About

Georges El Fakhri, PhD, DABR, is the Elizabeth Mears and House Jameson Professor of Radiology and Biomedical Imaging, Therapeutic Radiology, and of Biomedical Informatics & Data Science at Yale School of Medicine. He holds additional titles as Director of the PET Center, Vice Chair of Scientific Research in Radiology & Biomedical Imaging, and Director of the Yale Biomedical Imaging Institute. His research focuses on biomedical imaging, with a particular emphasis on PET imaging, cardiac extracellular volume mapping, and the development of advanced imaging techniques for health monitoring and disease diagnosis. Dr. El Fakhri has contributed extensively to the field through his work on signal modeling, contrast agent injection, and three-dimensional cardiac imaging, among other areas. His research output includes numerous peer-reviewed publications, and he collaborates frequently with other leading scientists in the field.

Research topics

• Computer Science
• Artificial Intelligence
• Biochemistry
• Psychology
• Medicinal chemistry
• Neuroscience
• Speech recognition
• Pathology
• Stereochemistry
• Biology
• Computer vision
• Chemistry
• Medicine
• Nuclear medicine
• Database
• Mathematics

Selected publications

• The ADAPT learning cancer treatment system: ARPA-H’s initiative to revolutionize cancer therapy

  Cancer Cell · 2026-01-08

  articleOpen access
  PublisherOA PDFDOI

• Fast 3D cardiac extracellular volume mapping and cine imaging via continuous free-breathing acquisition with mid-scan contrast agent injection

  Proceedings on CD-ROM - International Society for Magnetic Resonance in Medicine. Scientific Meeting and Exhibition/Proceedings of the International Society for Magnetic Resonance in Medicine, Scientific Meeting and Exhibition · 2025-09-16

  article

  Motivation: Three-dimensional cardiac extracellular volume (ECV) mapping and cine imaging are challenging due to respiratory and cardiac motions. Existing ECV mapping methods wait at least 10-minutes after contrast agent administration which increases the total scan time. Goal(s): To develop a time-efficient method for free-breathing 3D cardiac ECV mapping and cine imaging. Approach: Data was acquired continuously using inversion recovery spoiled gradient-echo (IR-SPGR) with mid-scan contrast agent administration. The Linear Tangent Space Alignment (LTSA) model was used for image reconstruction. Results: Three-dimensional cardiac ECV maps and cine images were obtained in a 20-minute free-breathing scan with contrast agent injection in between. Impact: We propose a highly time-efficient, free-breathing 3D cardiac imaging method that integrates ECV mapping and cine imaging in a single scan with mid-scan contrast agent injection.

  PublisherDOI

• Impact of Depth-of-Interaction and Spatially-Variant Point-Spread-Function Model on the Performance of Ultra-High Resolution Brain PET

  2025-11-01

  article

  Positron Emission Tomography (PET) has been a pivotal tool in brain research, enabling detailed analysis of cerebral flow, metabolism, and receptor occupancy. Spatial resolution is a critical parameter for brain-dedicated PET systems. To improve image quality in ultra-high-performance brain-dedicated PET scanners such as the NeuroEXPLORER (NX), depth of interaction (DOI) techniques and accurate point spread function (PSF) modeling are essential. In this study, we present a spatially-variant PSF model implemented in the open-source Yale Reconstruction Toolbox for PET (YRT-PET). Using Monte Carlo simulations, we estimated the spatially-variant PSF parameters from point source with background, and assessed how DOI and different PSF models affect spatial resolution and image quality. For comparison, reconstructions using the United Imaging Healthcare PET Reconstruction Toolbox (URT) platform were also performed. Results from simulated Mini-Derenzo and brain phantom show that the proposed spatially-variant PSF consistently outperforms both spatially-invariant single center-PSF and no-PSF approaches when DOI data is used. In addition, the YRT-PET reconstruction with spatially-variant PSF modeling demonstrated an enhanced contrast recovery in cortical structures, closely matching the vendor's reconstructions. These findings highlight the importance of DOI for achieving uniform resolution across the field of view and underscore the role of spatially-variant PSF in enhancing contrast in high-activity regions. Future investigations involve validating the proposed spatially-variant PSF model with real phantom and human scan, and investigating the convergence behavior of reconstructions employing different PSF models.

  PublisherDOI

• UniMRSeg: Unified Modality-Relax Segmentation via Hierarchical Self-Supervised Compensation

  ArXiv.org · 2025-09-19

  preprintOpen access

  Multi-modal image segmentation faces real-world deployment challenges from incomplete/corrupted modalities degrading performance. While existing methods address training-inference modality gaps via specialized per-combination models, they introduce high deployment costs by requiring exhaustive model subsets and model-modality matching. In this work, we propose a unified modality-relax segmentation network (UniMRSeg) through hierarchical self-supervised compensation (HSSC). Our approach hierarchically bridges representation gaps between complete and incomplete modalities across input, feature and output levels. % First, we adopt modality reconstruction with the hybrid shuffled-masking augmentation, encouraging the model to learn the intrinsic modality characteristics and generate meaningful representations for missing modalities through cross-modal fusion. % Next, modality-invariant contrastive learning implicitly compensates the feature space distance among incomplete-complete modality pairs. Furthermore, the proposed lightweight reverse attention adapter explicitly compensates for the weak perceptual semantics in the frozen encoder. Last, UniMRSeg is fine-tuned under the hybrid consistency constraint to ensure stable prediction under all modality combinations without large performance fluctuations. Without bells and whistles, UniMRSeg significantly outperforms the state-of-the-art methods under diverse missing modality scenarios on MRI-based brain tumor segmentation, RGB-D semantic segmentation, RGB-D/T salient object segmentation. The code will be released at https://github.com/Xiaoqi-Zhao-DLUT/UniMRSeg.

