| Time | Speaker | Affiliation | Talk Title |
|---|---|---|---|
| 12:40 | Thibault Marin | YBII, Yale Radiology | Physics-Informed Machine Learning for Quantitative PET Imaging |
| Dr. Marin is an Assistant Professor in the Department of Radiology and Biomedical Imaging at Yale University. His research focuses on the development of advanced algorithms for quantitative imaging in PET and MR. His work in PET includes the development of image reconstruction methods for next-generation PET scanners and the development of direct estimation methods combining image reconstruction and kinetic modeling. His work in MR imaging focuses on the development of subspace and manifold learning based methods for cardiac MRI. Additionally, he is interested in deep learning applications for model-based image denoising, parametric mapping and image segmentation. | |||
| 13:00 | Junhao Wen | Columbia | Multi-organ imaging offers new opportunities for whole-body AI modeling |
| Dr. Wen, PhD, is a computational neuroscientist with expertise in medical image computing, artificial intelligence/machine learning, multi-omics, and multi-organ bioinformatics. He is the director of imaging genetics research at Columbia's Center for Innovation in Imaging Biomarkers and Integrated Diagnostics (CIMBID). Dr. Wen’s research focuses on developing and applying artificial intelligence/machine learning techniques to analyze multi-organ and multi-omics biomedical data in the context of human aging and disease, with a particular emphasis on clinical and computational neuroscience to advance precision medicine. | |||
| 13:20 | Christof Karmonik | Houston Methodist Research Institute | Morphological Traits Classification of Arterial lesions in CLTI using Ultra-Short TE aided by a Convolutional Variational Autoencoder |
| A convolutional variational autoencoder (convVAE) was developed to classify below-the-knee arterial lesions in chronic limb-threatening ischemia using high-resolution 3D ultra-short echo time (UTE) MRI. The latent space enabled identification of six distinct lesion morphologies and generation of pseudo-composition maps for lesion characterization. Majority-vote aggregation across lesion cross-sections yielded highly robust lesion-level classification, supporting MRI-based treatment planning. | |||
| 13:27 | Shivi Kumar | University of Pennsylvania CHOP | MRI-Defined Glioblastoma Habitats Linked to Molecular Resistance Programs via Radiogenomic Integration of TCIA and TCGA Data |
| Glioblastoma (GBM) exhibits extreme intratumoral heterogeneity that limits therapeutic efficacy and contributes to recurrence. We present a radiogenomic imaging-to-therapy pipeline that links MRI-defined GBM habitats with molecular resistance programs and enables in silico therapeutic research from noninvasive imaging. This study develops a computational radiogenomic framework integrating multiparametric magnetic resonance imaging (MRI) from The Cancer Imaging Archive. | |||
| 13:34 | Armina Fani | Georgia State University, Center for Translational Research in Neuroimaging and Data Science | Domain Randomization for Robust Rodent Brain MRI Skull Stripping |
| Skull stripping is a critical preprocessing step in rodent brain MRI, yet robust automated tools remain scarce due to limited training data and high variability across acquisition protocols. We present a domain randomization approach (to our knowledge the first applied to rodents) that trains exclusively on synthetic images generated from label maps, enabling strong generalization across datasets without retraining or fine-tuning, in a model lightweight enough to run directly in the browser. | |||
| 13:41 | Sina Moayed Baharlou | Department of Electrical and Computer Engineering, Boston University | An end-to-end hybrid deep-learning approach for single-shot wavefront sensing and correction |
| This work presents a hybrid optical-computational imaging framework that combines a learned phase mask with deep learning for single-shot wavefront sensing and correction. The approach eliminates iterative phase retrieval, operates across broadband conditions, and is validated experimentally using SLM and metasurface implementations. | |||
| Time | Speaker | Affiliation | Talk Title |
|---|---|---|---|
| 14:05 | Qingyu Zhao | Cornell | Predictive Modeling in Developmental Neuroimaging |
| Dr. Zhao is an Assistant Professor in the Department of Radiology at Weill Cornell Medicine. His research focuses on machine-learning-based computational analysis of neuroimaging and neuropsychological data to explain brain-behavior relationships and determine biomedical phenotypes of neurological diseases. | |||
| 14:25 | Christopher Whitlow | Yale School of Medicine | AI in Diagnostic Radiology |
| Christopher Whitlow, MD, PhD, MHA, chair of radiology and biomedical imaging and assistant dean for translational research at Yale School of Medicine, focuses on advancing neuroimaging, computational data science, and translational innovation in biomedical imaging. Whitlow’s research integrates computational approaches such as artificial intelligence, deep learning, and graph theoretical network methods to analyze complex clinical and research data. His work aims to synthesize imaging, genomic, and clinical information to enhance precision diagnostics and improve patient care, particularly in oncology and neurosciences. | |||
| 14:45 | Javid Dadashkarimi | University of Pennsylvania | PIGMENT: A deep learning framework for Porcine Immunohistochemistry seGMENTation |
| PIGMENT is an annotation-efficient deep-learning framework for detecting and mapping APP-positive axonal pathology in porcine brain histology. Using a compact SegFormer model and pathology-specific augmentation, it transforms tile-level predictions into spatially organized tissue burden maps, supporting scalable analysis of traumatic brain injury and future micro–macro associations studies. | |||
| 14:52 | Qing Lyu | YBII, Yale Radiology | Self-Auditing Residual Drifting for Pathology-Preserving Accelerated Knee MRI |
| This abstract presents SA-RDM-DC, a self-auditing residual drifting model for accelerated knee MRI. The model learns residual corrections from zero-filled reconstructions, enforces measured k-space data consistency, and predicts dense error maps plus slice-level risk scores. It is evaluated on fastMRI knee data with pathology-aware analysis, showing strong structural fidelity, faster inference than iterative diffusion, and a practical signal for flagging unreliable reconstructions. | |||
| 14:59 | Yafei Dong | YBII, Yale Radiology | Clinical Information–Assisted 3D Latent Diffusion Transformer for Head and Neck Tumor Segmentation |
| Most existing segmentation approaches rely solely on radiological images and largely overlook the potential value of clinical information. We propose a clinical information-assisted 3D latent diffusion transformer model that integrates both volumetric imaging data and structured clinical information to achieve more accurate H&N tumor segmentation. Experimental results demonstrate a significant performance improvement after incorporating clinical information compared with image-only baselines. | |||
- Daniel Sodickson (Function Health)
- Cynthia McCollough (Mayo Clinic)
- Nicole Wake (GE HealthCare)
- Tony Fossile (Philips)
- Michael Bush (Siemens)
- Hongdi Li (United Imaging)
| Time | Speaker | Affiliation | Talk Title |
|---|---|---|---|
| 10:30 | Marc Normandin | YBII, Yale Radiology | Ultra-high-performance brain PET |
| Dr. Normandin, PhD, is Associate Professor of Radiology and Biomedical Imaging at Yale Medical School and the PET Core director. He is recognized as a leader in molecular imaging, especially quantitative methodology for PET and MRI. His research interests include pharmacokinetic modeling for quantification of physiological function, development and application of novel radiotracers for imaging and therapy, and multimodal integration of PET and MRI data in brain, cardiac, and cancer applications. | |||
