilitates identification and localization of point sources

The importance of performing exact image reconstruction from the minimum amount of data has been recognized for a longtime.Yangbo Ye et al. has proved that the interior problem can be exactly and stably solved if a subregion in an ROI/VOI in the FOV is known from fan-beam/cone-beam projection datasets using the analytic continuation technique . But the analytic continuation is very ill-posed question. The conventional methods are iterative or not suited to multi-region-known situation In this work we proposed a novel method based on fundamental solution method (FSM) analytic continuation to reconstruct whole object from truncated projection with a known subregion.First we translate the analytic continuation problem to an equivalent Cauchy problems of Laplace equation. And then we use the FSM method to solve the problem numrically .Because the problem is very ill-posed ,we adopt Tikhonov regularization in chord reconstruction.And We can use Singular value decomposition (SVD) to accelerate the algorithm The method is fast analytical algorithm and can be applied to multi-region-known situation.

Iterative algorithms such as ART, DROP, and CARP are commonly used in reconstructing computed tomography images, but only account for errors in the measurements. Errors in the predicted path and intersection lengths, or even blocks of missing measurements can result in degraded image quality. Robust techniques allow for errors in other areas of the model and produce good images that show less sensitivity. In this paper we introduce a robust version of DROP and compare its performance advantages to the standard DROP algorithm on real data.

Most of the maximum a posteriori (MAP) algorithms for transmission reconstruction use edge-preserving convex non-quadratic (CNQ) priors which have proven useful for reconstructing emission images. However, since the measured data from transmission scans are fundamentally different from emission data, questions regarding the appropriateness of the use of CNQ priors for transmission reconstruction remain mostly moot. In this work, we apply a weighted median prior (WMP) to transmission tomography. Our motivation lies in the fact that, unlike the CNQ priors which smooth noise in monotonic regions but allow limited smoothing at edges, the median prior simply preserves edges as well as locally monotonic regions by imposing the behavior that each pixel is attracted to its local median. The median prior (MP) used in this work is different from the median root prior (MRP) which is based on a heuristic empirical method. We basically follow the derivation of the MP proposed by Hsiao et al. but apply it to transmission tomography with modifications. To solve our MAP problem, we use an alternating joint minimization algorithm based on a separable paraboloidal surrogates algorithm. Our experimental results indicate that the edges restored by properly adjusting the center weight in the WMP are more accurate than those restored by adjusting the smoothing parameter in the standard MP.

A common strategy to speed-up image reconstruction in tomography is to use subsets, i.e. only part of the data is used to compute the update, as for instance in the OSEM algorithm. However, most subset algorithms do not convergence or have a limit cycle. Different strategies to solve this problem exist, for instance using relaxation. The conceptually easiest mechanism is to reduce the number of subsets gradually during iterations. However, the optimal point to reduce the number of subsets is usually depends on many factors such as initialisation, the object itself, amount of noise etc. In this paper, we propose a simple scheme to automatically compute if the number of subsets is too large (or too small) and adjust the size of the data to consider in the next update automatically. The scheme is based on idea of computing two image updates corresponding to different parts of the data. A comparison of these updates then allows to see if the updates were sufficiently consistent or not. We illustrate this idea using 2 different subset algorithms: OSEM and OSSPS.

This paper analyzes the performance of ensemble tree learning techniques for predicting the progression of Alzheimer's disease (AD) in a longitudinal dataset of 128 mild-cognitive impairment (MCI) subjects from the Alzheimer's disease Neuroimage Initiative (ADNI). The system is evaluated by means of tissue-segmented T1 magnetic resonance images (MRI) as well as Mini Mental State Examination (MMSE) and cognitive assessment subscale (ADAS-Cog) scores collected every 6 months during a three-year follow-up. The ensembles are trained by means of boosting and bagging while their performance is compared in terms of the classification error and the received operating characteristics curve (ROC) using cross validation.

DaTSCAN is a widely used radiotracer that binds to the dopamine transporters in the striatum and allows us visualizing a possible dopamine deficiency state. Thus, it has been commonly used to assist the diagnosis of Parkinsonism. However, when distinguishing Parkinson's disease (PD) from non-idiopathic parkinsonian variants the usefulness of DaTSCAN is reduced. In these cases, the 18F-Desmethoxyfallypride (DMFP) drug has recently emerged as an effective radiotracer. This work demonstrates a full automatic computer system based on DMFP-PET data to separate idiopathic and non-idiopathic PD patients. In order to deal with the information contained in the neuroimages, we used a multiple kernel learning approach. Two kernels were implemented: one for voxels in the striatum and other one for voxels in the remaining regions. Accurate rates up to 78% were obtained.

