Discounted short course fees are available
for some students and postdocs. Please click here for details.
Radiation
Detection and Measurement
Sunday and Monday, November 10-11, 2002
Time: 8:30 AM - 5:00 PM
Organizer: |
Prof. Glenn F. Knoll,
University of Michigan |
Instructors: |
Stephen Derenzo, Lawrence
Berkeley National Laboratory
Eugene Haller, UC Berkeley & Lawrence Berkeley National Lab
Glenn Knoll, University of Michigan
Fabio Sauli, CERN, Geneva
Helmuth Spieler, Lawrence Berkeley National Laboratory |
ABSTRACT
This 2-day course provides a short review of the basic principles
that underlie the operation of the major types of instruments
used in the detection and spectroscopy of charged particles,
gamma rays, and other forms of ionizing radiation. Examples both
of established applications and recent developments are drawn
from areas including particle physics, nuclear medicine, and
general radiation spectroscopy. Emphasis is on understanding
the fundamental processes that govern the operation of radiation
detectors, rather than on operational details that are unique
to specific commercial instruments. Topics are also included
on the pulse processing techniques that are needed to properly
record the information provided by the detection devices. This
course does not cover radiation dosimetry or health physics instrumentation.
The level of presentation is best suited to those with some prior
background in radiation measurements, but can also serve to introduce
topics that may be outside their experience base.
OUTLINE
1. General Properties of Detectors
2. Gas-Filled Detectors
3. Scintillation Counters
4. Semiconductor Detectors
5. Pulse Processing and Analysis
6. Summary and Intercomparison, New Detector Developments
The following will be provided as part of the course:
Class notes
Textbook - "Radiation Detection and Measurement", 3rd
Edition, by G. Knoll
Lunch both days
Refreshments at the morning and afternoon breaks
Certificate of completion
Fee: $400 for IEEE Members
Return to top of page
Triggering in Particle
Physics Experiments
Sunday, November 10, 2002
Time: 8:30 AM - 5:00 PM
Organizer: |
Peter Wilson,
Fermi National Accelerator Laboratory |
Instructors: |
Sidhara Dasu
- Univ of Wisconsin (BABAR, CMS)
Levan Babukhadia - SUNY at Stoneybrook (D0)
Giovanni Punzi - INFN Pisa (CDF) |
ABSTRACT
A critical component of particle physics experiment design
is determining which events to store for further analysis and
how to make that decision. The job of the Trigger is to quickly
discard uninteresting events while efficiently culling the most
interesting events in as unbiased a manner as possible. In most
experiments the rate at which detector data is sampled, such
as beam crossing rate for a colliding beam experiment, is much
higher than the rate of physics interactions of primary interest.
At the same time the volume of data from digitizing all readout
channels is frequently too high to be practically read-out by
a data acquisition system (DAQ) for later analysis let alone
be fully reconstructed in real time. Some reduction of needed
bandwidth can be achieved within the DAQ system by suppressing
channels with no interesting data or other data compression methods.
While sparsification can reduce the needed bandwidth by factors
of 10 or 100, suppression by factors of a million are often achieved
with a combination of triggering and data compression. For example,
the Run II trigger systems for the CDF and D0 experiments at
the Fermilab Tevatron collider each reduce an input of about
10 TBytes/sec to an output of about 20 Mbytes/sec recorded on
tape. This course will discuss the design of trigger systems
for particle physics ranging from cosmic ray experiments to future
colliding beam experiments such as those at the Large Hadron
Collider.
The course will cover overall trigger system design with particular
attention to impact of beam environment and data acquisition
design. It will also cover the design of trigger subsystems which
do fast partial event reconstruction and pass information to
more global decision hardware. This process is often referred
to as generating trigger primitives. The focus will be on primitives
that are common to many modern HEP experiments: charge track
reconstruction, calorimeter and muon triggers. Also covered will
be systems to reconstruct tracks from detached vertices which
is a more recent and complicated task. Specific examples from
past, current and future experiments will be used to illustrate
the techniques of each topic and the progression of those techniques
with improving technology. Comparisons will also be made for
different types of experiments (e.g. cosmic ray, fixed target,
colliding beam).