| 10:50 | Yanis Djebra | YBII, Yale Radiology | List-Mode Subspace-based Dynamic PET: 1-Second Frames on the NeuroEXPLORER |
| We present a list-mode subspace (low-rank) reconstruction for dynamic PET that jointly recovers all frames from a high-resolution NeuroEXPLORER scan. On an in-vivo non-human-primate [11C]UCB-J study, it reconstructs 1-second frames across the full 30-minute acquisition—resolving early perfusion and ungated respiratory motion—while matching 120-second OSEM image quality at low bias and cutting storage ~180× (1.8 TB to 10 GB). | |||
| 10:57 | Yue Zhuo | YBII, Yale Radiology | Comparison BaF2 and LYSO with Photonic Crystals for TOF-PET |
| We evaluated PhC designs with different geometries and surrounding media for BaF₂ and LYSO using MEEP FDTD and GATE. The optimal design depends on crystal refractive index: LYSO benefits from air-interface cylinders, while BaF₂ performs best with grease-embedded cylinders. Far-field k-space maps provide experimental validation. BaF₂ cross-luminescence offers a larger fast-photon budget enabling improved light coupling efficiency with PhCs, while LYSO achieves comparable performence through Cherenkov first-photon detection. | |||
| 11:04 | Zizhuo Xie | University of Science and Technology of China | BDM + PNI: A Modular Platform for Application-Specific PET |
| The BDM + PNI platform lets clinicians and researchers build application-specific PET systems without deep instrumentation expertise. Basic Detector Modules (BDMs) are modular particle-detector blocks that arrange into arbitrary geometries and output raw multi-voltage threshold (MVT) data, deferring all processing decisions to software. Plug-n-Image (PNI) provides swappable software modules that compose into a full pipeline — MVT curve fitting → singles → coincidence/listmode → image — where every stage can be replaced independently. The authors built a clinical in-beam PET system for proton therapy range monitoring entirely on BDM + PNI. | |||
| 11:11 | Qinyuan Yang | Yale, Applied Physics | Parallel acquisition of multi-contrast MRI and Deuterium Metabolic Imaging on a preclinical platform |
| This abstract aims on introducing new Magnetic Resonance Imaging (MRI) - Deuterium Metabolic Imaging (DMI) preclinical methods which implement parallel MRI-DMI acquisition to obtain anatomical and metabolic information. We present various MRI methods including T1-weighted (T1W) and T2-weighted (T2W) MRI, Diffusion Weighted Imaging (DWI) and Magnetization Transfer Contrast (MTC) MRI acquired in conjunction with steady state DMI. A practical example is demonstrated on a RG2 brain tumor in vivo. | |||
| Time | Speaker | Affiliation | Talk Title |
|---|---|---|---|
| 11:25 | Henk De Feyter | YBII, Yale Radiology | "Deuterium Metabolic Imaging (DMI) to characterize brain tumors" |
| Dr. De Feyter is an Associate Professor in the Department of Radiology and Biomedical Imaging. Originally from Belgium, where he got a degree as physical therapist, he moved to the Netherlands and got his PhD in biomedical NMR at the Eindhoven University of Technology, before joining the MRRC as a postdoc with Prof. Douglas Rothman. His research is focused on applying advanced magnetic resonance spectroscopy techniques combined with the use of stable isotopes to study metabolism in vivo, with a particular interest in brain tumors. | |||
| 11:45 | Christina Sun | Yale University, Biomedical Engineering Department | Deuterium Metabolic Imaging to Detect Early Treatment Response in Glioma |
| DMI offers a radiation-free alternative to FDG-PET for mapping brain tumor metabolism. By tracking deuterium-labeled glucose and its downstream products, lactate and Glx, DMI can capture the Warburg effect in gliomas. In a rat model undergoing chemoradiotherapy, treated tumors showed a trend toward reduced glycolytic activity before volumetric differences were detectable on structural MRI, suggesting DMI may enable earlier assessment of treatment response than conventional imaging alone. | |||
| 11:52 | Menghua Xia | Yale School of Medicine | Task-Based Evaluation of AI PET Denoising for Low-Count Lesion Localization Using CHO JAFROC Analysis |
| AI denoising, particularly diffusion or U-Net based denoising, is a valuable tool for improving image quality and potentially lesion detection in low-dose PET imaging. However, using CHO within an FROC study, we show that it is not a universal fix; its benefits are highly conditional on the image noise, lesion contrast, and anatomical background, demonstrating that visually "smoother" images do not automatically guarantee better clinical results in all conditions. | |||
| 11:59 | Rui Wang | Yale University, Biomedical Engineering Department | Flow Matching for Tumor Contrast Enhancement Modeling in Breast MRI Images |
| We investigate flow matching for synthesizing contrast-enhanced breast MRI from pre-contrast images. | |||
| 12:06 | John Onofrey | Yale, Biomedical Engineering Department | Deep Learning-based Segmentation of Prostate Cancer Lesions on PSMA PET/CT with Anatomical Context |
| Accurate lesion segmentation on PSMA PET/CT is essential for quantifying disease burden in advanced prostate cancer, yet manual delineation of the typically numerous lesions is time-consuming and subject to inter-observer variability. We introduce an anatomy-guided deep learning approach that incorporates organ masks and organ-specific data augmentation to provide anatomical context. The proposed method is validated on two datasets. | |||
| 12:13 | Moses Wilks | YBII, Yale Radiology | Peritumoral Injection of 177Lu-Nanoparticles Inhibits Tumor Growth With Low Organ Normal Radiation |
| Peritumoral injection of the 177Lu labeled nanoparticle ferumoxytol (FMX) was used to inhibit tumor growth and limit normal organ radiation doses. Mice were imaged by SPECT and post-mortem tissues were analyzed to calculate dosimetry. [177Lu]Lu-FMX greatly inhibited tumor growth and increased overall survival. SPECT showed 177Lu retention in or around the tumor, with little off-target dose. Absorbed doses to tumor (111.5 Gy) exceeded dose to healthy tissues (e.g. liver, marrow) by 10-100 fold. | |||
| Time | Speaker | Affiliation | Talk Title |
|---|---|---|---|
| 13:30 | Dustin Scheinhost | Yale Medical School | TBD |
| Dustin Scheinost, Ph.D., is an Associate Professor of Radiology & Biomedical Imaging, Biomedical Engineering, Statistics & Data Science, and at the Yale Child Study Center. His research focuses on neuroimaging method development and their application to development and mental health. | |||
| 13:50 | Leonardo Cardoso Saraiva | YSM, Department of Psychiatry | Whole-brain dynamical modeling reveals E/I imbalance and implicates neurotransmitter systems in OCD |
| Using a biophysical whole-brain model with PET-derived receptor densities, we mapped excitation/inhibition (E/I) balance across OCD, schizophrenia, and ADHD. We found transdiagnostic E/I alterations (right transverse temporal, left cuneus) linked to NMDA, VAChT, and DAT, plus OCD-specific changes driven by 5-HT1B and CB1. Results point to E/I imbalance as an upstream mechanism and highlight the endocannabinoid system as a novel OCD treatment target. | |||
| 13:57 | Maggie Davis | YBII, Department of Psychiatry | Central Sensitization, Kappa Opioid Receptor Availability, and Quality of Life in Trauma Survivors: Preliminary Results from Autonomic Testing and in vivo PET Imaging Studies |
| We present of preliminary analyses and conclusions from a set of complimentary studies examining the relationship between between central sensitization, autonomic dysfunction (Study 1), kappa opioid receptor availability (Study 2), PTSD severity, pain, and quality of life for the first time. | |||
| 14:04 | Richard Goyette | Department of Brain and Cognitive Sciences, MIT | Reward Processing in Childhood Depression: A Preliminary Look |
| There is a clear association between depression and abnormalities of reward processing. It is much less clear if the two share any causal link. Childhood depression provides insights into potential mechanistic relationships between depressive symptoms and reward processing abnormalities. | |||