Spatial normalization of brain images to a common template is a usual preprocessing step in several comparison studies, brain tissue segmentation methods and computer aided diagnosis systems. For [123I]FP-CIT SPECT and [18F]DMFP PET brain image modalities, the template building process is handicapped by their low intensity values outside the striatum. In this paper, four MNI space templates are built for both modalities using a functional database of these images without the co-registered MRI, a typical scenario in medical practice. Firstly, two standard methods are proposed based on the average of a set of [123I]FP-CIT SPECT and [18F]DMFP PET images. These images are previously co-registered to the MNI T1- weighted MRI template and to a single image registered to the MNI template, respectively. The third method iterates the registration and the average over the first template obtained. Finally, in the fourth method, a posterization of the source images into three levels (striatum, rest of the head and background), is performed. Analogously, a posterized version of the MNI template is designed using a labeled brain image (MNI atlas). The posterized images are then used to differently weight the intensity values in the striatum, head and background and to obtain a posterized transform between anatomical/functional modalities. This transform allows us to register the source images to the MNI template with subsequent averaging.

Breast-conserving surgery is used frequently to treat patients with non-palpable breast cancer. Techniques like radioguided occult lesion localization and radioactive seed localization are used to reduce the number of incomplete excisions. We hypothesize that the number of incomplete excisions can be reduced further if intraoperative information about the location of the radioactive seed is combined with preoperative information about the extent of the tumor. A system for computer-assisted resection of non-palpable breast lesions was proposed in this study to visualize and track a sphere-shaped target volume which is centered at the location of the radioactive seed and which circumscribes the tumor. Physical experiments were conducted to assess the accuracy (defined as trueness and precision) of the proposed system. A gamma camera system (this system was used to establish the location of the radioactive seed) was composed of two Phillips Forte gamma camera heads and two low-energy high-resolution parallel-hole collimators. A range camera system (this system was used to visualize the sphere-shaped target volume) was composed of one MESA SR4500 time-of-flight camera. The trueness along the vertical axis was less than 0.5 cm when the phantom thickness was less than 5.5 cm. The trueness along the horizontal plane was less than 0.5 cm when the phantom thickness was less or equal than 5.5 cm. This is important since an accuracy of 0.5 cm is required to ensure that negative surgical margins can be achieved. We conclude that the proposed system is feasible for the application in breast-conserving surgery.

4D MRI is currently gaining attention as an imaging modality which is able to capture inter-cycle variability of respiratory motion. Such information is beneficial for example in radiotherapy planning and delivery. However the motion extracted from 4D MRI is dependant on the method of processing the images. Typically a deformable registration algorithm is used so that the 3D motion of every point in the image can be found. We have used several popular registration algorithms which have been recently published with publicly available data. Our initial findings suggest that due to the low contrast of 4D MRI, the registration is only accurate near clear boundaries, and less accurate elsewhere, to different degrees depending on the algorithm used. We thus suggest modelling organ deformation as a piecewise-affine transformation, which is then embedded into the deformation field for more realistic modelling of respiratory motion.

Cadmium zinc telluride (CZT) detectors are highly promising for advanced PET imaging due to their high energy and spatial resolution. In this contribution, an improved data analysis is suggested for the small animal PET system currently under construction at Stanford. Good energy resolution for this system is of certain interest, as it leads to enhanced signal to noise ratios and is necessary for advanced strategies to increase coincidence sensitivity in PET. For the proposed analysis strategy, two different approaches are studied where the first approach is an automated regulation process to determine appropriate fit parameters for each channel. The second approach identifies channels with over-corrections and changes the data analysis chain to avoid over-corrections. Following the measured data set in this analysis, the two approaches are make a relative improve- ment of the energy resolution by 15% and 0.1%, respectively, when reasonable margins are set. This equates to an absolute improvement from 8.54% to 7.25% FWHM at 511 keV for a combined processing.