The overall system design of the trigger is very closely coupled
to the structure of the beam of the experiment. The particle
type, beam energy and timing structure all have a large impact
on the rate of particle interactions both interesting (signal)
and uninteresting (background). Beam environment can vary from
neutrinos produced in the sun (no timing structure) to proton-anti-proton
collisions in bunches separated by 25ns. The complexity of the
detector systems also impact the trigger design: what types of
events will the experiment need to detect, how many channels
are there? The trigger design is also intimately related to the
DAQ architecture since the DAQ must feed data to the trigger
and the trigger must tell the DAQ what to do with the data. We
will discuss how these issues impact the trigger system design.
For example, how many decision levels are needed, which levels
will be implemented with hardware, which levels with software
and which as combination of hardware and software. We will show
how these decisions have changed over time with improving technology
(impact of Moore's Law).
The design of subsystems to generate trigger primitives must
be closely connected to the design of the detector subsystems
and the front-end electronics which read them out. These systems
do fast reconstruction of event data with a very focussed purpose.
To minimize execution time, only certain classes of objects are
reconstructed (e.g. Tracks above a minimum momentum threshold).
We will focus on reconstruction of physics objects in lower level
parts of trigger systems. Since most Level 3 triggers are based
on computing farms running off-line type reconstruction code,
the design requirements are not particularly unique to the trigger
application. The most frequently used trigger primitives are
from calorimeters for electron, photon and jet reconstruction
along with muons from muon detectors. These have provided and
continue to provide standard signatures for many types of particle
decays. We will cover these along with the next most common trigger
type from reconstruction of charged particle tracks in tracking
chambers. These charged tracks are used on their own or matched
to objects found in calorimeters (e.g. electrons) or muon detectors.
Recently, very powerful trigger processors have been developed
to exploit the long lifetime of heavy quarks (b or c quarks)
from the presence of displaced tracks or detached track vertices.
These detached vertex triggers are very challenging but are already
revolutionizing triggering in hadron collider experiments.
OUTLINE
System Design for HEP Triggers - Lecturer: Peter Wilson
- I. What does the Trigger do?
II. Design Constraints - DAQ Bandwidth vs Physics Rates
- · Types of Experiment (eg Cosmic Ray, Fixed Target,
Collider)
· Beam Structures (DC, Pulse trains etc)
· Cross-section of Physics of interest vs backgrounds
· Basic methods of event selection - particle identification
and
· kinematics etc
· Need to select data of greatest interest for physics
analysis
· Bandwidth, Bandwidth, Bandwidth
- III. Intimate relationship with DAQ system
- · DAQ readout bandwidth
· data storage and transfer architecture
· data sparsification
· Impact of Trigger on DAQ design
- IV. Need for and design of multi-level triggers
- · Tied directly to DAQ capabilities
· Rate reduction too large to be done in single fast processing
step
- V. Differences for different types of experiment:
- · non-accelerator experiments
· fixed target experiment
· e+e- collider
· ep collider
· Hadron collider
- VI. Impact of Trigger design on DAQ and Front-End Electronics
Designs
VII. Design for testing and validation of Trigger operation
- · Measurement of Trigger Efficiency
· Readout of data from Trigger decision process
· Trigger Hardware as Data Source for DAQ
- VIII. Historical Perspective
- · Progression of HEP accelerator luminosities
· Improvements in Experiment bandwidths
Charge Particle Track Processors for Hardware Triggering -
Lecturer: Levan Babukhadia
- I. Relationship of tracking algorithms to detector architecture
II. Relationship of hardware implementation to front-end and
DAQ designs
III. Technological progression of Track Processors
- · Fixed target to colliding beam
· With electronics capabilities
- IV. Specific implementation examples/differences
- · e+e- (eg Babar, Belle)
· p-pbar (CDF, D0)
· Being built (LHC)
- V. What the future may bring?
Triggering on Particle Types: Calorimeter and Muon Based Triggers
- Lecturer: Sidhara Dasu
- I. Calorimeter based triggering
- · Electrons, Photons, and Jets
· Global energy triggers
· Energy clustering algorithms
- II. Triggering on Muons
- · Triggering on Muon detectors alone
· Connection to charge track processors (matching)
- III. Relationship of hardware implementation to front-end
and DAQ designs
IV. Specific implementation examples/differences
- · Fixed target
· e+e- (eg Babar, Belle)
· p-pbar (CDF, D0)
· Being built (CMS, CMS)
- V. Technologies: ASICs, FPGAs,...