| 14:11 | Elisenda Bueicheku | YBII, Yale Radiology | Connectomic Motifs Reveal Molecular Vulnerability Underlying Sequential Tau Spreading in the Human Brain |
| This research investigates systems-level tau spreading in late-onset Alzheimer’s disease. Using longitudinal tau-PET imaging data, we apply graph-based connectome models (network motifs) while accounting for ApoE e4 and amyloid-beta status. Finally, we integrate the Allen Human Brain Atlas information to explore the intersection between neuroimaging phenotypes and gene transcriptional profiles. | |||
| 14:18 | Taekyung Kang | Yale, Biomedical Engineering Department | fMRI-Compatible Simultaneous Cortical Wide-Field Fluorescent Ca2+ and Intrinsic Hemodynamic Imaging |
| Resting-state BOLD fMRI maps brain networks but indirectly reflects neuronal, vascular, and oxygenation dynamics. To better separate these components, we previously developed an fMRI-compatible wide-field optical platform that simultaneously measures cortical WF-Ca²⁺ fluorescence and now our system also measures intrinsic hemoglobin reflectance, providing preliminary neural and vascular data for future integration with whole-brain BOLD fMRI. | |||
| 14:25 | Basav Sanganahalli | YBII, Yale Radiology | Translational MRI of Structural and Functional Network Dysfunction Following Traumatic Brain Injury |
| Traumatic brain injury (TBI) often causes persistent neurological deficits that are not detected by conventional imaging. Using a translational multimodal MRI approach combining structural MRI, diffusion tensor imaging (DTI), and resting-state fMRI, we identify subtle alterations in brain microstructure and network connectivity following TBI. These imaging biomarkers provide insight into injury progression, recovery, and therapeutic response, advancing precision diagnostics and treatment strategy. | |||
| Time | Speaker | Affiliation | Talk Title |
|---|---|---|---|
| 15:55 | Gerald Shulman | Yale School of Medicine | Imaging Human Metabolism: From Molecular Mechanism to Precision Therapeutics |
| Dr. Shulman is the George R. Cowgill Professor of Medicine and Cellular & Molecular Physiology at Yale. He is also an Investigator Emeritus of the Howard Hughes Medical Institute and Co-Director of the Yale Diabetes Research Center. Dr. Shulman has pioneered the use of magnetic resonance spectroscopy combined with mass spectrometry to non-invasively examine intracellular glucose and fat metabolism in humans and transgenic rodent models | |||
| 16:15 | Xuanang Xu | YBII, Yale Radiology | Estimating Cardiac Material Parameters through Physics-Informed Gradient Descent Optimization |
| This research presents a physics-informed optimization framework to non-invasively estimate intrinsic cardiac material properties (e.g., stiffness and contractility) directly from 4D ultrasound and MRI data. By embedding continuum mechanics into a gradient descent pipeline, the algorithm can extract fundamental material parameters from observable heart motion. This approach paves the way for subject-specific biomechanical profiling to better monitor cardiac remodeling and disease progression. | |||
| 16:23 | Paul Han | YBII, Yale Radiology | Look-Locker with Variable Flip Angle Imaging for Free-Running 3D Cardiac T1 Mapping |
| This work presents a novel free-running method for cardiac T1 mapping. The method is characterized by a continuous data acquisition scheme of Look-Locker with variable flip angle imaging to maximize the data acquisition efficiency and improve the through-plane spatial coverage, resolution, and patient comfort, while enhancing the robustness of T1 estimation against B0 and B1+ inhomogeneities. The feasibility of the method is demonstrated via in vivo studies with five healthy volunteers at 3T. | |||
| 16:30 | Jie Xiang | YBII, Yale Radiology | CAIPI3 accelerated 4D flow bSSFP at 3T for improved diastolic function evaluation |
| We proposed a 4D flow bSSFP at 3T accelerated by CAIPI3, for rapid diastolic function evaluation in a short 5min free-breathing scan. | |||
Abstract Summaries
Abstracts are listed in alphabetical order by the presenting author's last name
White Matter Vulnerability Throughout Menopause: Memory Pathways at Risk in Women
Oumayma Agdali
Yale University
Women represent two-thirds of Alzheimer's cases and experience faster cognitive decline, potentially linked to the menopause transition. Using DTI connectomes from HCP-A, we assessed white matter microstructure across menopausal stages and age-matched men. Perimenopausal women show the greatest disruption in memory-critical limbic circuitry. Namely, the fornix, cingulum, and uncinate fasciculus, suggesting menopause as a key inflection point in female-specific dementia vulnerability.
Poster #01
Neuronal activity and amyloid-beta promote tau seeding in the entorhinal cortex in Alzheimer's disease
Christoffer Gretarsson Alexandersen
Yale University
Why does tau pathology begin in the entorhinal cortex in Alzheimer’s disease? We combined brain connectivity with FDG- and amyloid-PET data in a computational model of tau transport. Across ADNI and HABS, neuronal activity biased tau seeding toward the medial temporal lobe, while amyloid-β selectively amplified entorhinal vulnerability. Subject-level predictions also tracked measured entorhinal tau.
Poster #02
Mindfulness-based Neurofeedback to augment DBT for Borderline Personality Disorder
Jitendra Awasthi
Department of Radiology and Biomedical Imaging, Yale University School of Medicine, New Haven, United States of America
Borderline personality disorder (BPD) is a mental health condition characterized by affective instability, impulsivity, and difficulties in interpersonal relationships. The most widely disseminated evidence-based treatment is Dialectical Behavior Therapy (DBT). A core component of DBT is mindfulness training. This study tests whether the mindfulness-based neurofeedback could help BPD patients get better outcomes with DBT by improving their mindfulness.
Poster #03
Reactive Oxygen Species (ROS)-Sensitive Imaging: From Fundamental Insights to Practical Guidelines
Shubham Bansal
Department of Chemistry and Center for Diagnostics and Therapeutics, Georgia State University, Atlanta, Georgia 30301 USA
Oxidative stress resulting in elevated levels of reactive oxygen species (ROS) have been associated with a large number of pathological conditions. Therefore, there is a widespread interest in utilizing these elevated levels of ROS for diagnostic and therapeutic applications. However, in such studies, there are major concerns of inappropriate use of various methods that lead to erroneous results. Along this line, we have described several factors that reshape ROS research. For example, commonly
Poster #04
Artificial Intelligence for the Federated Screening of ATTR Cardiomyopathy in the US: A Multicenter Analysis from the TRACE-AI Network
Bruno Batinica
Yale University
Across the TRACE-AI network of US health systems, AI-Echo (AUROC 0.88) and AI-ECG (0.78) screened cardiac imaging for transthyretin amyloid cardiomyopathy (ATTR-CM). Combining modalities sharpened detection: PYP confirmation rose from 6.6% (both negative) to 61% (Echo+) and 77% (both positive). Dual positivity also predicted higher mortality. Federated, multimodal imaging-based AI offers a tiered pathway to prioritize patients for confirmatory testing.
Poster #05
Analysis of behavioral and neural differences in learning and memory in borderline personality disorder
Cassie Boulis
Yale University, Fineberg Lab
In this project, we use the RAVLT to analyze the behavioral and neural differences in learning and memory in people with BPD compared to healthy controls (HC) while recording EEG. We find that people with BPD perform significantly worse on the RAVLT than HC, consistent with prior studies, and anticipate finding a decreased P2 ERP amplification. We seek to explore whether performance on the RAVLT manifests as a neural signature that presents differently in people with BPD vs. HC.
