{what we do}


Enabling Ultra Low Dose PET/CT Imaging

Imaging of Myocardial Blood Flow

PET and SPECT Scanner Simulation Packages

Quantitative PET/CT and SPECT/CT Imaging

Detector and Timing Electronics and Data Acquisition Systems

Imaging Integration with Radiation Treatment Planning

Novel PET system designs for preclinical and application-specific imaging

Dedicated Breast PET coupled to X-RAY Mammography (Breast PET/X)

Cancer Imaging Research Protocols and Clinical Trials Using PET/CT

Past and Ongoing Research Topics

Enabling Ultra-Low Dose PET/CT Imaging

Respiratory motion correction methods, such as respiartory-gated and breath-hold PET/CT, can cause tumor tracer activity measures to increase and tumor volume measures to decrease.


Image left:
Standard PET scan

Image right:
Respiratory gated PET scan showing small tumor


To enable respiratory motion compensation for PET imaging, it is necessary to extend the duration of the CT scan to better match the PET scan data for both motion and range motion data

Extending the duration of CT scanning, however, increases patient radiation dose, which is a critical barrier to progress. Radiation dose from CT scanning is a growing concern in both the medical community and the public.
The goal of this project is to develop methods for ultra-low-dose CT imaging to be used for CT-based attenuation correction in PET. Our motivation is two-fold:

  1. To lower the CT radiation dose in PET/CT scanning. This is possible due to the reduced noise and resolution requirements for attenuation correction CT imaging as compared to diagnostic CT imaging, and also from the introduction of new acquisition techniques combined with innovative noise and artifact suppression methods.
  2. To improve the quantitative accuracy of PET imaging by enabling methods that compensate for degradation from respiratory motion.

We have developed three approaches for ultra low dose CT-based attenuation correction:
1. Dose minimization by extending standard techniques
2. Dose minimization by tube pulsing and compressed sensing
3. Dose minimization by advanced physics modeling in image reconstruction (figure below)

3torso phantom

The figures above show coronal sections of helical x-ray CT scan (120kVp, 12.5mAs, pitch 1.375) of the RSD anthropomorphic phantom with iodine in liver chamber. Images are reconstructed with conventional FBP and MBIR.

To date we shown that CT tube current can be reduced much lower for CT-based attenuation correction than that used in low dose diagnostic CT protocols. This allows for a further reduction of CT dose with no penalty for PET image quantitation.



imaging of myocardial Blood flow

One of our goals is to develop clinically viable strategies for the quantitative estimation of myocardial blood flow (MBF) from dynamic CT and PET studies. Despite the proven clinical value of quantifying MBF (in ml/g/min), there are no widespread clinical methods to easily measure MBF in absolute units.  We have developed software applications, models, and protocols to measure MBF with PET studies

Fig 1. (right) Standard error of flow estimates derived from simulated noise analysis of dynamic ammonia PET studies versus relative count rates for both the axially distributed model (solid lines) and 3-compartment model (dashed lines). Temporal sampling schemes based on initial 1 sec, 2 sec, 5 sec and 10 sec sampling are presented demonstrating that overall count rate has a larger impact on error than sampling choices (when sampling is 5 sec or shorter) (link).

error of flow

A critical barrier limiting the use of computed tomography (CT) for dynamic perfusion studies is that these studies impart a large radiation dose to the patient. We hypothesize that data- and image- restoration techniques developed for dynamic PET imaging can be adapted to low-dose CT studies to produce quantitatively accurate and precise perfusion images (in absolute units of ml/g/min), with radiation doses to patients less than those in other common cardiac imaging procedures that use ionizing radiation (<6 mSv per exam).   We have developed an accurate simulator for contrast kinetics and CT acquisitions in order to evaluate new dose reduction strategies.



Fig 2. (left) Visualization of qualitative (left) and absolute (right) regional myocardial blood flow estimates derived from dynamic cardiac PET imaging at rest and stress states. These maps are generated with custom software (UW-QPP) developed at the University of Washington for processing and visualization of dynamic PET data. This software is used clinically for all cardiac PET/CT exams performed at UW.

Fig 3. Representative simulated CT images for a dynamic acquisition of contrast enhancement in the myocardium.  These ground truth guide the development of new blood flow estimation algorithms and acquisition protocols.



PET and SPECT Scanner Simulation Packages

The IRL has developed many tools to assist in the improvement of imaging technologies. Notably these include our simulation software packages ASIM and SimSET. Developed by members of the lab and regularly improved upon, these tools have become primary resources for many nuclear medicine imaging research groups around the world.