Data-driven estimation methods have several advantages over the use of external device to determine respiratory motion (RM). Our goal is to evaluate the effect of myocardial uptake, image noise level, and data analysis method on the extraction and estimation accuracy of RM from 4D cardiac gated list-mode PET data. Cardiac PET projection data were generated from the 4D XCAT phantom with realistic anatomy and RM and cardiac motion (CM) using the GATE Monte Carlo simulation code. The RM of the phantom was replaced by an extracted RM from a normal patient data. Four myocardium-to-background (M/B) uptake ratios and 4 count levels were used in the simulation. Simulated list-mode data were derived from the finely sampled simulated projection data. The RM signal was extracted as the center-of-mass (COM) of a region-of-interest that contains the heart in the projections in 200 msec frames calculated from the list-mode data. Background (BKG) counts within the ROI were estimated and subtracted with and without smoothing. The frequency spectrum of the RM motion was obtained from the Fourier transform (FT) of the time sequence of COM and the average RM frequency was identified from the highest amplitude peak. The full RM signal used for respiratory gating was obtained from the inverse FT of the smoothed frequency spectrum. Our results show the RM signal magnitude is increased and closer to the ‘true’ value with than without BKG subtraction, but with increased signal noise. Smoothed BKG subtraction improves the estimate with smaller signal noise increase than direct subtraction. Lower count level reduces the SNR of the RM signal for all 3 methods. Lower M/B reduces the SNR of RM signal and its extraction fails when the M/B and count level fall below 0.3 and ½, respectively, of the average values found in clinical studies. In conclusion, RM extraction and estimation accuracy improve with BKG subtraction and are negatively affected by low M/B and high image noise.

Modern day, minimally invasive x-ray image guided endovascular interventional treament of neurovascular diseases such as aneurysms involve the use of devices such as stents and coils. These devices have very small features. Current state of the art imaging systems using flat panel detectors have inherent pixel sizes of 160 – 200 µm and are insufficient to faithfully image the treatment devices. As a solution to this a Micro Angiographic Fluoroscope (MAF-CCD) with an effective pixel size of 35µm was built and successfully used during interventional treatments. While this improves the spatial resolution, applying image processing techniques such as noise reduction, dynamic range management, edge enhancement and other methodologies implemented on current clinical platforms, further improvements to device visualization can be achieved. An image processing algorithm, similar to that utilized on a clinical system, was implemented on clinical MAF-CCD images of stent and coil deployments during aneurysm treatments. The processed and unprocessed images were compared and quantitative metrics such as relative noise and contrast to noise ratios (CNR) were measured. The relative noise over anatomically uniform regions was reduced by 72% with the application of a Super Noise Reduction Filter (SNRF).The CNR of coils before and after image processing was measured to be 9.22 and 19.208 respectively, with an improvement factor of 2.08. Using advanced image processing techniques on images obtained during interventional treatments using high resolution detectors can provide critical information such as details of stent structures, coil structure and voids within the coil masses. Such improved visualization can act as a beneficial aid to the interventionist and improve the quality of the intervention.

Image denoising is an important task in many medical imaging applications. There have been many developed algorithms such as gaussian kernel based filtering, anisotropic diffusion, bilaterial filtering, non local means filtering, and total variation based denoising. In the computer vision community, the guided image filtering (GF) has been introduced, which demonstrated powerful denoising performance and fast computation time. GF was also extended to Iterative GF to improve the filter output quality (e.g. rolling guidance filtering or RGF). Recently, GF was applied to PET denoising problem in simultaneous PET-MR with matched spatial resolution of MR. In this work, we propose a new iterative GF method called iterative guided image filtering (IGF). Unlike RGF, our new method uses an intermediate guide image. The proposed method was evaluated using a toy example as well as SPECT-CT Monte Carlo (MC) simulation with XCAT phantom. When there were matched edges between noisy image and guide image, then both GF and IGF showed significant improvement in image quality due to edge-preserving property of GF for toy example images. However, when there are mismatched structures between noisy and guide images, severe blurring was observed in GF result image, but relatively good edge information recovery was observed and IGF using side information image yielded the best the root mean squared error (RMSE). Similar tendency was observed in the SPECT-CT MC simulation. Both GF with CT side information and IGF without CT information yielded about 40% reduction in RMSE. When both IGF and CT information were used, 55% RMSE reduction was achieved. Our proposed method can potentially be applied to many multimodal medical imaging applications including PET-CT, SPECT-CT, and simultaneous PET-MR.

Portable functional imaging devices can be used in oncological surgeries to locate residual tumors for better patient recovery and survival. Taking the patient dose and the limited time of surgery into account, the count in such images could be very low. In this study, we investigate effectiveness of different spatial denoising methods, such as Gaussian filtering, bilateral filtering, Rudin-Osher and Fatemin (ROF) denoising, and non-local means filtering, on low count functional images. We also propose a new denoising method based on maximum a posteriori (MAP) criterion. The simulation study shows that the simple methods, such as Gaussian and bilateral filtering, may be as effective as the advanced searching or iterative methods as measured by the relative root mean square error when the count is low. Further investigations using more realistic simulations or real functional images and tumor detection performance are needed to evaluate these methods at high noise levels.