  PublisherOA PDFDOI

• Predicting tumor recurrence site after reirradiation in head and neck cancer: a retrospective external validation of a published [18F]-FDG PET radiomic signature

  La radiologia medica · 2025-08-20

  articleSenior author
  PublisherDOI

• Variance Extrapolated Class-Imbalance-Aware Domain Adaptive Myocardial Segmentation in Multi-Sequence Cardiac MRI

  IEEE Journal of Biomedical and Health Informatics · 2025-12-31

  articleOpen access

  Fully automated myocardial segmentation from cardiac magnetic resonance imaging (MRI) is vital for efficient diagnosis and treatment planning. Although numerous automated methods have been proposed, they typically focus on single MRI sequences and therefore have difficulties in generalizing across vendors and across cardiac MRI protocols. Simultaneous analysis of complementary cardiac MRI sequences, such as cine, T1 mapping, and late gadolinium enhancement (LGE) MRI, remains challenging due to their distinct image characteristics and scanner-specific variations. To address these issues, we propose an unsupervised domain adaptation approach that allows robust myocardial segmentation across multi-vendor cine, T1, and LGE MRI data. In particular, we introduce a class- imbalance self-training framework to transfer information learned from a source domain with labels to any unlabeled target domain, while maintaining consistent performance across different MRI sequences. Our framework iteratively refines segmentation accuracy by generating pseudo-labels for target data using a hardness-aware strategy, thus effectively addressing the problem of class imbalance in cardiac MRI segmentation. To mitigate data scarcity following pseudo-label selection, we employ a variance-guided vicinal feature extrapolation, which expands data points in the feature space into a probabilistic distribution. This, in turn, facilitates joint source-target training by generating a larger intersection in the feature space. Experimental results demonstrate that our framework outperforms existing methods when assessed using the Dice coefficient and Hausdorff distance. Our framework enables cardiac evaluation across MRI protocols without sequence-specific manual annotations.

  PublisherOA PDFDOI

• LR-PET: A Subspace-Based Dynamic PET Imaging via Explicit Non-Negative Low-Rank Factorization

  2025-11-01

  article

  Dynamic PET provides valuable insights by tracking tracer kinetics over time. Detected events are divided into T shorter frames, each with only a fraction of the total counts, and each frame is typically reconstructed independently. This approach, however, ignores the inherent spatiotemporal correlation of the data --- especially in brain studies, where motion is minimal and kinetic changes are dominant. To exploit this redundancy, we propose to model the dynamic images as residing in a rank-K lower-dimensional subspace with explicit low-rank factorization. The proposed method was validated on simulated data and on an [18F]-MK6240 dynamic PET study, demonstrating higher-quality images, lower variability at equal bias, faster reconstruction times by a ratio <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\sim \mathrm{K} / \mathrm{T}$</tex>, and reduced memory requirements; all within a compact and simple EM-based framework.

  PublisherDOI

• Ultrafast J-resolved magnetic resonance spectroscopic imaging for high-resolution metabolic brain imaging

  Nature Biomedical Engineering · 2025-06-20 · 2 citations

  article
  PublisherDOI

• Multicenter PET Image Harmonization Using Style-Guided CycleGAN in Primary Central Nervous System Lymphoma : InStyleGAN

  Lecture notes in computer science · 2025-11-13

  book-chapter
  PublisherDOI

• Dopaminergic frontostriatal pathways in major depressive disorder and childhood sexual abuse: a multimodal neuroimaging investigation

  Molecular Psychiatry · 2025-09-09 · 1 citations

  articleOpen access
  PublisherOA PDFDOI

Recent grants

• NIH Grant R01HL110241

  NIH · $1.9M · 2017

• Postgraduate Training Program in Medical Imaging (PTPMI)

  NIH · $237k · 2011–2021

• NIH Grant T32EB002102

  NIH · $864k · 2011

• Ultra High Resolution Brain PET Scanner for in-vivo Autoradiography Imaging

  NIH · $4.6M · 2018–2026

• NIH Grant R21EB012326

  NIH · $470k · 2014

Frequent coauthors

• Marc D. Normandin

  Gordon Center for Medical Imaging

  768 shared

• Jinsong Ouyang

  Yale University

  564 shared

• Jonghye Woo

  486 shared

• Nicolas J. Guehl

  Harvard University

  456 shared

• Fangxu Xing

  Harvard University

  430 shared

• Quanzheng Li

  Harvard University

  428 shared

• Yoann Petibon

  Takeda (United States)

  331 shared

• Xiaofeng Liu

  Harvard University

  320 shared

Labs

• Yale Biomedical Imaging InstitutePI