VI. What may the future bring?
Triggering on Tracks from Detached Vertices - Lecturer: Giovanni
Punzi
- I. Technological progression of vertex triggers Track Processors
II. Relationship of tracking algorithms to detector architecture
(eg strips vs pixels, forward vs Central Geometry)
III. Relationship of hardware implementation to front-end and
DAQ designs
IV. Hardware Implementation differences: ASICs, FPGAs, Commercial
Processors
V. Specific Implementation examples/differences
- · Fixed Target Experience?
· CDF and D0 Silicon Vertex Trackers
· BTeV Detached Vertex Trigger
The following will be provided as part of the course:
Class notes
Lunch
Refreshments at the morning & afternoon break
Certificate of completion
Fee: $280 for IEEE Members
Return to top of page
Integrated
Circuit Front Ends for Nuclear Pulse Processing
Monday, November 11, 2002
Time: 8:30am - 5:00pm
Organizer: |
Chuck Britton, Oak
Ridge National Laboratory |
Instructors: |
Veljko Radeka, Brookhaven
National Laboratory
Paul O'Connor, Brookhaven National Laboratory
Alan Wintenberg, Oak Ridge National Laboratory |
ABSTRACT
This one-day course will cover integrated circuits developed
for nuclear pulse processing applications with an emphasis on
charge measurement. We will discuss bipolar and MOS transistor
operation, signal processing for pulse measurements, charge-sensitive
preamplifiers, photomultiplier preamplifiers, pulse-shaping circuits,
sample/holds, and analog/digital converters.
This course has been targeted to three types of attendees.
The first is the engineer/physicist who desires understanding
of the basics of integrated circuits and pulse-shaping networks
in order to begin creating circuits for systems. The second is
the engineer/physicist/manager who needs to be able to understand
the basics of these technologies and their achievable performance
in order to manage or work with a development team utilizing
these technologies. The third type is one who desires an overview
for personal technical development.
The morning session will be an overview of the theory of pulse
processing from a theoretical viewpoint. It will cover noise
sources and pile up and their effect on resolution. Charge-sensitive
preamplifiers and their design in integrated circuit processes
will be covered with an emphasis on implementation.
The afternoon session will cover integrated circuits for photomultiplier
tube readout and associated circuits for the system aspects such
as variations of gain and timing. Analog/digital converters and
their associated circuitry (sample/hold and peak stretchers)
will be discussed.
In all cases, numerous examples will presented of the present
state-of-the-art.
The following will be provided as part of the course:
Class notes
Textbook - "Analog Integrated Circuit Design", by David
Johns and Ken Martin
Lunch
Refreshments at the morning & afternoon break
Certificate of completion
Fee: $340 for IEEE Members
Return to top of page
Nuclear Emission
Imaging Detectors, Systems and Methods for Breast Cancer Evaluation
Monday, November 11, 2002
Time: 1:00pm - 5:30pm
Co-Organizers: |
Martin Tornai, PhD
Duke University Medical Center, Department of Radiology
Duke University, Department of Biomedical Engineering
Craig Levin, PhD
UC San Diego School of Medicine, Department of Radiology
San Diego VA Medical Center, Nuclear Medicine Division |
Instructors: |
James Bowsher, PhD, Duke
University Medical Center,
Department of Radiology |
ABSTRACT
In recent years, the possibility of using nuclear emission
imaging to alleviate some drawbacks of conventional methods for
detection, diagnosis and staging of breast cancer has been an
active field of research. Dedicated nuclear cameras used in conjunction
with breast cancer specific radiotracers offer the potential
for more specific and sensitive identification of breast cancer
than conventional imaging techniques. Drawbacks of standard clinical
nuclear imaging methods for breast imaging such as planar scintigraphy,
Single Photon Emission Computed Tomography (SPECT), and Positron
Emission Tomography (PET) are that the all-purpose camera systems'
geometry and performance are not optimized for breast cancer
imaging. In addition, the relatively expensive all-purpose cameras
keep study costs high compared to standard breast imaging techniques,
which raises questions of cost-effectiveness. For these reasons
there has been great interest in development of dedicated breast
imaging systems and techniques. Through close-proximity breast
imaging and new detector materials, components and configurations,
such systems extend the performance limits available to nuclear
imaging.