ASIM is an analytic PET simulator.  It helps primarily in investigations of statisitcal noise, biologic variability and image resolution. An analytic simulation directly calculates values resulting from user defined inputs. These inputs describe both scanner and phantom geometry, the expected number of random and scatter events, the half-life of the radioactivy being imaged, and several other parameters. ASIM can be thought of as a simulation package divided into three applications. The core application is an analytic simulator for positron emission scanners that estimates what the average sinogram (output from the tomograph) would be for the given inputs. The second application applies statistical noise to the sinogram, generating an estimated sinogram for a single noisy scan. Finally a normalize application can be utilized to apply corrections to sinogram data. Because these simulations can be generated quickly, hundreds or thousands of realizations can be produced. The tradeoffs to analytic simulations include a lack of accuracy in the modeling of contanimants  (scatter and random events especially) and simplified tomograph models when compared to most photon tracking simulations.

Image left: Output from a model for the uptake of fluoromisonidazole (FMISO) in gliomas (left) which defined the emission activity map for an ASIM PET simulation. An image was reconstructed from the simulation output (middle); it shows a striking similarity to a clinical PET image of a patient with such a tumor (right).

The Simulation System for Emission Tomography (SimSET) is a photon-tracking simulation for investigations requiring more detailed tomograph models or more accurate estimates of contaminants. It is and easy-to-use, efficient, portable simulation for both PET and SPECT. The SimSET package uses Monte Carlo techniques to model the physical processes and instrumentation used in emission imaging. Using a series of modules, the system follows the photon from annihilation to detection, including all possible interactions along the way. A binning module allows the user to histogram detected events by detected position, energy, and type (scattered or unscattered event, or, for PET, random event) for further analysis. The main tradeoff of photon-tracking simulations is long computational times, though SimSET uses several acceleration techniques to reduce this problem.

Image right: Photon tracking simulations are often used to investigate the impact of scattered photons on images. A SPECT simulation using SimSET created projection data of a torso phantom separately for unscattered (upper left) and scattered (upper right) photons. Image reconstructions of these projections show the impact of unscattered (bottom left) and scattered (bottom right) photons on our estimates of radiotracer distribution.




Quantitative PET/CT and SPECT/CT Imaging

The tremendous expansion of quantitative imaging possibilities provided by modern SPECT/CT scanners and the wide array of SPECT pharmaceuticals makes the possibilities of SPECT/CT quantitation in clinical procedures feasible.   This is an area that may well have a significant impact on improving existing Nuclear Medicine examinations.  While many groups have had great success in developing image reconstruction tools to advance quantitative SPECT (and some manufactures are starting to do the same), practical implementation of quantitative SPECT in the clinic has remained elusive.   Under NIH funding, our group is working to understand the issues to allow implementation of quantitative SPECT in routine clinical operations. 
fig1 Figure 1:  Plot of calibration (conversion) factors for 99mTc cylindrical phantom.  Blue: Precedence data with 20 cm  diameter (6 liter) cylindrical fillable phantom;  Green:  XCT data with 20 cm diameter (6 liter) fillable phantom;  Black: Precedence data using the ACR SPECT phantom;  Red: other data including data from the NEMA image quality phantom and a 2 litter cylindrical phantom.  Note the expanded vertical scale.
fig2 Figure 2:  Plot of calibration (conversion) factors for 57Co cylindrical phantom.  Red: Precedence data;  Blue:  XCT data.   Note the expanded vertical scale.  The large bias in the XCT data in early 2013 was fixed by re-aligning the CT and table systems.   The apparent drift in the XCT values over time is yet to be explained.
A first step is understanding the stability of existing SPECT/CT systems and the potential of using a simple phantom (such as a 57Co cylinder) to provide a basic system calibration that can them be recalled for use with a wide range of collimators and isotopes.   The stability issue is the first step in this process.   To that end we have been collecting data on two SPECT/CT systems (one with a diagnostic 16 slice spiral CT and one with a cone beam flat panel detector CT).  While the data show good stability in general, it also points out apparent trends that need to be fully understood.  In particular, we have found instances where the CT system was responsible for shifts in the system calibration factors even though the daily QC tests were passed.  We are now developing new QA tools and software for the clinical workstations to both track stability and to generate calibration factors which can be used to convert images from counts/voxel to activity/cc.   The work will be extended to include other vendor SPECT/CT systems becoming available in our clinic.
The next steps are quantifying and understanding the sources of regional biases in images of different sized objects with different collimators and isotopes.  That work is expected to lead to an investigation of further improvements in the models used in various reconstruction codes for scatter and collimator response.
circle histogram
Image of cylinder phantom with regions of interest (regions 1 – 6) and the resulting plot of the measured concentration values in the entire phantom and each ROI.  The regional bias effects have proven to be reproducible and an current area of study.