Poster #06
Connectomic Motifs Reveal Molecular Vulnerability Underlying Sequential Tau Spreading in the Human Brain
Elisenda Bueichekú
Yale Biomedical Imaging Institute; Department of Radiology and Biomedical Imaging
This research investigates systems-level tau spreading in late-onset Alzheimer’s disease. Using longitudinal tau-PET imaging data, we apply graph-based connectome models (network motifs) while accounting for ApoE e4 and amyloid-beta status. Finally, we integrate the Allen Human Brain Atlas information to explore the intersection between neuroimaging phenotypes and gene transcriptional profiles.
Session 3: Neuroimaging
Poster #07
Whole-brain dynamical modeling reveals E/I imbalance and implicates neurotransmitter systems in OCD
Leonardo Cardoso Saraiva
Department of Psychiatry, Yale University School of Medicine
Using a biophysical whole-brain model with PET-derived receptor densities, we mapped excitation/inhibition (E/I) balance across OCD, schizophrenia, and ADHD. We found transdiagnostic E/I alterations (right transverse temporal, left cuneus) linked to NMDA, VAChT, and DAT, plus OCD-specific changes driven by 5-HT1B and CB1. Results point to E/I imbalance as an upstream mechanism and highlight the endocannabinoid system as a novel OCD treatment target.
Session 3: Neuroimaging
Poster #08
Cura 1T: Specialized Model for Agentic Healthcare
Haolin Chen
actAVA AI
We present Cura 1T, a healthcare-specialized LLM trained through a human-gated self-evolution loop. In each evolution round, a training agent plans a target capability, trains the model, evaluates benchmark trajectories, and refines the data mixture from observed failures. Cura 1T ranks at or near the top among frontier baselines, while remaining competitive on out-of-domain reasoning and agentic benchmarks.
Poster #09
Systematic Differences Between Standardized and Routine Clinical Centiloid Reporting in Real-World Amyloid PET
Katelle Colvin
Yale University
Centiloid (CL) values increasingly inform anti-amyloid therapy eligibility, with decision thresholds concentrated at the low end of the scale. In routine practice these values are generated by heterogeneous vendor software rather than a single standardized implementation. We applied a standardized automated pipeline to a real-world clinical cohort and asked how closely routine reported values track it, and whether the two agree well enough to be interchangeable at the individual-patient level.
Poster #10
PIGMENT: A deep learning framework for Porcine Immunohistochemistry seGMENTation
Javid Dadashkarimi
University of Pennsylvania
PIGMENT is an annotation-efficient deep-learning framework for detecting and mapping APP-positive axonal pathology in porcine brain histology. Using a compact SegFormer model and pathology-specific augmentation, it transforms tile-level predictions into spatially organized tissue burden maps, supporting scalable analysis of traumatic brain injury and future micro–macro associations studies.
Workshop 2: Translational use of AI in Medical Imaging
Central Sensitization, Kappa Opioid Receptor Availability, and Quality of Life in Trauma Survivors: Preliminary Results from Autonomic Testing and in vivo PET Imaging Studies
Margaret Davis
Yale School of Medicine, Department of Psychiatry (secondary appointment in Psychology)
We present of preliminary analyses and conclusions from a set of complimentary studies examining the relationship between between central sensitization, autonomic dysfunction (Study 1), kappa opioid receptor availability (Study 2), PTSD severity, pain, and quality of life for the first time.
Session 3: Neuroimaging
Preliminary assessment of [18F]3F4AP PET for Imaging Demyelination in Alzheimer’s Disease
Maeva Dhaynaut
Yale School of Medicine
This study presents the first evaluation of [18F]3F4AP PET for imaging myelin-related alterations in Alzheimer's disease. Using multimodal PET/MRI, we investigated relationships between demyelination, tau pathology, amyloid burden, and myelin water fraction. Preliminary findings support [18F]3F4AP as a promising biomarker of myelin integrity and highlight the value of multi-tracer imaging in Alzheimer's disease.
Poster #11
Parametric mapping with the tau PET tracer [18F]MK6240: validation against regional quantification
Maeva Dhaynaut
Yale School of Medicine
This study validates parametric imaging methods for dynamic [18F]MK6240 tau PET by comparing voxel-wise estimates of tau burden (DVR) and relative tracer delivery (R1) with standard regional quantification. The findings demonstrate excellent performance of R1 maps and strong agreement of DVR maps with ROI-based analyses, supporting the use of parametric imaging for whole-brain tau PET studies while highlighting important quantitative limitations.
Poster #11
Evaluation of novel ¹⁸F-labeled PET tracers for μ-opioid receptor imaging in non-human primates
Alexandra DiFilippo
Biomedical Imaging Institute
[11C]CFN is the established PET radiotracer for in vivo μ-opioid receptor (MOR) imaging but is limited by short 11C half-life and extreme tracer potency. This study evaluated 3 novel fluorinated MOR radiotracers in nonhuman primates. All [18F]FMOR regional VT values showed strong correlations with [11C]CFN and naloxone produced complete blockade of all receptors, consistent with MOR binding. These findings support further investigation of all [18F]FMOR tracers as 18F alternatives to [11C]CFN.
Poster #12
List-Mode Subspace-based Dynamic PET: 1-Second Frames on the NeuroEXPLORER
Yanis Djebra
Yale Biomedical Imaging Institute, Yale School of Medicine
We present a list-mode subspace (low-rank) reconstruction for dynamic PET that jointly recovers all frames from a high-resolution NeuroEXPLORER scan. On an in-vivo non-human-primate [11C]UCB-J study, it reconstructs 1-second frames across the full 30-minute acquisition—resolving early perfusion and ungated respiratory motion—while matching 120-second OSEM image quality at low bias and cutting storage ~180× (1.8 TB to 10 GB).
Session 1: Image Technology
Clinical Information–Assisted 3D Latent Diffusion Transformer for Head and Neck Tumor Segmentation
Yafei Dong
Yale University
Most existing segmentation approaches rely solely on radiological images and largely overlook the potential value of clinical information. We propose a clinical information-assisted 3D latent diffusion transformer model that integrates both volumetric imaging data and structured clinical information to achieve more accurate H&N tumor segmentation. Experimental results demonstrate a significant performance improvement after incorporating clinical information compared with image-only baselines.
Workshop 2: Translational use of AI in Medical Imaging
Cannabis use in trauma-related pathology: Neurobiological differences and relationships with anxiety and mindfulness
Elizabeth Duraney
Yale School of Medicine
Comorbid cannabis use is common among individuals with trauma-related psychopathology (TRP). Evidence of the neurobiological effects of cannabis in TRP is needed. Kappa opioid receptor availability (KOR) has been implicated in the pathophysiology of TRP. It is unknown how cannabis use might impact KOR availability in this population. The current study examined differences in KOR availability in individuals with TRP based on cannabis use.
Poster #13
First Dual-Tracer PET Study of Mu and Kappa Opioid Receptor Availability in Alcohol Use Disorder
Gaelle M. Emvalomenos
Radiology and Biomedical Imaging
This is the first study to image both MOR and KOR in the same individuals with AUD. Both receptor systems were dysregulated: KOR was reduced in the striatum and amygdala (confirming hypotheses), but MOR was unexpectedly lower across all regions, notably in the insula. This suggests that opioid receptor availability may shift across abstinence stages, with implications for AUD treatment targeting.