This course is designed for the scientist and engineer who
wants to learn more about or review the details of issues specific
to breast imaging with nuclear emission cameras. This discussion
includes system development issues relevant to both single and
coincident photon imaging systems. Issues relevant to both conventional
clinical and dedicated breast imaging systems will be covered.
The course begins with a discussion of basic detector design
issues for breast imaging with nuclear emission cameras. Practical
information on how to build a dedicated breast imaging system
will be covered such as detector components, electronics, and
event positioning algorithms. Next, we present a thorough discussion
of recent systems and methods that have been developed by various
researchers in the field for nuclear breast imaging. A comparison
of the variety of different approaches will give the course attendee
perspective on the important system issues under consideration.
Data generated from phantom studies will be presented to understand
the limitations of the various approaches. Practical clinical
imaging applications of these systems and methods will also be
presented to demonstrate the utility of these systems. The session
ends with a comprehensive discussion on complete-orbit and image
reconstruction issues relevant to nuclear emission breast imaging
systems. The particular geometry of the breast in relation to
the detector systems yields some very unique problems and solutions
for tomographic image reconstruction.
OUTLINE
Detector Design Issues for Breast Imaging with Nuclear Emission
Cameras
(Craig Levin)
- I. Overview of Breast Imaging with Nuclear Cameras
- A. Motivation
- B. General Design Principles
II. Scintillation Detector Designs of Nuclear Imagers
- A. Scintillation Crystal Design
- B. Collimation Schemes for Scintillation Crystal Designs
- C. Photodetector Design
- D. Electronic Readout of Position Sensitive Photodetectors
- E. Electronic Processing and Data Acquisition for Imaging
- F. Event Positioning Schemes and Image Formation
-
- III. Semiconductor Detector Designs of Nuclear Imagers
- A. Semiconductor Crystal Materials
- B. Semiconductor vs. Scintillation Crystal Designs
- C. Semiconductor Imaging Array Configurations
-
- IV. Promising Detector Designs for Dedicated Nuclear Breast
Imagers
- A. Small FOV Gamma Ray Imagers
- B. Small FOV Coincidence Imagers
Systems Approaches to Nuclear Emission Breast Imaging
(Martin P. Tornai)
- I. Overview - Imaging Approaches
- A. Primary Concerns of Nuclear Imaging Systems
- B. Single Photon Planar Imaging
- C. Single Photon Emission Computed Tomographic Imaging
- D. Coincident Photon Planar Imaging
- E. Coincident Photon Tomographic Imaging
- F. Useful "Add-On" Features of Systems
-
- II. Single Photon Planar Imaging (GEM, PhEM, Scintimammography,
SPEM)
- A. Overview of Uncompressed and Compressed Breast Imaging
- B. Clinical Scintimammography Cameras & Systems
- C. Dedicated, Compact Cameras & Systems
- D. Needle Biopsy Guidance with Compact Systems
-
- III. Mammotomography with Single Photon Emission Computed
Tomography (ASETT, PICO-SPECT, RSH-SPECT, SPECT)
- A. Overview of Uncompressed Breast Imaging
- B. Clinical SPECT Cameras & Systems
- C. Compact SPECT Cameras & Systems
- D. Specialized SPECT Systems
-
- IV. Coincident Photon Planar Imaging (PEM)
- A. Overview of Uncompressed and Partially-Compressed Breast
Imaging
- B. Dedicated, Compact Cameras & Systems
- C. Needle Biopsy Guidance with Compact Systems
-
- V. Mammotomography with Coincident Photon Tomographic Imaging
(B-PET, mammoPET, maxPET, PET)
- A. Overview of Uncompressed and Partially-Compressed Breast
Imaging
- B. Clinical PET Cameras & Systems
- C. Dedicated, Compact PET Cameras & Systems
-
Image Reconstruction and More Nearly Complete Orbits in Mammotomography
(James E. Bowsher)
- I. General Perspective on Image Reconstruction and Orbits
for Mammotomography
-
- II. Orbits for Nearly Completely Sampling the Breast Region
with SPECT
- A. Parallel-Hole Collimation
- 1. Basic resolution-sensitivity characteristics of parallel-hole
collimation
- 2. Orlov's condition
- 3. Parallel-hole, horizontal axis of rotation orbits
- 4. Parallel-hole, vertical axis of rotation orbits