Detector and Timing Electronics and Data Acquisition Systems

The development of new detectors and the supporting electronics is a major activity in our laboratory.   The focus is detector designs that can locate an event in x,y, and z in scintillators using photosensors on a single surface (single ended readout).  This focus reflects our basic goal of designing detectors and electronics that can be easily scaled for use in a wide range of imaging system designs (form pre-clinical to human whole body).

The electronics design effort is based around a single board design that can be easily adapted to many different data acquisition topologies and tasks.  This board (which we term phase II since it is a second generation system in our laboratory) is a complex board as is indicated in the block diagram.  


The board has 12 layers and supports up to 64 channels of 70 mHz ADCs as well as a single high speed ADC channel (currently running at 300 mHz).  The board is centered around a large Altera FPGA to handle all of the pulse processing and reflects the trend in our laboratory to put as much processing as possible into FPGAs rather than large analog boards.  The board also supports expansion connectors to allow easy adaptation to other tasks that we have not provided direct support for in our initial designs.   The current board is based on USB 3.0 for communications to the host computer (earlier versions used FireWire).  There are also provisions for direct board to board connections using high speed serial commiunications (several different protocols).  The end result is a very versitile basic building block that can support a wide variety of imaging system designs.

Image left: Phase II card top view and and the X-ray image (generated by the Eagle layout software) showing the 12 layers of traces

Our detector designs also cover a wide range of applications.   We currently have four major designs on-going in the laboratory.  The original cMiCE module is based on a multi-anode photomultiplier tube (PMT) coupled to a monolithic crystal.  It requires 64 channels of readout along with a timing pickoff channel.  In addition, the module is designed to be used with statistical based position  algorithms to estimate the  x, y, and z coordinates of the event.  Our designs call for those algorithms to be implemented at the detector module level to further reduce the amount of data required to be sent to the host computer.  Those algorithms have been developed and implemented in FPGAs.

The dMiCE approach (using controlled light sharing between crystal pairs) requires individual readouts for each crystal pair.   The current module designs are based on arrays of 10x10 to 20x20 and will use silicon photomultiplier (SiPM) devices for the light collection.  To reduce the number of readout channels, a row/column/diagonal summing ASIC is under development that would reduce the number of channels to be digitized for a 20x20 array to 60 plus a timing pick off channel that is a sum of all the array elements.



Image left: cMiCE detector module, requires 64 readout channels plus a timing pickoff channel.


Figure left: dMiCE detector concept. (a) DOI detector unit.  (b) PMT ratio plots [A/(A+B)].  A significant amount of light is shared when an event is detected near the entrance face of the detector unit.  Less sharing occurs for interactions near the PMT interface

SES cMiCE detector module (sensors on front surface of the monolithic cyrstal) with flex circuit for routing signals to the backside of the module.

Image above: Two Phase II cards (A) supporting two cMiCE detectors (B) for testing.  Also shown is an Altera De0 Nano FPGA card used as the coincidence controller (C) and a clock fan out board (D) developed in our laboratory.


A variant on the cMiCE approach (termed SES) is to put the sensor array is put on the entrance surface of the monolithic crystal, closer to where the majority of first interactions occur.  The result is in an improvement in the estimation of the position of the event.  The initial versions of this design use an 8x8 array of SiPMs mounted on a flex circuit to allow routing of the signals to the backside of the detector module where the connectors and impedance matching networks are located.

Our newest detector design (TSC) uses segmented crystals (tapered) in the transaxial direction and continuous decoding in the axial direction are utilized.   The current module design requires 16 channels of readout, again with a timing pickoff.



Image right: TSC detector module prototype (2x8) requiring 16 channels of data per detector.

Image above: Comparison of the summed slow ADC channels and the high-speed timing ADC for a single event from the cMiCE detector as recorded by the Phase II digital card.

We are also developing techniques using Sub Surface Laser Etching to control the light distribution inside crystals.  One application is to use the SSLE approach to produce the light interface between “crystals” in the dMiCE design.   We have done preliminary work with a surplus SSLE system designed and used for over a decade for the making of souvenirs in a shopping mall.  We are now awaiting delivery of a new system from Crystalix that is specifically designed for our application.