Poster #14
Deuterium Imaging: The Hype, the Hope, and the Real-World Impact
Roozbeh Eskandari
Albert Einstein College of Medicine
Deuterium metabolic imaging (DMI) is an emerging, cost-effective technique for visualizing metabolic reprogramming associated with chronic diseases. Because of its favorable MR properties, DMI enables high-resolution detection of metabolic fluxes and downstream metabolites. This abstract shows recent probe development, their biological applications, and key considerations for designing next-generation deuterium probes for metabolic imaging.
Poster #15
Domain Randomization for Robust Rodent Brain MRI Skull Stripping
Armina Fani
Georgia State University, Center for Translational Research in Neuroimaging and Data Science
Skull stripping is a critical preprocessing step in rodent brain MRI, yet robust automated tools remain scarce due to limited training data and high variability across acquisition protocols. We present a domain randomization approach (to our knowledge the first applied to rodents) that trains exclusively on synthetic images generated from label maps, enabling strong generalization across datasets without retraining or fine-tuning, in a model lightweight enough to run directly in the browser.
Workshop 1: Technical Developments in AI in Medical Imaging
Poster #16
Advanced Optical Microscopy for Visualizing Spinal Cord Circuits at Depth
Sulekh Fernando-Peiris
Department of Biomedical Engineering, Yale University
Imaging the spinal cord at cellular resolution has long been limited to its most superficial layers. This work uses three-photon microscopy to image over twice as deep into the dorsal horn, characterizing tissue optical properties, establishing safe laser power parameters via injury markers, and demonstrating reliable calcium transients during hindlimb thermal stimulation beyond 200 µm. These findings provide practical, translatable guidelines for deep imaging of spinal cord circuitry.
Poster #17
Reward Processing in Childhood Depression: A Preliminary Look
Richard Goyette
Department of Brain and Cognitive Sciences, MIT
There is a clear association between depression and abnormalities of reward processing. It is much less clear if the two share any causal link. Childhood depression provides insights into potential mechanistic relationships between depressive symptoms and reward processing abnormalities.
Session 3: Neuroimaging
Look-Locker with Variable Flip Angle Imaging for Free-Running 3D Cardiac T1 Mapping
Paul Han
Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, CT, USA
This work presents a novel free-running method for cardiac T1 mapping. The method is characterized by a continuous data acquisition scheme of Look-Locker with variable flip angle imaging to maximize the data acquisition efficiency and improve the through-plane spatial coverage, resolution, and patient comfort, while enhancing the robustness of T1 estimation against B0 and B1+ inhomogeneities. The feasibility of the method is demonstrated via in vivo studies with five healthy volunteers at 3T.
Session 4: Body Imaging
Regional Amyloid Distribution Stratifies Dementia Risk Within Amyloid-Positive MCI
Jeremy Hudson
Department of Radiology and Biomedical Imaging, Yale School of Medicine
This exploratory ADNI PET/MR study tests whether amyloid’s regional distribution, not just total burden, refines dementia risk within amyloid-positive MCI. A limbic-hub amyloid distribution score stratified 3-year progression risk beyond scalar amyloid measures and aligned with FDG hypometabolism, medial-temporal atrophy, WMH, and ventricular burden.
Poster #18
fMRI-Compatible Simultaneous Cortical Wide-Field Fluorescent Ca²⁺ and Intrinsic Hemodynamic Imaging
Taekyung Kang
Biomedical Engineering
Resting-state BOLD fMRI maps brain networks but indirectly reflects neuronal, vascular, and oxygenation dynamics. To better separate these components, we previously developed an fMRI-compatible wide-field optical platform that simultaneously measures cortical WF-Ca²⁺ fluorescence and now our system also measures intrinsic hemoglobin reflectance, providing preliminary neural and vascular data for future integration with whole-brain BOLD fMRI.
Session 3: Neuroimaging
Morphological Traits Classification of Arterial lesions in CLTI using Ultra-Short TE aided by a Convolutional Variational Autoencoder
Christof Karmonik
Houston Methodist Research Institute
A convolutional variational autoencoder (convVAE) was developed to classify below-the-knee arterial lesions in chronic limb-threatening ischemia using high-resolution 3D ultra-short echo time (UTE) MRI. The latent space enabled identification of six distinct lesion morphologies and generation of pseudo-composition maps for lesion characterization. Majority-vote aggregation across lesion cross-sections yielded highly robust lesion-level classification, supporting MRI-based treatment planning.
Workshop 1: Technical Developments in AI in Medical Imaging
Poster #19
Single-Season Changes in MRI-Derived Brain Age Following Repetitive Head Impact Exposure in High-School Football
Mohammad Iyas Kawas
Yale School of Medicine, Department of Radiology and Biomedical Imaging
High-school footballers experience significant repetitive head impacts during a critical period of brain development. This study examined whether a single season of collision sport alters brain aging trajectories measured by MRI. Using CentileBrain.org models, analysis of 37 football players and 13 athlete controls, found that football players showed a ~1-year relative reduction in MR-predicted brain age over one season; a result that challenges assumptions about head-impact-related brain aging
Poster #20
Cortical Myelin Mapping in Individuals at Clinical High Risk (CHR) for Psychosis: An Accelerated Medicines Partnership Schizophrenia (AMP SCZ) Study
Victoria King
Yale University
The temporal convergence of cortical myelin maturation and the onset of psychosis-spectrum disorders suggests that abnormal myelination may represent a core neurodevelopmental mechanism underlying psychosis vulnerability. We analyzed T1w/T2w ratio myelin maps in 679 individuals at clinical high risk for psychosis (CHR) from the Accelerating Medicines Partnership Schizophrenia (AMP SCZ) dataset to determine whether myelin-related abnormalities are present before the onset of the illness.
Poster #21
MRI-Defined Glioblastoma Habitats Linked to Molecular Resistance Programs via Radiogenomic Integration of TCIA and TCGA Data
Shivi Kumar
University of Pennsylvania CHOP
Glioblastoma (GBM) exhibits extreme intratumoral heterogeneity that limits therapeutic efficacy and contributes to recurrence. We present a radiogenomic imaging-to-therapy pipeline that links MRI-defined GBM habitats with molecular resistance programs and enables in silico therapeutic research from noninvasive imaging. This study develops a computational radiogenomic framework integrating multiparametric magnetic resonance imaging (MRI) from The Cancer Imaging Archive.
Workshop 1: Technical Developments in AI in Medical Imaging
Improved Perfusion Imaging Using Pseudo-Continuous Arterial Spin Labeling with Radial Balanced Steady-State Free Precession
Jaehun Lee
Department of Radiology and Biomedical Imaging, Yale School of Medicine
This work presents a novel arterial spin labeling (ASL) method for non-invasive perfusion imaging. The method is characterized by pseudo-continuous arterial spin labeling (pCASL), balanced steady-state free precession (bSSFP) readout, and a radial sampling scheme to improve the SNR, enhance robustness to motion, and reduce scan time in ASL perfusion imaging. The performance of the proposed method is evaluated via in vivo studies with four healthy volunteers.
Poster #22
Network-Guided Synthetic Image-Based Deformable Registration for Dynamic Myocardial T1 Mapping
Wonil Lee
Yale School of Medicine, Department of Radiology and Biomedical Imaging
We developed a network-guided synthetic image based deformable registration framework for free-breathing myocardial T1 mapping. A signal-to-parameter estimation network generates contrast-matched synthetic images, enabling accurate motion correction despite dynamic contrast changes. The proposed method reduced respiratory motion, improved myocardial wall delineation, and produced myocardial T1 values closer to the breath-hold MOLLI reference, without requiring ground-truth deformation fields.