Proton computerized tomography (pCT) allows the reconstruction of a 3D map of relative stopping power (RSP), which can be directly used for a proton treatment planning, avoiding the errors due to the conversion from x-ray CT numbers to RSP. The complete pCT image reconstruction requires pre-processing of a large amount of raw data, including tracked coordinates and water equivalent path length (WEPL) values of >100M proton histories per scan. Before an iterative reconstruction algorithm is applied to the coordinate and WEPL data, an initial image and image boundary is reconstructed using filtered back projection (FBP). In this work we present an updated C++ based data processing platform, capable of processing ~200 Mbytes of data from ~2.5×106 proton histories in about 20 seconds. In addition, we investigate different image processing techniques, applied to the FBP image to improve the quality of the initial image using the Catphan CTP404 phantom module. When comparing thresholding methods to a Gaussian blur filter applied two or three times, we found that both techinques avoided artifacts inherent in the FBP image. The Gaussian blur filter achieved better RSP accuracy than the thresholding methods with fewer numbers of iterations.

In this project, we assessed the clinical value of tumor heterogeneity measured with 18F-FLT as a biomarker for lung cancer diagnosis and staging, and compared the performance to traditional image features using final pathology as gold standard. We also proposed to apply support vector machine (SVM) to train a vector of image features including heterogeneity extracted from PET/CT image to improve the diagnosis and staging for lung cancer. Thirty-two subjects with lung nodules (19 M, 13 F, age 70 ± 9 y) who underwent 18F-FLT PET/CT scans were included in our study. The reference diagnosis and staging, including benign, early and advanced malignance, were determined by final pathology from biopsy or surgery. We applied the global Moran I(d) analysis to characterize the intra-tumor heterogeneity on PET images 1h post-injection. Other than texture analysis that widely used in heterogeneity prediction, I(d) statistic is a measure of spatial autocorrelation characterized by the correlation among 3D neighboring voxels. Other image features including SUV and CT texture were extracted from PET/CT images. Then we trained and applied a support vector machine (SVM) based statistical machine learning tool to fuse the features and test the SVM performance in classifying patient groups: benign/early malignant and early/advanced malignant. In conclusion, we obtained a novel measurement of intra-tumor heterogeneity which has promising performance for diagnosis and staging of lung cancer. We demonstrated the feasibility of performing SVM based cancer staging using multiple image features in PET/CT, the ROC of SVM with multiple image features outperforms other methods. SVM analysis and classification with combination of effective features has great potential to augment diagnostic accuracy and improve tumor staging in oncological practice.

This work investigates the attenuation of a proton beam to reconstruct the map of the inelastic macroscopic cross section of an object. Tomographic reconstructions using most likely path (MLP) estimates were performed using a filtered back-projection (FBP) algorithm with a distance driven binning. Comparison of the spatial resolution of attenuation proton computed tomography (pCT) images with a conventional binning shows that incorporating the MLP estimate during reconstruction improves the spatial resolution of attenuation pCT. The statistical limitation of this method has also been investigated and it was found that the noise in attenuation pCT images is about two orders of magnitures higher than the noise in energy loss pCT images for the same imaging dose. Finally, the impact of the energy dependence of the inelastic cross section below 100 MeV has been evaluated and it was found that an inhomogeneity artefact is visible in a 20 cm water cylinder with a 175 MeV proton beam but it is corrected using a 200 MeV proton beam.

Recently we developed a Bayesian-FBP (Filtered Backprojection) algorithm for CT image reconstruction. This algorithm is also referred to as the FBP-MAP (FBP Maximum a Posteriori) algorithm. This non-iterative Bayesian algorithm has been applied to real-time MRI, in which the k-space is under-sampled. This current paper investigates the possibility to extend this FBP-MAP algorithm by introducing more controlling parameters. Thus, our original FBP-MAP algorithm became a special case of the extended FBP-MAP algorithm. A cardiac patient data set is used in this paper to evaluate the extended FBP-MAP algorithm, and the result from a well-establish iterative algorithm with L1-norms is used as the gold standard. If the parameters are selected properly, the extended FBP-MAP algorithm can outperform the original FBP-MAP algorithm.

The aim of this work consists in the full simulation and development of a GEM (Gas Electron Multiplier) detector for a possible application as dose monitor for treatment plans verification in hadrontherapy at CNAO Center (National Center for Oncological Hadrontherapy). A triple GEM detector (15 x 15 cm2 area) could provide a monitoring for the treatment plan dose verification with good resolution and optimal radiation background control with respect to the existing devices. The Monte Carlo Geant4 10.01 toolkit was used to simulate the complete CNAO extraction beamline (beam delivered with active scanning modality) , with the active and passive components along it. The simulation allowed the characterization of the GEM response to protons and carbon ions with respect to reference detectors (EBT3 radiochromic films) and the optimization of the geometry of the readout electrodes. To achieve the required spatial resolution (consistent with the FWHM of the beam), different readout electrode configurations with pad segmentation was exploited. In the simulation a new particle generation method was developed in order to reproduce the experimental treatment plan by importing the DICOM RT-PLAN file, which contains all the information on the irradiation geometries and sequences (treatment plan parameters). The GEM was previously characterized by simulated irradiation, giving good results, both for single beam spot, scanned radiation fields and simple treatment plans. A real GEM is going to be developed and tested: important beam parameters such as the transverse FWHM and scanned radiation field's uniformity is compared with experimental measurements at CNAO Centre.