Photo left: two dMiCE crystals with SSLE interfaces. A pair of photodetectors reads out light from the left end.



Imaging Integration with Radiation Treatment Planning

Radiation therapy has long been utilized in the treatment of malignant and benign neoplasms, both in curative and palliative settings.  Recent technological advances, including the modulation of radiation beam shape and intensity during computerized planning and delivery, have increased the achievable spatial and temporal resolution of patient treatments. Our goal is to achieve synergy between radiation therapy planning and advances in quantitative multimodality imaging to precisely target tumor regions at highest risk of treatment resistance or avoid normal tissue at highest risk of radiation damage. In lung cancer patients, we are integrating respiratory motion compensation into PET/CT imaging, treatment planning and delivery.

The figure above shows an example patients PET/CT images of a lung mediastinal tumor, where the upper row is a static image with respiratory motion blurring and the lower row is a respiratory-gated (4D) image at end-exhale phase that has higher contrast recovery

The figure to the right shows a radiation treatment planning workflow, beginning with a PET/CT image, transforming it to a radiation dose prescription and deliverable dose distribution by a radiation therapy linear accelerator.


The figure to the left shows a patient with an integrated PET/CT-based treatment plan with non-uniform radiation dose distribution (dose painting). The upper row represents a plan based on a static PET/CT image, and the bottom row a plan based on a respiratory-gated (4D) PET/CT at end-exhale phase.

Patient-specific, motion-managed, and PET/CT-guided radiation therapy will potentially allow dose escalation to resistant tumor regions while avoiding irradiation of functional normal tissues, which could improve clinical outcomes in future prospective trials.



Novel PET system designs for preclinical and
application-specific imaging

The University of Washington is developing a number of novel PET detector designs as well as systems for preclinical and application-specific imaging.  On the PET detector front there are three main designs that we are pursing.  The first is the sensor on the entrance surface (SES) continuous miniature crystal element (cMiCE) detector.  In this design, the photo-sensor array is placed on the entrance surface of a monolithic crystal PET detector.  Using this detector geometry, the three-dimensional intrinsic positioning performance of the detector is improved versus conventional placement of the photosensors on the back surface of the detector crystal. 

Thse figures illustrate the improved intrinsic positioning for the SES versus conventional cMiCE PET detector design.  The red dots represent testing locations and the black-ring contours represent the FWHM of the event positioning results.  The intrinsic spatial resolution and bias associated with positioning is also visually displayed in the figure.

A second design under development is the trapezoidal, slat crystal (TSC) detector.  In this design, trapezoidal, slat crystals are fabricated to create an ultra compact PET detector that supports high detector packing fraction (for a ring geometry detector system) and depth of interaction decoding capability. Goals for this design are to provide between than 1 mm intrinsic positioning performance in X and Y and <3 mm intrinsic positioning performance in Z (depth of interaction).  The third detector technology we are exploring is the use of sub-surface laser etching to shape the light response function in scintillation detectors to improve their three-dimensional intrinsic positioning performance.  The goal of this work is to not discretize the crystal assembly but to adjust the light response function while maintaining pseudo-continuous event positioning within the detector.
The average intrinsic spatial resolution along the length of the slat for the TSC PET detector is 1.0 mm FHWM after correcting for the width of the testing beam.  The depth of interaction decoding is <2.5 mm FWHM after correcting for intrinisic spatial resolution of the detection and beam width of the testing beam. The average peak to valley ratio for the slat decoding is 5.7.

In addition, to detector development we are currently working on two PET systems. The first is an MRI compatible, small animal PET system using our SES cMiCE detector design. To support this work, we are also developing high performance front end detector electronics and high bandwidth data acquisition boards. The goals of this design are to provide better than 1 mm FWHM image resolution using high sensitivity (i.e., 15 mm thick ) monolithic crystal detectors. 

The second system under development is called PETX (prototype pictured right).  It is a PET system that will be mounted onto a commercial mammography unit to provide co-registered PET and mammography imaging.  A key feature of PETX is an emphasis on the quantiative imaging capabilities of the PET imaging system.  Unlike other currently available positron emission mammography systems, PETX is fully tomographic and will provide quantiatively accurate images.  The initial targeted clinical application of PETX will be to provide accurate, quantiative PET imaging to assess response to therapy for breast cancer patients.