Poster #23
Self-Auditing Residual Drifting for Pathology-Preserving Accelerated Knee MRI
Qing Lyu
Yale School of Medicine
This abstract presents SA-RDM-DC, a self-auditing residual drifting model for accelerated knee MRI. The model learns residual corrections from zero-filled reconstructions, enforces measured k-space data consistency, and predicts dense error maps plus slice-level risk scores. It is evaluated on fastMRI knee data with pathology-aware analysis, showing strong structural fidelity, faster inference than iterative diffusion, and a practical signal for flagging unreliable reconstructions.
Workshop 2: Translational use of AI in Medical Imaging
Modality-Adaptive Fusion of Segmentation Information for Medical Image Classification Across Heterogeneous Imaging Modalities
Sarthak Mahapatra
Yale School of Medicine
We present a modality-adaptive deep learning framework for segmentation-guided medical image classification across heterogeneous imaging modalities. The model learns how to balance lesion-focused and contextual information using interpretable gating mechanisms, enabling modality-specific reasoning for breast ultrasound, dermoscopy, chest radiography, and retinal fundus imaging.
Poster #24
[18F]F-Lasoxifene: a non-steroidal radiotracer for estrogen receptor (ER) imaging
Riya Mallik
Yale School of Medicine, Postdoctoral Associate
Estrogen receptor (ER)-targeted PET imaging has become clinically relevant as it provides important information in hormone-responsive breast cancer (BCa). Herein, we have developed non-steroidal [18F]F-LASi for ER-PET imaging. We retrosynthesized ER-inhibitor lasofoxifene and introduced a piperazine-linker to generate ERi-NH. Subsequently, we radiolabeled with an aryl [18F]F-fluoride group yielding [18F]F-LASi in >99% radiochemical purity, 7.7% decay-corrected radiochemical yield, and molar acti
Poster #25
AI-Enabled Multimodal Fast Force Microscopy correlated with Brain MRI and Cryo-Electron Microscopy for Quantitative Imaging of Physical Forces that cause Diverse Bacterial Infections and Inflammation
Nikhil Malvankar
YSM and FAS
We have developed multimodal microscopy that can visualize and quantify electrical and mechanical forces between brain cells and pathogens, both in single cells and biofilms, using a nanoelectrode array and fast force microscopy. We found that Neisseria meningitidis, the causative agent of cerebrospinal meningitis, transfers electrons via aromatic residues in pili to overcome the electrostatic repulsion from human brain cells that initiates infection, which can be suppressed by targeting pili.
Poster #26
Improving Medical Image Generative Models with Fréchet Distance Loss
Andrew Marshall
Yale University, Biomedical Engineering
We propose fine-tuning segmentation-guided diffusion models with a Fréchet Distance (FD) loss that aligns synthetic and real image feature distributions. On two liver cancer CT datasets, downstream segmentation models trained on FD-regularized synthetic data outperform those trained on non-FD synthetic data and heavily-augmented real data, improving tumor Dice and boundary accuracy on real patient data.
Poster #27
Lifespan brain age prediction from structural and functional connectomes using connectome-based predictive modeling
Marie McCusker
Yale University
This study uses connectome-based predictive modeling to estimate chronological age from diffusion and resting-state functional MRI brain connectivity across the human lifespan. Using large-scale Human Connectome Project cohorts, it examines how structural and functional brain networks predict age from development through older adulthood and characterizes shared and distinct network-level organization of brain age signals.
Poster #28
An end-to-end hybrid deep-learning approach for single-shot wavefront sensing and correction
Sina Moayed Baharlou
Department of Electrical and Computer Engineering, Boston University, Boston, MA, USA
This work presents a hybrid optical-computational imaging framework that combines a learned phase mask with deep learning for single-shot wavefront sensing and correction. The approach eliminates iterative phase retrieval, operates across broadband conditions, and is validated experimentally using SLM and metasurface implementations.
Workshop 1: Technical Developments in AI in Medical Imaging
The Association between Physical Activity and Higher Corticolimbic Synaptic Density is Independent of Body Mass Index
Karina Moisieienko
Ylae Medical School ( Psychiatry Department)
Using in vivo PET imaging with the synaptic vesicle glycoprotein 2A (SV2A) radiotracer [18F]SynVesT-1, we investigated whether physical activity is associated with synaptic density in the human brain. Physically active individuals showed higher synaptic density in corticolimbic regions, particularly in prefrontal circuits. These associations remained after accounting for body mass index and depression severity, suggesting a potential link between physical activity and synaptic health.
Poster #29
A Unified Framework for PET Motion Correction and Quantitative MR Imaging Using Manifold Learning-Based Real-Time MRI
Ismael Mounime
Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, CT, USA, and LTCI, Télécom Paris, Institut Polytechnique de Paris, Paris, France
Cardiac and respiratory motion degrade PET image quality and quantification. We propose a manifold learning-based real-time MRI framework that simultaneously enables PET motion correction and quantitative MR imaging within a single PET/MR acquisition. In a preliminary 18F-TPP study, MR-derived motion fields improved PET image quality and revealed fine anatomical structures while also providing whole-heart ECV maps, demonstrating the feasibility of a unified PET/MR imaging approach.
Poster #30
Application of non-negative low-rank constraint for dynamic image reconstruction on the high-resolution SAVANT scanner
Yassir Najmaoui
Yale Biomedical Imaging Institute
High-resolution PET scanners tend to reduce their overall sensitivity. This leads to a possible compromise between dynamic frame duration, spatial resolution, and noise. The proposed dynamic image reconstruction method leverages the spatiotemporal redundancy to reconstruct all frames simultaneously. The method, based on MLEM, updates alternatively a matrix representing the spatial distribution for each temporal basis and a matrix representing the translation from temporal bases to dynamic frame.
Poster #31
Deep Learning-based Segmentation of Prostate Cancer Lesions on PSMA PET/CT with Anatomical Context
John Onofrey
Biomedical Engineering
Accurate lesion segmentation on PSMA PET/CT is essential for quantifying disease burden in advanced prostate cancer, yet manual delineation of the typically numerous lesions is time-consuming and subject to inter-observer variability. We introduce an anatomy-guided deep learning approach that incorporates organ masks and organ-specific data augmentation to provide anatomical context. The proposed method is validated on two datasets.
Session 2: Imaging Cancer
SpatialGen: A Bayesian Framework with Localized Error Mapping for Coherent MR-to-CT Skull Segmentation
Lucas Papamitsakis
Yale College
This framework adapts the MetaCOG model to correct U-Net skull segmentations by shifting from temporal permanence to spatial coherence. We register 3D scans to an atlas to isolate the cranial vault and define a normalized slice index, and we train a conditional VAE with soft Dice loss to learn slice-dependent skull geometry as an anatomical prior. Instead of rigid global variables, a patch-based grid of localized Beta priors captures independent error rates to reduce noise post-hoc.
Poster #32
Neurofeedback to reduce anhedonia and improve outcomes in opioid use disorder: a feasibility study.
Sarah Paprotna
Yale University
Anhedonia and poor autobiographical memory (AM) recall are transdiagnostic clinical phenotypes of depression and opioid-use disorder (OUD). Prior work has used fMRI neurofeedback to treat depression by reducing anhedonia and improving AM recall. In this feasibility trial, we will adapt this work to treat OUD and evaluate its impacts on clinical and neuropsychological function. This work represents a novel approach to OUD treatment with potential to improve quality of life for those affected.
Poster #33
Relaxivity-sensitive Inversion Recovery for LGE (RSIR)
Dana Peters
Yale University, Department of Radiology and Biomedical Imaging
We present preliminary study of RSIR, relaxivity-sensitive inversion recovery, for improved late gadolinium enhancement. The theoretical analysis and practical application are demonstrated.