Among the successful gamma camera designs with angular sensitivity, most rely either on Compton imaging or a coded aperture – in any case requiring a position-sensitive sensor and involved real-time data analysis. In some use scenarios, such as tracking of a point source of radiation, design constraints can be relaxed from full gamma imaging to a mere determination of the gamma radiation direction. We describe an extremely cost-effective camera design built around a single stationary scintillation counter and a single specifically shaped rotary obstruction that nevertheless measures both the azimuthal and the elevation coordinates of the impinging radiation, and report on quantitative performance of the prototype with its design parameters tuned to real-time monitoring of a typical pulsed HDR brachytherapy session. We could also foresee its use in homeland security applications.

We have previously been successful obtaining proton radiographs for thin objects (< 4 cm) using a binning technique, however for thicker objects, multiple Coulomb scattering (MCS) of protons in the materials has a larger effect on proton path. Thus, proton paths in the object can no longer be approximated as straight lines. The consequence is that the angular and spatial bins, to which protons are initially assigned in preprocessing, are less representative of the true average position of the proton inside the object. We are developing an algorithm that utilizes the most likely path (MLP) and the water equivalent path length (WEPL) of each proton, in order to estimate the water equivalent thickness (WET) of each voxel in the image space. To test this algorithm, a step phantom composed of water, cortical bone and lung was simulated in Geant4 and a single projection was taken. We compared the known WETs of the steps with the reconstructed WETs for both the binning technique and the new algorithm. The MLP algorithm for proton radiography produces high image-quality radiographs that represent the WET of the object being imaged within approximately 3% for objects of 2-20 cm in thickness. This technique may be useful for alignment verification during adaptive proton radiation therapy.

Treatment monitoring in hadron-therapy is one of the key issues for treatment quality assessment. The INSIDE project is building an in-beam treatment monitoring system based on a dual-head PET scanner and a fiber-based charged particles tracker, to be used at CNAO synchrotron hadron-therapy facility in Pavia, Italy. In this work the figures of merit and the results of a test of PET modules with proton beams are presented. Materials and methods: The first two modules of the INSIDE PET scanner have been tested with proton beams. Each module is made of Multi-Pixel Photon Counters matrix of 16x16 channels with 3.2 mm pitch, pixels are coupled 1:1 to LFS scintillator crystals 20 mm thick. Measurements have been performed in the 68-100 MeV energy range, with 50 cm face-to-face distance between modules. A Maximum-Likelihood Expectation Maximization algorithm is used for image reconstruction. Results and Conclusions: The first two modules of the INSIDE PET scanner have been successfully tested in the expected working conditions. Data were acquired seamlessly and image reconstruction is feasible also with beam on. Activity profiles measured in the PET images are in good agreement with the expected range.

Proton computed tomography (pCT) can produce accurate images based on relative stopping power (RSP) calculation for proton treatment planning. The Proton CT system is composed of 4 silicon strip detectors (SSDs) for particle tracking and a calorimeter to measure residual energy of individual protons. In this study, design parameters affecting pCT performances were characterized and optimized by GEANT4.9.6 simulations. The effects of the detector thickness and interval between detectors on the amount of multiple Coulomb scattering (MCS) and system sensitivity were characterized. Also, the initial proton energy affecting the accuracy of RSP was analyzed. Finally, the accuracy of RSP depending on the threshold level of the residual proton energy was investigated.

In this study we investigate optimised strategies of prompt gamma (PG) detections for Bragg peak (BP) tracking in proton radiation therapy. We expect that the PG signal with its spatial, spectral and timing characteristics could provide us the information of optimal PG detection strategy. Extensive Geant4 Monte Carlo simulations have been used to study the energy spectral, spatial and timing characteristics of PG emission signal from high-energy proton radiations. Cylindrical water and PMMA phantoms (F30cm x50cm) have been used in the simulation with a 200 MeV pencil proton beam. An ideal detection sphere surrounding phantom with 50 cm radius was introduced for both the registrations of PG signals and neutrons. Our preliminary results show that there exists an optimal energy window around 4.44 MeV for PG detection in both phantoms. The PG signal in this energy window has a better correlation with BP and higher PG gamma/neutron ratio. There is also a prominent PG emission around 110 degree with higher PG gamma/neutron yield ratio. The timing properties of PG emission also show a narrow time-of-flight (TOF) window around 3ns which is well differentiated from interference background of neutrons. These results indicate that there exists an optimized strategy for PG signal detection. Utilising these optimal energy window, angular window and time-of-flight window, PG image formation can be significantly improved for Bragg peak tracking.