See research topic below for more detail on this novel PETX system




Breast cancer continues to be one of the most prevalent cancers in the U.S. and other developed countries. Advancements in the understanding of basic molecular mechanisms of the disease has led to the development of several new therapies targeted specifically at the implicated molecular pathways. However, there are still large percentages of breast cancer patients that do not respond to targeted therapies despite expressing the target pathologically. Positron emission tomography offers the potential to study the dynamics of molecular pathways in vivo, non-invasively, and in situ. In situ imaging allows the study of bio-physiology of a tumor in its native setting with all associated molecular pathways active. This can reveal interdependencies that may not be captured with an in vitro tissue sample.


Clinical researchers at the Seattle Cancer Care Alliance (SCCA) (and elsewhere) are investigating the use of PET to examine efficacy of therapies in specific patients. The figure above shows schematically the concept of a ‘window of opportunity’ study in which a patient, after diagnosis, undergoes the following regimen: prior to treating the primary tumor a baseline PET image is taken, followed by a trial therapy believed to be appropriate based on earlier patient work-up.A short time (~1-4 weeks) after administration of the trial therapy a follow-up PET scan is taken to observe changes in the tumor. Based on response/non-response determined by the change in PET tracer uptake, the therapy would be selected/rejected for adjuvant treatment. To determine this change using PET it is important that the PET scan measure of tracer uptake be quantitatively accurate. This has posed a challenge for whole-body PET scanners in early-stage breast cancer due to the small size of the tumors. The breast PET/X system is being developed to facilitate these studies in patients with early-stage disease.

To integrate these studies into the breast-imaging clinic we are developing the dedicated PET device to mount onto standard mammography equipment, and designing a gantry to allow acquisition of both the PET image and an x-ray mammogram without moving the breast. We will also pursue integration of the PET detectors with breast tomosynthesis.

The PET detector design is a rectangular box (see figure). This design differs from whole-body PET scanners that are assembled as a circular ring surrounding the patient. The different detector geometry creates different challenges and optimization methods that we are currently studying via Monte Carlo computer simulations. The figure below shows estimated spatial resolution and photon detection sensitivity vs. scintillation crystal thickness on ‘main’ (tM) and ‘side’ (tS) detectors. The study modeled a system with/without depth-of-interaction (3D-CoM/2D-CoM)

Using the rectangular detector also changes the data sampling schemes with respect to circular ring detector PET scanners. The traditional sinogram shown in the figure demonstrates the irregular sampling pattern of the PET/X geometry; the red “+” signs indicate coincident line-of-response data samples, and extreme differences in sampling density are evident. In a ring-geometry system this sampling pattern is essentially uniform. We are in the process of developing image reconstruction programs that account for the distinct data acquisition patterns of PET/X.



Integrating the PET detector hardware onto a mammography gantry requires a custom mounting device. We will use the attachment rails built into mammography systems to mount PET/X, and design the mounting stage such that the PET detectors are removable without moving the patient between PET and mammogram scans. A prototype of the mounting system is shown in a figure below. This is only a mechanical prototype, without operating detector components. The investigation of necessary ergonomic features for the mounting stage is underway.



Cancer Imaging Research Protocols and Clinical Trials Using PET/CT

Cancer remains the second leading cause of death at 23%, closely following heart disease (26%). Progress has been made in treating breast cancer and a few other forms of cancer. However, many other forms of cancer, including lung cancer, have seen no improvement in survival rates over the last 30+ years. This is in spite of US government expenditures of over $200 billion, in inflation-adjusted dollars, since the 1971 National Cancer Act, i.e. the start of the 'War on Cancer'. One reason for this lack of progress is that many cancer therapies are ineffective. This is compounded by the increasing difficulty in conducting clinical trials to evaluate new therapies.


image right: Hallmarks of Cancer. Hanahan & Weinberg Cell 2011


Positron emission tomography (PET) combined with x-ray computed tomography (CT) has become a standard component of cancer diagnosis and staging over the last decade. Many of the techniques used in PET/CT scanners were developed by members of the Imaging Research Laboratory (IRL). We are now developing quantitative PET/CT methods for assessment of an individual’s response to therapy. These methods can also be used for clinical trials of new cancer therapies, since PET/CT can measure changes in tumor metabolism, receptor signaling proliferation rates, and other functional changes.

There are several projects, including new calibration sources (called phantoms), modeling patient biology, working with multiple manufactures, application to ongoing patient studies, and developing national standards.




UW Home > UW Medical Center > Dept of Radiology > Imaging Research Laboratory