Poster #34
Triadic Connectivity: A Voxelwise Graph-Based Approach for Characterizing Cortico-Striato-Thalamo-Cortical (CSTC) Circuits
Lester Rodriguez-Santos
Yale University
Cortico-striato-thalamo-cortical (CSTC) circuits are implicated in psychiatric disorders, but conventional connectivity methods assess only pairwise relationships. We developed a voxelwise graph-based approach to identify CSTC triadic motifs and applied it to resting-state fMRI data from healthy controls (N=46) and individuals with cocaine use disorder (N=61). We identified motif hubs across groups and found that more years of cocaine use is associated with reduced connectivity in the precuneus.
Poster #35
Translational MRI of Structural and Functional Network Dysfunction Following Traumatic Brain Injury
Basavaraju Sanganahalli
Radiology and Biomedical Imaging
Traumatic brain injury (TBI) often causes persistent neurological deficits that are not detected by conventional imaging. Using a translational multimodal MRI approach combining structural MRI, diffusion tensor imaging (DTI), and resting-state fMRI, we identify subtle alterations in brain microstructure and network connectivity following TBI. These imaging biomarkers provide insight into injury progression, recovery, and therapeutic response, advancing precision diagnostics and treatment strateg
Session 3: Neuroimaging
Comparison of 99% and 90% enriched [6,6’-2H2]-D-Glucose for Deuterium Metabolic Imaging of Glioma in Rats
Christina Sun
Yale University, Department of Biomedical Engineering
DMI was used to compare 99% and 90% enrichment [6,6'-2H2]-D-glucose in a rat glioma model. Glc/water signal ratios were comparable between groups, indicating similar glucose delivery regardless of enrichment level. Both formulations captured the glioma Warburg phenotype, with lactate/(lactate+Glx) ratios consistently higher in tumor than normal-appearing brain. These preliminary findings suggest 90% enrichment glucose may be a cost-effective alternative for DMI studies of brain tumor metabolism.
Poster #37
Deuterium Metabolic Imaging to Detect Early Treatment Response in Glioma
Christina Sun
Department of Biomedical Engineering
DMI offers a radiation-free alternative to FDG-PET for mapping brain tumor metabolism. By tracking deuterium-labeled glucose and its downstream products, lactate and Glx, DMI can capture the Warburg effect in gliomas. In a rat model undergoing chemoradiotherapy, treated tumors showed a trend toward reduced glycolytic activity before volumetric differences were detectable on structural MRI, suggesting DMI may enable earlier assessment of treatment response than conventional imaging alone.
Session 2: Imaging Cancer
Development of the cognitive energy landscape from infancy to adolescence
Huili Sun
Yale University
Using network control theory and 3,712 developmental dMRI scans, we map the structural 'control energy' to activate 100 NeuroSynth cognitive states from infancy through adolescence. Energy decreases for 96/100 states; social/perceptual functions reach peak efficiency earlier (~100 mo) than higher-order cognition (~206 mo). Prenatal neurogenesis shapes infant energy; myelination acts across all ages and the broadest cognitive scope. Transition architecture stays stable but becomes modularized.
Poster #36
Relating Psychological and Clinical Scales to Network Connectivity Change from Mindfulness-Based Neurofeedback in Borderline Personality Disorder
Benjamin Swinchoski
Yale School of Medicine
Aberrant default mode network (DMN) activity is observed in people with borderline personality disorder (BPD). In our mindfulness-based neurofeedback clinical trial, participants with BPD employ mindfulness practice to modulate their DMN and frontoparietal control network (FPCN). Analyses test whether baseline clinical and psychological scales predict change in within-DMN and/or FPCN-DMN connectivity, as well as if state mindfulness change correlates to connectivity change.
Poster #38
Unified Optimization Framework for the Accurate Mapping of Effective Dose (ED50)
Dimitri Szezurek
Yale Biomedical Imaging Institute and Department of Radiology and Biomedical Imaging, Yale University School of Medicine, New Haven, CT, USA; Sorbonne Université, AP-HP, Hôpitaux Universitaires Pitié-Salpêtrière, Service de Médecine Nucléaire and LIB, INSERM, Paris, France
We propose a framework to jointly estimate parametric maps of ED50, RO, and nondisplaceable binding potential simultaneously from a set of pairs of scans: before and after administration of blocking agent levetiracetam. The method is tested on simulated NeuroEXPLORER scans, showing significant improvements in map accuracy and reduced error across brain structures compared to conventional indirect fitting.
Poster #39
Preclinical Evaluation of [¹⁸F]BPARP, a Brain-Penetrant PARP1 PET Tracer, in Glioma Rats and NHP
Peng Wen Tan
Yale Biomedical Imaging Institute
This study presents the preclinical evaluation of 18F-PARP10, a brain-penetrant PET radiotracer developed to image PARP-1 expression in glioma. Using dynamic PET imaging, biodistribution, blocking studies with AZD9574, radiometabolite analysis, and histological correlation in BT142 tumor-bearing and healthy rats, the study characterizes tracer uptake, specificity, pharmacokinetics, and metabolic stability to support its potential for future clinical translation in brain tumor imaging.
Poster #40
Reviving a Dormant Lead: Cardiac Sympathetic Innervation Imaging in Anxiety and Trauma-Related Disorders—A Narrative Review
Viona Tavakkoli
New York University Department of Psychology
Cardiac sympathetic innervation imaging (MIBG SPECT, HED PET) is well validated in heart failure and Lewy body disease, but its only application to primary anxiety disorders is a single unreplicated 2004 panic disorder study. A 2024 Parkinson’s disease study offers indirect, modern support, linking cardiac denervation specifically to anxiety. We argue this gap is a stalled, not exhausted, line of inquiry, and propose reviving it with modern PET methods in panic disorder, GAD, and PTSD.
Poster #41
Spatiotemporal Collagen Remodeling Drives Mechanical Immune Exclusion Niches in Melanoma
Noriko Toyosato Bracken
Yale School of Medicine
Spatiotemporal analysis of the development of stromal-immune suppressive niches in melanoma focusing on two-photon and second harmonic generation profiling of collagen in the invasive margins show the evolution of immune exclusion and formation of a Treg-CAF (cancer associated fibroblast) niche.
Poster #42
Dissecting neurovascular coupling of odor-evoked responses in the olfactory bulb: Interventions with lamotrigine and L-NAME
Justus Verhagen
Department of Radiology & Biomedical Imaging, Yale University
To investigate neurovascular coupling embedded within the BOLD signal, we combined fMRI with optical calcium imaging of the rat olfactory bulb and by pharmacological interventions we disrupted the neurovascular coupling. The concordant effects of NOS inhibition on presynaptic Ca2+ and BOLD responses, coupled with absence of lamotrigine effects, demonstrate that olfactory bulb BOLD signals predominantly reflect presynaptic glomerular activity and its nitric oxide-dependent neurovascular coupling.
Poster #43
The Neurofeedback and Neuroimaging Experience: Exploratory results from an ongoing mindfulness-based fMRI neurofeedback study in Borderline Personality Disorder
Tenzin Walker
Yale School of Medicine, Department of Psychiatry
Neuroimaging is crucial to psychiatric research and clinical practice, yet we rarely examine the actual patient experience. This exploratory analysis evaluates satisfaction data from an ongoing Yale fMRI study, finding high overall satisfaction and significant individual variation. Leveraging these insights will help researchers optimize future scan protocols, maximize data yield, and improve clinical efficacy across neuroimaging applications.