Proton therapy is expanding due to its superior dose distribution with improved outcomes for some types of cancer and reduced side effects. While proton beam characteristics allow better beam conformity of the dose distribution, uncertainties in patient positioning can compromise treatment effectiveness. Therefore, there are ongoing efforts to develop and improve image guidance for proton therapy. For treatment planning in proton therapy, a CT data set of the patient is acquired and CT numbers are converted into relative stopping power (RSP). This conversion is a major source for range uncertainties, which are normally around 3.5%. Proton CT (pCT) has been proposed as a method to reduce this uncertainty and a scanner prototype has been developed. Besides treatment planning, pCT could also be useful for image guidance in the treatment room, due to its superior image quality, i.e., the absence of major artifacts, and lower dose than X-ray cone beam CT. The present study evaluates a pCT-to-pCT 3D registration algorithm with a realistic pediatric head phantom. An experimental pCT scan of the phantom, carried out with the Phase-II Head Scanner, was modified by prescribing 10 random 3D errors and applying the algorithm to correct them. The mean absolute residual errors were found to be less than 0.5 mm. Based on these preliminary results, registration of a pCT 3D head image acquired at the time of treatment to a pCT 3D reference head image should result in optimal positioning accuracy.

Proton computed tomography (pCT) is a promising alternative to conventional proton radiation therapy planning methods. We are carrying out tests of our pre-clinical Phase II pCT head scanner, which consists of a tracking detector and a multistage energy detector. It is capable of a readout rate of over one million proton histories per second. A specially designed step-shaped phantom allows us to collect all necessary calibration data for each step during a single experimental setup. In this study we apply Machine Learning methods to simplify offline processing of the calibration data and to improve the calibration for protons that deposited a significant part of energy at the interfaces of the energy detector stages. We describe the variables and training procedure and compare the performance with the original cut-based calibration approach.

As part of the BRAIN initiative, we are designing a mobile molecular brain imager that can be used on healthy subjects to study the functioning of the human brain during motion. The Ambulatory Micro-Dose Wearable PET Brain Imager (AMPET) must be as light as possible while still providing adequate sensitivity and resolution. Methods: We used a modified version of the Simulation System for Emission Tomography (SimSET) to estimate the sensitivity of head-mounted PET scanners with truncated sphere geometry to both whole-brain activity and to activity in mid-brain structures. With the equator of the truncated sphere set just above the lenses of the eyes, we investigated 10 and 12 cm radius spheres with axial truncations at the equator and 4 cm below for the lower truncation and at 4, 8, and, for the 12 cm sphere, 12 cm above the equator for the upper truncation (so the last truncation is a closed top sphere). Results: Whole brain sensitivity improved markedly as the detectors were extended in either direction axially, but sensitivity for two mid-brain structures, the hippocampus and amygdala, improved most markedly with extension of the sphere below the equator, which would cover the eyes. Conclusions: Optimal sensitivity may not be achievable for some biodistributions. In particular, eyes-free designs may have low sensitivity for some brain structures. Rotating the sphere towards the rear of the head will help for some target structures, like the hippocampus. In general, however, the utility of rotating the detectors will depend on the structure of interest and individual anatomy.

Drug delivery to solid tumors can be expressed physically using biomechanical phenomena such as convection, diffusion of drug in extracellular matrices, and drug extravasation from microvessels. Applying computational methods to solve governing conservation equations clarifies the mechanisms of drug delivery from the injection site to a solid tumor. In this study, multiple tumor geometries were obtained from PET/CT images. An advanced numerical method is used to solve fluid flow and solute transport equations simultaneously to investigate the effect of tumor size on drug delivery to lung tumors. Data from 20 patients with lung tumors were analyzed and the tumor geometrical information including size, shape, and aspect ratios were classified. An ideal bolus injection was considered for the model. The results show similar average drug concentration profiles in tumors with different sizes. Smaller tumors have higher drug concentration; however, the maximum difference between average drug concentrations is less than 10% for the smallest and largest tumors. The small differences in drug concentration in different tumor sizes was due to the minimum effect of convection in lung tumors.

