Training and Dissemination

View code on Github

This code and shared simulation datasets can be used to reproduce the simulation results published in our work, and also help the community further develop approaches combining deep learning and iterative reconstruction. Below is the abstract of this work.

PET image reconstruction is challenging due to the ill-poseness of the inverse problem and limited number of detected photons. Recently, the deep neural networks have been widely and successfully used in computer vision tasks and attracted growing interests in medical imaging. In this work, we trained a deep residual convolutional neural network to improve PET image quality by using the existing inter-patient information. An innovative feature of the proposed method is that we embed the neural network in the iterative reconstruction framework for image representation, rather than using it as a post-processing tool. We formulated the objective function as a constrained optimization problem and solved it using the alternating direction method of multipliers algorithm. Quantification results show that our proposed iterative neural network method can outperform the neural network denoising and conventional penalized maximum likelihood methods.

Questions? Contact Kuang Gong

We have developed an analysis pipeline in Matlab for processing pharmacokinetic PET studies performed in non-human primates and in humans. The workflow includes registration of the images to a standardized template space, automated extraction of regional brain time activity curves (TACs) using a set of brain atlases, processing of blood and plasma data, if available, application of kinetic models for accurate quantification of the physiological processes of interest and computation of parametric maps.

The list of implemented mathematical models includes blood-based techniques such as full compartment models in different configurations (one-tissue or two-tissue in reversible and irreversible modes with and without inclusion of the vascular contribution to the PET signal as a model parameter), graphical methods (such as Logan, MA1 and Patlak) and reference-tissue based techniques (such as SRTM, Logan DVR, MRTM). The inputs that are required for the pipeline consists of the individual structural MR images and the dynamic PET images either in dicom or in nifti format. If blood-based models are selected, the user also needs to provide the whole-blood and plasma concentrations measured over the course of the study as well as corresponding parent fraction measurements for generation of metabolite-corrected arterial input function.

Questions? Contact Nicolas Guehl

A manifold learning-based method that leverages the intrinsic low-dimensional structure of MR Spectroscopic Imaging (MRSI) signals for joint spectral quantification.

The available code consists of MATLAB®-based scripts and functions along with corresponding example simulated datasets.

Download software

Please email developer Chao Ma for help and comments.

C. Ma, et. al. “Joint spectral quantification of MR spectroscopic imaging using linear tangent space alignment (LTSA)-based manifold learning” Magn Reson Med. 2022 Nov 20;. doi: 10.1002/mrm.29526.

PLEASE NOTE: The software available on this page is provided free of charge and comes without any warranty. CMITT and Yale do not take any liability for problems or damage of any kind resulting from the use of the files provided. Operation of the software is solely at the user’s own risk. The software developments provided are not medical products and must not be used for making diagnostic decisions.

The software is provided for non-commercial, non-clinical academic use only. Usage or distribution of the software for commercial purpose is prohibited. If you use the software for academic work, please give credit to the author in publications and cite the related publications.

Some of the software developments provided are still work-in-progress such that the software is not completely free of errors.

YRT-PET is a tomographic image reconstruction toolkit developed to overcome limitations in existing PET reconstruction software while fostering innovation, reproducibility, and scalability. Designed to be modular and flexible, YRT-PET is compatible with a wide range of commercial and research PET scanners, including modern total-body systems. Its GPU-accelerated performance enables efficient execution of advanced iterative reconstruction methods. A Python interface through pybind11 facilitates rapid prototyping and seamless integration with modern scientific workflows. The toolkit also supports event-by-event motion correction, enhancing the accuracy of dynamic PET reconstructions. YRT-PET’s flexibility is further demonstrated through its interoperability with open data formats, such as ETSI list-mode files (in progress), and its native support for deep learning integration via PyTorch. This allows researchers to combine classical PET models with methods such as the deep image prior (DIP) or unrolled networks, leveraging transparent GPU memory management for advanced image processing. Committed to open science, YRT-PET provides thorough documentation and unit tests to ensure reproducibility and scalability in PET imaging research. By empowering researchers and clinicians with a versatile, high-performance platform, YRT-PET is pushing the boundaries of quantitative PET imaging and molecular medicine.

https://github.com/YaleBioImaging/yrt-pet

For further inquiries, please contact Thibault Marin (thibault.marin@yale.edu).

CMITT provides research training in medical imaging, as currently applied to disciplines such as nuclear medicine (PET and SPECT), magnetic resonance imaging (MRI) and computed tomography (CT). The goal of the training is to provide an avenue for doctoral scientists in physics, engineering, mathematics, statistics, as well as related disciplines to enter and be successful in medical imaging research. This is achieved by providing training in radiological sciences to trainees with a strong quantitative background via structured didactic courses and seminars, enabling graduates to critically evaluate the field and formulate their own research ideas. Trainees participate in leading-edge research, with the opportunity to interact with a world-class faculty in a setting that includes the resources of Yale School of Medicine and Yale University as a whole. CMITT stresses the methodology needed to advance medical imaging research specifically (in contrast to research training in imaging probes, animal models, or the study of disease mechanisms). The major foci of the didactic program are: (1) the physics of image formation with radiation, magnetic resonance and computed tomography, (2) the use of image processing to enhance the quantitative diagnostic and therapeutic capabilities of medical imaging, (3) the kinetic modeling of physiological processes needed to test hypotheses with cross-sectional PET, MRI and fMRI, as well as proton therapy and to advance quantitative functional imaging and molecular medicine.