- 5. Slanted parallel-hole collimation
- B. Pinhole Collimation
- 1. Basic resolution-sensitivity characteristics of pinhole
collimation
- 2. Tuy-Smith conditions
- 3. Partial-circle, 2D pinhole orbits
- 4. 3D pinhole orbits
- 5. Combined pinhole/parallel-hole configurations
- C. Cone Beam Collimation
- 1. Basic resolution-sensitivity characteristics of cone-beam
collimation
- 2. Cone beam orbits
- D. Connections to PET Mammotomography
-
- III. Reconstruction Issues for Mammotomography
- A. The value of modeling scatter, attenuation, and detector
geometry within iterative reconstruction for SPECT and PET
- B. Iterative, statistical reconstruction of multi-modality
acquisitions using highly informative models of cross-modality
a priori information
The following will be provided as part of the course:
Class notes
Refreshments at the afternoon break
Certificate of completion
Fee: $175 for IEEE Members
Return to top of page
Multi-Modality
Imaging Devices
Tuesday, November 12, 2002
Time: 8:00am - 12:30pm.
Organizer: |
David W Townsend, Ph.D.
Department of Radiology
University of Pittsburgh |
Instructors: |
Bruce Hasegawa, Ph.D.
Professor, Physics Research Laboratory
University of California at San Francisco
Simon R Cherry, Ph.D.
Professor, Department of Biomedical Engineering
University of California at Davis
|
ABSTRACT
The
importance of aligning image sets from two different modalities
in regions of the body other than the brain has long been recognized,
particularly where the modalities represent complementary aspects
of disease. Functional imaging modalities such as PET and SPECT
offer little anatomical localization, whereas anatomical imaging
modalities such as CT or MR generally contain very little functional
information. However, the imaging of function, accurately localized
within an anatomical framework, could offer a powerful approach
to the diagnosis and staging of disease, and the monitoring of
treatment. Despite increasing sophistication, software fusion
techniques cannot compete outside the brain with the convenience
and accuracy of a hardware approach where the imaging technologies
themselves are fused, rather than the images registered post
hoc. This course will review the motivation for combined functional
and anatomical imaging, particularly emphasizing the areas in
which the software approach can be problematic. The recent development
of combined SPECT/CT and PET/CT designs for imaging patients,
and SPECT/CT, PET/CT and PET/MR designs for imaging small animals,
will be presented, summarizing the unique challenges created
by the differing scale of the animal and human instrumentation.
The use of the anatomical µ-map to correct the functional
data for photon attenuation is also a key feature of these devices.
Finally, the clinical impact of the new systems will be assessed
and illustrated with patient studies in oncology and cardiology.
The
following will be provided as part of the course:
Class notes
Refreshments at the morning break
Certificate of completion
Fee: $175 for IEEE Members
Return to top of page
Analytical Reconstruction Methods
Tuesday,
November 12, 2002
Time: 8:00 a.m.-12:30 p.m.
Organizer: |
Michel Defrise, Department
of Nuclear Medicine
Vrije Universiteit Brussel, Brussels, Belgium. |
Instructors: |
Pierre Grangeat, LETI, CEA-DTA,
Grenoble, France.
Frédéric Noo, Department of Radiology, University
of Utah. |
ABSTRACT
Analytic reconstruction methods describe the unknown image
and the data as continuous functions, and model the data acquisition
by a transform operator mapping the image onto the data. In SPECT,
PET and CT, this operator is the Radon or x-ray transform in
two and three dimensions. Explicit inversion formulae for these
operators are discretized to obtain algorithms which take into
account the sampling of the data and of the image. Beside providing
a unique insight into issues such as sampling, stability and
data sufficiency, analytic algorithms are the methods of choice
whenever the data set and the image matrix are too large to apply
iterative reconstruction techniques. Despite an already long
history, the academic and industrial research on analytic methods
is still extremely active and has recently produced remarkable
solutions to problems which had been open for many years.