X-ray imaging and external beam radiation therapy in cancer treatment are conducted with X-ray sources generating energy spectra. The photon beam is generally hardened by filtering out low energy photons. In radiotherapy, monoenergetic photons can be produced from radioactive sources and these are more predictable in their interaction in tissue by optimally targeting the tumors. In this work, we compared photon energy spectra against their corresponding effective energies by Monte Carlo simulations (MCS). Methods: We simulated with MCS the detection of a thin gold wire of 20 µm in diameter for spatial resolution assessment, and a water phantom, for contrast and noise assessment, made of 4 cylinders of 7 mm diameter and composed of air, fat, brain and bone. We also simulated a 28 g mouse for dosimetry in a lung tumor of 1.4 mm diameter. The energies used were 120 kVp and 50 kVp, and their corresponding effective energies of 44 keV and 33 keV. Results: The attenuation of the photons in the phantoms were different between the kVp energy spectra and their respective effective energy in keV. The spatial resolutions were found as 73, 70, 69 and 71 µm for a detector pixel resolution of 100 µm and respectively for 120 kVp, 44 keV, 50 kVp and 33 keV. In the contrast phantom, the contrast with respect to water reaches 10% in the bone for 120 kVp with respect to 44 keV and 13% for 50 kVp with respect to 33 keV. For the absorbed dose, the energy deposit in the tumor and other tissues behave differently between energies in kVp en in keV. Conclusions: It is expected to fundamentally describe phenomena by using monoenergetic radiations. However, since experimentations with X-rays provide energy spectra including characteristic X-rays, the simulations have to conform to the experimentations either in imaging or in radiation therapy. We have shown in this work the difference between the two approaches.

We are modifying the SimSET photon-tracking simulation to use angular response functions (ARFs) to speed up collimator response modeling. In this paper we report on the development of ARFs for Gallium-67. Ga-67 is clinically used for tumor imaging, but the quantitative accuracy is not well understood. Ga-67 is challenging to image, with gamma emissions at nine different energies from 90 to 900 keV. In our clinic, Ga-67 is imaged using three 20% FWHM symmetric energy windows, one at each of the three most abundant energy peaks (93.3, 184.6, and 300.2 keV). To date we have generated ARFs for a General Electric 670 SPECT/CT scanner using a MEGP collimator for each energy window at each of Ga-67’s emissions peaks. We found significant contamination from collimator penetration and scatter. The highest energy emission, 887.8 keV, contributes significantly despite its low abundance (0.1450% mean emissions/decay); for instance, in the energy window centered at 300.2 keV, contributes over 10% as many counts as geometrically collimated 300.2 keV photons. In total, collimator penetration and scatter contributes increase detected counts by around 12.58% in the 93.3 keV window, 16.52% in the 184.6 keV window, and 50.88% in the 300.2 keV window. ARFs are derived from simulation of point sources in air; in patient imaging, there will be significant additional contamination from photons scattered in the patient. Thus the contamination from collimator penetration and scatter derived from ARFs significantly understates the full impact of penetration and scatter; we will be investigating this aspect once SimSET has been fully modified to use ARFs.

Monolithic detector is a promising design for PET due to its low cost, high packing fraction, intrinsic depth-of-interaction (DOI) capability as well as potentially better energy and timing resolution compared to the conventional segmented detector design. In monolithic PET detector design, choosing an appropriate crystal surface treatment method plays a crucial role in achieving good positioning, energy and timing performance as it determines the optical photon transportation process in the crystal. To optimize crystal surface treatment of monolithic detector for better detector performance, we studied the positioning, energy and timing performance of a monolithic detector under five different crystal surface treatments using GATE Monte Carlo simulation. The detector positioning performance was evaluated using the Cramer-Rao lower bound to eliminate the influence of position estimation algorithm and thus to achieve an objective comparison between different crystal surface treatments. The results demonstrate that detector with crystal surface painted with black film has the best positioning but the worst energy and timing performance among the five crystal surface treatments studied. A diffuse coating film will lead to an improved positioning performance compared to specular coating film regardless of crystal surface roughness. As for energy performance, there is no observable difference between different crystal surface treatments as long as the reflectivity of the coating film is the same. Detector with specular coating film shows slightly better timing performance than that with diffuse film. Based on the results, we conclude that polished or rough crystal coated with diffuse film is a promising crystal surface treatment for monolithic detector design due to its best overall performance in terms of positioning, energy and timing. This work also provides guidance in choosing crystal surface treatment to facilitate monolithic detector design.