Trainees can directly participate in unique training where trainees learn PET/MR theory, techniques, the underlying technical methods, and a wide range of clinical applications.

We will also provide educational and hands-on training to medical students, graduate students, and postdoctoral fellows and leverage the extended resources of our PET/MR program.

A small number of participants will be provided in-depth training in specific PET/MR technology. If you are interested in participating in the visiting scientist program, please contact our director and administrative coordinator Georges El Fakhri and Chris Agro with your CV and research interests.

P-41 Symposium at the World Molecular Imaging Congress (WMIC)

• September 9-13, 2024
• Montréal, Canada
• Palais des congrès de Montréal
• Organizer: Georges El Fakhri, Martin Pomper, Robert Gropler
• Details to come

IEEE-MIC

• October 30-Nov 2, Tampa, Florida
• “PET Kinetic Modeling And Parametric Imaging”
• Organizers: Marc Normandin and Guobao Wang

ISMRM Workshop on MR Spectroscopy: Frontiers in Molecular & Metabolic Imaging

• Organizer: Chao Ma
• October 15-18, 2024, Boston, Massachusetts
• Details to come

P-41 Symposium at the World Molecular Imaging Congress (WMIC)

Prague, Czechia

September 5th – September 9th 2023

The National Institute of Biomedical Imaging and Bioengineering (NIBIB) supports a large network of National Centers for Biomedical Imaging and Bioengineering (NCBIB) through the P41 grant mechanism. Three of these Centers comprise the faction of Molecular Imaging Technology Centers:

The Center for Molecular Imaging Technology & Translation (CMITT), Massachusetts General Hospital, led by Dr. Georges El Fakhri, aims to develop and apply new imaging technologies that will revolutionize the way scientists and physicians view and use PET and MRI.

The Resource for Molecular Imaging Agents in Precision Medicine, Johns Hopkins University, led by Dr. Martin Pomper, encompasses projects that extend from the development of reagents to detect and promote an immune reactive tumor microenvironment to the synthesis of nanodrones to treat cancer and combined small-molecule diagnostic and theranostic agents.

The PET Radiotracer Translation and Resource Center (PET-RTRC), Washington University School of Medicine, led by Dr. Robert Gropler, seeks to develop new PET radiotracers that will image biologic targets modulating the ubiquitous disease processes of inflammation and oxidative stress.

The synergistic model of the P41 mechanism interaction of service and expertise within the Centers and other collaborating laboratories ensures other researchers may gain access to the newest Molecular Imaging Technology. During this session, participants will learn about cutting-edge molecular imaging innovations, in addition to the collaborating, service, and training opportunities disseminating from the three complementary, but non-overlapping programs that may benefit your own research and projects.

IEEE Nuclear Science and Medical Imaging Symposium in November 2023 in Vancouver, Canada

Short Course: “PET Kinetic Modeling And Parametric Imaging”

• Monday, November 6 , 2023 from 8:00 am – 6:00 pm

“Open Kinetic Modeling Workshop”

• Tuesday, November 7th, 2023 from 1:00-4:15 pm
• Vancouver Convention Centre, Room MR 118/9/0
• Organizer: Guobao Wang

IEEE Nuclear Science and Medical Imaging Symposium in November 2022 in Milan, Italy

Short Course: "PET Kinetic Modeling: Theory and Applications"

• The course topics comprised "Reference Tissue Modeling Methods", " Graphical and Linearized Models", and " Direct Estimation of Kinetic Parameters".
• Tuesday, 8 November 2022 – 8:30 – 17:30

(PTO Serial No. 62/146,004)

The present invention overcomes the aforementioned drawbacks by providing a method for reconstructing an image from limited angle positron emission tomography (“PET”) data. The method includes providing limited angle PET data to a computer system. This limited angle PET data indicates gamma ray activity in a subject. A reduced angle system matrix that includes rows associated with only view angles represented in the limited angle PET data is selected. The image is then iteratively reconstructed from the limited angle PET data by iteratively solving an optimization problem that includes the selected reduced angle system matrix.

(PCT/ US20150246143 A1, CA2920334A1, CN104718299A, EP2882863A1, EP2882863A4, WO2014028392A1)

Inventors: El Fakhri and Alpert

PCT / US20160296766 A1)

Inventors: El Fakhri, Alpert, Zhu and Grogg

(PTO Serial No. 61/817/750, S201600773993A1, WO2014179443A1)

The present disclosure relates generally to medical imaging and, more particularly, to systems and methods for tracking motion during medical imaging procedures using motion tracking coils.

(PCT/ 20150262389A1, WO2014074148A1)

Inventors: Li, and El Fakhri

(PCT/US2017/039896, WO2018005737A1)

Disclosed are nanocarriers that include one or more cyclodextrin moieties conjugated to a polymer. The cyclodextrin moieties can complex therapeutic (e.g., anticancer) agents, and can be used to treat diseases such as cancer.

Inventors: Hak Soo Choi, Homan Kang, Georges El Fakhri

A Method for Quantitative Biological Measurements Using Labeled Responsive Contrast Materials

Methods to Automatically Tuning Spatially Variant Hyperparameter for Penalized Maximum Likelihood Estimation.