M5A2  Signal and Image Processing

Friday, Nov. 6  08:30-09:45  Pacific Salon 1&2

Session Chair:  Arkadiusz Sitek, Philips Research North America, United States; Andrew Reader, King's College London, United Kingdom

Show/Hide All Abstracts

(08:30) M5A2-1, PET/MR Attenuation Correction in Brain Imaging Using RESOLUTE: a MR-AC Method Employing Continuous Bone Signal Derived from UTE Imaging

C. Ladefoged, D. Benoit, I. Law, S. Holm, L. H�jgaard, A. E. Hansen, F. Andersen

Department of Clinical Physiology, Nuclear Medicine, and PET, Rigshospitalet, Copenhagen, Denmark
In the absence of transmission sources in combined clinical PET/MR systems, MR images are used for MR-based attenuation correction (MR-AC). The main challenge of MR-AC is to separate bone and air, as neither have a signal in the traditional MR images, and to assign the correct linear attenuation coefficient of bone. The ultra-short echo time (UTE) MR sequence was proposed as a basis for MR-AC as this sequence shows a small signal in bone. The segmented UTE MRI signal is assigned a single attenuation coefficient for bone in the vendor provided attenuation map. The purpose of this study was to develop a new segmentation-based MR-AC method (MR-ACUPSTAIRS) with continuous bone values, and evaluate it on a large patient cohort. METHODS. 154 [18F]-FDG PET/MR patients were included in this study. PET/MR data were acquired over a single bed position of 25.8 cm covering the head and neck. MR-AC was based on either standard MR-ACCT, MR-ACUTE or MR-ACUPSTAIRS. Our method segments the air, brain, cerebral spinal fluid, and soft tissue voxels on the raw UTE images, and uses a mapping of R2* values to CT HU to measure the density in bone voxels. The reconstructed PET images were evaluated in the whole brain, as well as regionally using a ROI-based analysis. RESULTS. Joint histogram of the intensities in the PET image produced with our method shows high correlation to the PETCT image (r2=0.92), compared to UTE (r2=0.78). The average error in the brain is 0.1%, and less than 1.2% in all regions of the brain. On average 95% of the brain is within �10% of PETCT, compared to 72% when using UTE. CONCLUSIONS. The proposed method uses the available UTE images to segment tissue classes, and uses the R2* map to measure a continuous bone signal. The improvements over the vendor provided UTE reduces both the global and local error on the reconstructed PET images, as well as limits the number and extent of the outliers.

(08:45) M5A2-2, Evaluation of a BGO-Based PET System for Single Cell Tracking: Simulation and Phantom Studies

Y. Ouyang, T. J. Kim, K. S. Lee, G. Pratx

Radiation Oncology (Physics), Stanford University, Palo Alto, CA, USA
Background � Current in vivo imaging methods are unable to track single cells throughout the whole body. A recent method was developed for tracking a moving positron emitter, and reconstruction of trajectories was demonstrated with a simulated Inveon preclinical PET system. However, lutetium oxyorthosilicate (LSO) background activity was not simulated. Here, we explore the feasibility of single cell tracking using the Genisys4, a PET system based on bismuth germinate which has no intrinsic background and may be better suited to tracking low activity. Methods � Systems (Genisys4, Inveon with and without background from real data) were simulated in GATE with static and helically moving 18F point sources. The position-dependent sensitivity in the FOV of the Genisys4 was simulated using a static source. Trajectories of moving sources of varying activity and velocity were reconstructed directly from list mode data by minimizing the distance between recorded lines of response and the trajectory estimate, represented as a B-spline function. Tracking with the Genisys4 was validated using a phantom of helically moving droplets (< 0.5 mm diameter) of [18F]FDG (1-2.5 mCi/cc). Results � Localization error of the Genisys4 simulation data was comparable to that of the Inveon (within 0.2 mm for the region of activity and velocity where localization error was < 2 mm) without LSO background. With added background, data suggests that localization error with the Inveon is adversely affected. Trajectories of several [18F]FDG droplets were reconstructed in phantom studies with the Genisys4. Correlation between the simulated sensitivity and the measured activity was R = 0.86 in a representative example. Conclusion � We have investigated the feasibility of tracking the movement of small point source-like objects with the Genisys4. These results suggest the merit of tracking low activity, single cells with this system.