One important application of Positron Emission Tomography (PET) is for high resolution brain research to obtain high resolution imaging of the neurophysiological activity for human brain research. Compared with functional magnetic resonance imaging (fMRI), PET can provide relatively silent condition so that the brain activity will not be disturbed by acoustic noise. Conventional whole body or brain dedicated PET systems require patients to hold in one position for long time to reduce the motion effect. Their bore size has to be larger than then human head and requires more detectors to cover the brain region.The objective of this work is to develop a wearable brain PET system which allows patients move within limited places has been developed. Two wearable PET designs with 14 mm and 7 mm crystal length were simulated and compared. Spatial resolution and sensitivity are taken into consideration to evaluate system performance. Thinner crystal has better performance in spatial resolution at a cost of poor sensitivity. Radial resolution drops dramatically near the edge of FOV. In future work, a DOI PET detector will be evaluated in system design to eliminate this effect.

Monolithic scintillator detectors have been shown to provide various advantages in PET systems. An accurate light distribution model is important for investigating the monolithic detector performance and develop event location estimation methods. We proposed a generic analytical model for monolithic scintillator with different crystal surface treatments, including specular crystal surface combined with black films (BS), with specular films (SS), and with diffuse films (DS). This model takes into account detector geometry, reflection and refraction in crystal surfaces, light absorption within crystal and surface treatment methods. Monte Carlo simulation was used to validate accuracy of this model. Two indices were defined to measure difference between the proposed model and simulation: correlation coefficient (CC) for measuring difference in shape of light distribution, and normalized mean square error (NMSE), to measure difference in value of light distribution. Excellent agreement (CC above 0.99 and NMSE below 0.13) achieved in all investigated cases, indicating the accuracy of this model. This work provides an accurate and fast generic analytical model for calculating optical photon transportation in monolithic PET detector. It has very promising potential in position information extraction for monolithic scintillator detector, and is useful to provide guidance in design of monolithic scintillator detector.

The objective of this work is to derive a formula - given the key parameters of a typical cylindrical PET scanner – for calculating the spatial resolution at any point in the field of view. This formula is to bridge the gap between the cumbersome full system modeling (Monte Carlo or other) methods and the existing over-simplified analytical resolution formula that ignores parallax effect. We extend an analytical line-of-response geometrical model to a closed form expression with which the full-width-at-half-maximum (FWHM) of the geometrical response of any line-of-response (LOR) at any given point on the LOR can be calculated. By systematically changing the values of 4 variables - scintillator crystal width and length, photon incidence angle on crystal, the location (depth) along LOR, a series of FWHM values are calculated for deriving the formula. Other factors contributing to system spatial resolution, that is, positron range, annihilation photon’s acolinearity, and detector multiplex-coding are considered to be known quadratic terms and ready to be integrated with the geometric response. The calculated response functions and their corresponding FWHM values show distribution characteristics agreeing with existing knowledge. Derivation of the formula is being pursued through analyzing the FWHM’s dependency on the identified variables and parameterizing the relationship between FWHM and the variables.

F-18 Sodium Fluoride (NaF) has been investigated as an imaging biomarker for coronary vulnerable plaque. The goal of this work is to simulate the effects of plaque uptake, image noise, respiratory and cardiac (R&C) motion, and R&C gating techniques on the detectability of coronary artery plaque. PET projection data from the 4D XCAT phantom with typical organ uptake and realistic modeling of R&C motion were generated using Monte Carlo simulation. Plaque with multiple dimensions and 6 plaque-to-background (P/B) uptake ratios were placed at 3 specific locations in the coronary arteries (LAD, LCX, and RCA) of the beating heart of the XCAT phantom. In addition, 4 respiratory motion (RM) amplitudes ranging from 6 to 15 mm with 6 image noise levels including almost noise-free simulations were generated. For each projection dataset, reconstructed images without any gating, with only 6-frame respiratory gating (RG), only 8-frame cardiac gating (CG), and dual R&C gating were used to evaluate the detectability of plaque. In reconstructed images without RG and without CG, only the LAD plaque was barely visible in the almost noise-free case. All 3 plaques were visible in dual R&C gated images in the almost noise-free case, the visibility decreased dramatically with increase of image noise. The LAD and LCX plaque were barely visible in the almost noise-free RG images with 71% contrast reduction, while all 3 plaques were visible in the CG images even with the largest 15 mm RM amplitude but with 50% contrast reduction. Larger RM required higher P/B uptake ratio in order to detect the plaque. In all gating schemes, plaques with lower P/B uptake ratio required low noise level to be detected. Results of this study provide useful guides for coronary artery plaque PET imaging. The LAD plaque was less sensitive to RM and CM and was the easiest to detect. Dual RG and CG reduces blurring at the cost of increased image noise. CG provides improved plaque visibility over RG.