The course will provide an overview of the reconstruction
methods which are currently used in clinical scanners, as well
as of the most recent advances in this field. It will be assumed
that the attendees have some prior knowledge of the basic principles
of 2D tomography (notes will be provided beforehand), but these
principles will nevertheless be carefully summarized. The course
will then concentrate on more advanced topics, especially 3D
reconstruction in PET and spiral CT, and dynamic (4D) reconstruction.
OUTLINE
I. Image reconstruction from parallel-projections (M.
Defrise)
2D and 3D x-ray transform, relevance for PET and SPECT, Fourier
properties (central section theorem and direct Fourier reconstruction),
data sufficiency condition (Orlov), Filtered-backprojection (FBP),
truncated 3D data and the Reprojection algorithm.
Rebinning algorithms for 3D PET. Single-slice rebinning, approximate
and exact Fourier rebinning, hybrid algorithms and the statistical
distribution of rebinned data.
II. Image reconstruction from divergent projections
(F. Noo)
Fan-beam FBP, Grangeat's theorem, data sufficiency (Tuy),
the RADON algorithm, 3D cone-beam FBP for non-truncated cone-beam
projections, redundancy and the reduction to the ramp filter,
Circular acquisition (approximate Feldkamp's methods).
Exact methods for helical cone-beam CT: PI-lines, Tam's window,
Grangeat's formula for truncated projections, long-object problem.
Cone-beam FBP algorithm of Katsevich and implementation details.
Rebinning methods for helical cone-beam CT: advanced single-slice
rebinning and related methods (AMPR,..).
III. Reconstruction of dynamic tomographic data (P.
Grangeat)
Temporal filtering of image sequences, sliding window principle,
affine motion compensation, gated tomography, reconstruction
of periodic motion, voxel specific motion compensation.
Applications to CT fluoroscopy, cardiac tomography and image
image reconstruction for radiotherapy.
The following will be provided as part of the course:
Class notes
Refreshments at the morning break
Certificate of completion
Fee: $175 for IEEE Members
Return to top of page
Task Based Assessment of Image Quality
Tuesday,
November 12, 2002
Time: 1:00 p.m.-5:30 p.m.
Organizer: |
Michael King, University
of Massachusetts Medical School |
Instructors: |
Charles Metz, University
of Chicago
Harrison Barrett, University of Arizona |
ABSTRACT
Medical images are acquired for the purpose of diagnosis,
delineation of disease state, and monitoring therapy. Thus the
relative merits of different imaging, acquisition, reconstruction,
and processing strategies would be best determined from objective
comparisons of the imaging systems, protocols, and images at
performing tasks closely related to the clinical ones for which
imaging is to be performed. The purpose of this short course
is to introduce the participant to the objective assessment of
image quality and give them the needed information to start conducting
studies of task performance using human and numerical observers.
This course will be divided into two sessions with a combined
discussion, and question and answer session following the second.
The first session will cover the conduction of lesion detection
studies using human observers. It will start with the underlying
model of ROC studies. This will be followed by discussion of
the design, conduction, and analysis of ROC studies. Specific
topics will include the collection of data, definition of "truth",
avoidance of bias in study design, curve fitting, comparison
criteria, and statistical testing. The session will conclude
with discussion of alternative observer testing methodologies
such as LROC, FROC and alternative forced choice.
The second session will deal with classification and estimation
tasks as assessed by numerical observers. It will start with
a review of statistical decision theory and the statistical properties
of medical images. With this background, optimum strategies for
performing the tasks (ideal observers) will be formulated, and
computational difficulties in actually implementing the optimal
strategy will be identified. Various suboptimal strategies will
be presented, including models that incorporate limitations of
the human visual system. Some examples of applications will be
presented.
The following will be provided as part of the course:
Class notes
Refreshments at the afternoon break
Certificate of completion
Fee: $175 for IEEE Members
Return to top of page
Statistical Methods for Image Reconstruction
Tuesday,
November 12, 2002
Time: 1:00 p.m.-5:30 p.m.