(09:00) M5A2-3, Practical Time Mark Estimators for Multichannel Digital Silicon Photomultipliers

E. Venialgo1, S. Mandai1, T. Gong1, D. Schaart2, E. Charbon1

1CAS, TUDelft, Delft, Netherlands
2Radiation Science and Technology, TUDelft, Delft, Netherlands
Multiple time-to-digital converters coupled with silicon photomultipliers allow to timestamp several light photons generated by a scintillation event. Multichannel digital silicon photomultipliers opened the possibility to estimate a gamma-photon time mark by using several photoelectrons timestamps. We studied the already-existing statistics models of photoeletron time-stamping generation, while extending the current models by adding the skipping effect. Which accounts for the inability of the system to timestamp a continuous set of photoelectrons. In addition, we proposed two multiple photoelectron timemark estimators based on the best linear unbiased and the maximum likelihood estimation methods. We calculated the Cram\'er Rao lower bound for several system parameter and compared it to the proposed estimators' performance. We concluded that under certain system configurations the proposed estimators are efficient. Moreover, we investigated the effect of dark count rate on the timing performance. Also, we introduced a filtering method that is based on measuring the time distance between adjacent timestamps. We performed a full Monte Carlo simulation to evaluate the proposed filter efficiency. Finally, we performed a full Monte Carlo simulation to compare the timemark estimators' performance. We concluded that the best linear unbiased estimator is as efficient as the maximum likelihood estimator. In addition, it was verified that multichannel digital silicon photomultipliers have a stronger tolerance to dark counts in comparison with current digital silicon photomultiplier architectures.

(09:15) M5A2-4, Data-Driven Respiratory Signal Extraction for SPECT Imaging Using Laplacian Eigenmaps

J. C. Sanders1, P. Ritt2, T. Kuwert2, A. H. Vija3, J. Hornegger1

1Pattern Recognition Lab, University of Erlangen-Nuremberg, Erlangen, Germany
2Clinic of Nuclear Medicine, University Hospital Erlangen, Erlangen, Germany
3Siemens Molecular Imaging, Hoffman Estates, IL, USA
In Single Photon Emission-Computed Tomography (SPECT) imaging, respiratory motion leads to blurring and loss of quantitative accuracy. A respiratory surrogate signal is useful for management of this motion, and data-driven solutions requiring no additional hardware are desirable. This work compares two data-driven methods based on dimensionality reduction: Principle Component Analysis (PCA) and Laplacian Eigenmaps (LE). Our aim is to apply both to SPECT and assess the feasibility of data-driven gating for this modality. We expect that LE, which is less sensitive to outliers in data, will outperform PCA at high levels of image noise. Two phantom acquisitions were performed: one in which a sphere in cold background was translated axially by a piston actuator (dynamic), and a warm background with no sphere (static). Using binomial subsampling, both datasets were combined at various Signal to Noise Ratios (SNRs, defined as ratio of mean dynamic count density to square root of mean static count density). LE and PCA surrogate signals were computed and compared via Pearson’s correlation to the truth signal obtained from the actuator. As a follow-up, LE and PCA estimates from 27 cardiac SPECT acquisitions were compared to a simultaneously acquired signal from a pressure sensor embedded in an elastic belt. In the phantom experiment, correlations between LE/PCA and truth were >0.9 for all SNR>5. For SNR<5, PCA deteriorated rapidly, whereas LE remained stable through SNR=2.5. For the patient validation, LE and PCA yielded average correlations of 0.86±0.14 and 0.37±0.26, respectively. The phantom experiment indicated that LE outperforms PCA for low-SNR data. This conclusion was supported by the superior performance of LE for patient datasets, where noise may be high. However, it is limited by the simplistic motion present in the phantom experiment and the limited scope of the patient validation.

(09:30) M5A2-5, Spatially Encoded Joint Entropy Prior for PET Image Deblurring

J. Dutta, G. El Fakhri, Q. Li

Center for Advanced Medical Imaging Sciences, Department of Radiology, Massachusetts General Hospital, Boston, MA, USA
The quantitative accuracy of PET is degraded by partial volume effects caused by the limited spatial resolution capabilities of PET scanners. We describe here an image deblurring technique that relies on the joint entropy (JE) between the PET image and an anatomical image to stabilize the deconvolution of the PET image using a spatially varying point spread function. One challenge associated with the use of a JE prior for PET image deblurring is the formation of spurious peaks in the joint probability density function. In this paper, we present a spatial encoding scheme that leads to a weighted JE regularizer which suppresses the effect of the spurious peaks. To validate this method, we have performed simulations on a digital phantom. Our results show that the resultant technique reduces mean squared error in the deblurred PET image. We also show that the spatially encoded JE prior is more robust than ordinary JE in the presence of structural discrepancies between the PET and the anatomical images and suppresses artifacts arising from such discrepancies.