Organizer: |
Jeffrey A. Fessler,
Associate Professor
Department of Electrical Engineering and Computer Science
Department of Biomedical Imaging
Nuclear Medicine Division of Department of Radiology
University of Michigan |
ABSTRACT
The
recent commercial introduction of iterative algorithms for tomographic
image reconstruction, and the increasing interest in scanners
with nonstandard imaging geometries, has brought new relevance
and timeliness to the topic of statistical methods for image
reconstruction. This course will provide an orderly overview
of the potpourri of statistical reconstruction methods that have
been proposed recently. Rather than advocating any particular
method, this course will emphasize the fundamental issues that
one must consider when choosing between different reconstruction
approaches. The intended audience is anyone who would like to
reconstruct "better" images from photon-limited measurements,
and who wants to make informed choices between the various methods.
Recent advances in convergent forms of "ordered subsets"
algorithms will be given particular attention, since these algorithms
can be both practical for routine use, while also having desirable
theoretical properties. Both emission tomography and transmission
tomography algorithms will be discussed.
Background
of Participants:
Attendees should be familiar with photon-counting imaging systems
at the level presented in the Medical Imaging short course offered
in previous years. Some past attendees have commented that at
least a little experience with some type of iterative reconstruction
(e.g. ART or OS-EM) would be helpful for getting the most value
from this course.
IMPORTANT NOTICE:
All attendees who advance register for this
course (and who supply a valid
email address with their registration information) will be sent
by email a link and a password for downloading the annotated
lecture notes for this short course about 2 weeks before the
meeting. These advance registrants can then choose whether to
print those notes in advance (about 75 4-up pages) so as to have
hard copy during the course for taking notes, or to download
the PDF file of the notes into a laptop, etc. On-site course
registrants will be given the access information during the course
and may download and print the course notes after the meeting.
Hard copies of the notes will NOT be available at the course,
so advance registration is recommended.
OUTLINE
- A.
Introduction
- Overview
- The
Poisson statistical model
- Mathematical
statement of the reconstruction problem
-
- B.
The Statistical Framework
- Image
parameterization
- Bases
- System
physical modeling
- general
- line
/ strip integrals
- detector
response etc.
- projector/backprojector
cautions
- Statistical
modeling of measurements
- Poisson
- Gaussian
(data-weighted least squares)
- Reweighted
least squares
- Deviations,
e.g. deadtime
- Shifted
Poisson (precorrected random coincidences)
- Emission
vs Transmission scans
- Objective
functions
- Contrast
with "algebraic" methods
- Bayesian
estimation: Maximum a posteriori (MAP) methods
- Data-fit
terms
- likelihood
- quadratic
- robust
- Regularization
- none
- separable
- quadratic
- convex
- nonconvex,
entropy, ...
- Object
constraints
-
- BREAK
-
- C.
Iterative algorithms for statistical image reconstruction
- EM
based
- (EM,
GEM, SAGE, OSEM)
- Direct
optimization
- (Coordinate
Descent, Conjugate Gradient, Surrogate Functions)
- Considerations
- nonnegativity
- parallelizability
- simultaneous
vs sequential
- convergence
rate
- monotonicity
- global
convergence
- Optimization
transfer / surrogate functions
-
- BREAK
-
- D.
Additional topics
- Ordered
subsets / block iterative algorithms
- acceleration
properties interpreted geometrically
- convergence
issues
- Properties
- Spatial
resolution properties / modified penalty functions
- Noise
properties
- Performance
in detection tasks relative to FBP
- Applications
to real PET and SPECT data
- (and
associated practical issues)
- Model
mismatch
- Precorrected
data
- Comparisons
to FBP
- Pseudo-3D
PET reconstruction from Fourier rebinned data
Biographical
Sketch:
Jeff Fessler earned a Ph.D. in electrical engineering in 1990
from Stanford University. He has since worked at the University
of Michigan, first as a DoE Alexander Hollaender post-doctoral
fellow and then as an Assistant Professor in the Division of
Nuclear Medicine. Since 1995 he has been with the EECS Department,
where he is an Associate Professor.
The
following will be provided as part of the course:
Class notes & bibliography
Refreshments at the afternoon break
Certificate of completion
Fee:
$175 for IEEE Members
Return to top of page