Poster #44
Flow Matching for Tumor Contrast Enhancement Modeling in Breast MRI Images
Rui Wang
Biomedical Engineering Department, Yale University
We investigate flow matching for synthesizing contrast-enhanced breast MRI from pre-contrast images.
Session 2: Imaging Cancer
Poster #45
Adaptive Optics on Three-Photon Microscopy Enables Deep-Tissue Imaging in Mouse Brain
Shengqi Wang
Biomedical Engineering Department, Yale University
We present 3P-DASH, an adaptive optics technique integrated with three-photon microscopy for deep-tissue imaging in the mouse brain. By correcting wavefront aberrations through focus interference, 3P-DASH achieves a 9.3-fold signal improvement in CA1 neurons at ~1 mm depth beneath white matter. This fast, single-SLM approach offers a practical solution for high-quality in vivo neuroimaging in highly scattering tissue.
Poster #46
Kappa opioid receptor availability, impulsivity, and emotion dysregulation among individuals with hoarding behaviors: Preliminary evidence using [11C]EKAP PET.
Emily Weiss
Yale University School of Medicine, Department of Psychiatry
Hoarding disorder (HD) is a psychiatric condition characterized by compulsive saving and acquiring behaviors leading to excessively cluttered living spaces and profound functional impairment. The kappa opioid receptor (KOR) may be a relevant treatment target, but has not been studied in HD. This [11C]EKAP PET pilot data examines cingulo-opercular KOR in-vivo in 9 subjects with hoarding behaviors compared to age- and sex-matched healthy controls, and explores links between KOR and features of HD.
Poster #47
Whole-brain GABA and Glx mapping using SLOW and joint LTSA reconstruction
Guodong Weng
University of Bern
Whole-brain mapping of GABA and Glx is limited by the low signal-to-noise ratio of edited MRSI. This work applies two reconstruction methods, SPICE and LTSA, to 7T SLOW-editing MRSI data. Spectra reconstructed using SPICE from a 6-minute acquisition closely match 20-minute raw data, and LTSA further improves GABA and Glx map quality by jointly exploiting the shared manifold structure of SLOW-partial and SLOW-difference data, enabling faster, higher-quality whole-brain edited metabolite mapping.
Poster #48
Peritumoral Injection of 177Lu-Nanoparticles Inhibits Tumor Growth With Low Organ Normal Radiation
Moses Wilks
Yale School of Medicine
Peritumoral injection of the 177Lu labeled nanoparticle ferumoxytol (FMX) was used to inhibit tumor growth and limit normal organ radiation doses. Mice were imaged by SPECT and post-mortem tissues were analyzed to calculate dosimetry. [177Lu]Lu-FMX greatly inhibited tumor growth and increased overall survival. SPECT showed 177Lu retention in or around the tumor, with little off-target dose. Absorbed doses to tumor (111.5 Gy) exceeded dose to healthy tissues (e.g. liver, marrow) by 10-100 fold.
Session 2: Imaging Cancer
Task-Based Evaluation of AI PET Denoising for Low-Count Lesion Localization Using CHO JAFROC Analysis
Menghua Xia
Yale Biomedical Imaging Institute
AI denoising, particularly diffusion or U-Net based denoising, is a valuable tool for improving image quality and potentially lesion detection in low-dose PET imaging. However, using CHO within an FROC study, we show that it is not a universal fix; its benefits are highly conditional on the image noise, lesion contrast, and anatomical background, demonstrating that visually "smoother" images do not automatically guarantee better clinical results in all conditions.
Session 2: Imaging Cancer
CAIPI3 accelerated 4D flow bSSFP at 3T for improved diastolic function evaluation
Jie Xiang
Radiology and Biomedical Imaging
We proposed a 4D flow bSSFP at 3T accelerated by CAIPI3, for rapid diastolic function evaluation in a short 5min free-breathing scan.
Session 4: Body Imaging
A Modular Platform (BDM + PNI) for Application-Specific PET: An All-digital In-Beam System for Proton Range Monitoring
Zizhuo Xie
University of Science and Technology of China
The BDM + PNI platform lets clinicians and researchers build application-specific PET systems without instrumentation expertise.The BDM + PNI platform lets clinicians and researchers build application-specific PET systems without instrumentation expertise. BDM detector blocks output raw MVT data; PNI composes a swappable software pipeline (MVT → singles → coincidence → image). The authors built an in-beam PET system for proton range monitoring entirely on it, resolvi
Session 1: Image Technology
Poster #49
Estimating Cardiac Material Parameters through Physics-Informed Gradient Descent Optimization
Xuanang Xu
Yale University
This research presents a physics-informed optimization framework to non-invasively estimate intrinsic cardiac material properties (e.g., stiffness and contractility) directly from 4D ultrasound and MRI data. By embedding continuum mechanics into a gradient descent pipeline, the algorithm can extract fundamental material parameters from observable heart motion. This approach paves the way for subject-specific biomechanical profiling to better monitor cardiac remodeling and disease progression.
Session 4: Body Imaging
Imaging Adhesion of Diverse Pathogenic Bacteria to Host Surfaces through Type4-Pili via Electron Transfer
Sibel Ebru Yalcin
Molecular Biophysics and Biochemistry, Microbial Sciences & Biomedical Imaging Institutes
Infections are initiated via adhesion of bacteria to host surfaces, although there is an electrostatic repulsion among these surfaces. Diverse bacteria use filamentous appendages called Type 4 pili (T4P) to overcome repulsion. Combining single-cell imaging and electrostatic force microscopy on Neisseria, Pseudomonas and Paenibacillus, we show that bacteria use T4P to facilitate adhesion using stacked aromatic residues in pilus for electron transfer offering novel strategy to prevent infections.
Poster #50
Parallel acquisition of multi-contrast MRI and Deuterium Metabolic Imaging on a preclinical platform
Qinyuan Yang
Yale University
This abstract aims on introducing new Magnetic Resonance Imaging (MRI) - Deuterium Metabolic Imaging (DMI) preclinical methods which implement parallel MRI-DMI acquisition to obtain anatomical and metabolic information. We present various MRI methods including T1-weighted (T1W) and T2-weighted (T2W) MRI, Diffusion Weighted Imaging (DWI) and Magnetization Transfer Contrast (MTC) MRI acquired in conjunction with steady state DMI. A practical example is demonstrated on a RG2 brain tumor in vivo.
Session 1: Image Technology
Neurochemical and genetic organization of head impact effects on cortical neurophysiology
Kevin Yu
Department of Radiology & Biomedical Imaging, Yale School of Medicine
Neurophysiological changes related to head impact exposure are spatially organized by cortical neurochemical and genetic gradients, and are associated with cognitive symptom severity.
Poster #51
Correcting Pretrained Diffusion Priors for Unrolled Accelerated MRI Reconstruction
Tian Yu
Yale University
Diffusion priors are trained along a Gaussian perturbation path, but unrolled MAP reconstruction visits iterates that leave it. We integrate a pretrained EDM denoiser into an unrolled network and add a 437k-parameter learned score correction adapted to the reconstruction trajectory. On simulated single-coil knee MRI at R=4, the corrected prior improves NMSE, PSNR and SSIM over MoDL, diffusion posterior sampling baselines, and the uncorrected variant.
Poster #52
Spatially-Variant Double-Gaussian Resolution Modeling for PET Reconstruction
Tianyi Zeng
Yale University
We developed a spatially-variant double-Gaussian resolution model for PET reconstruction that accounts for detector crosstalk and inter-crystal scatter. Using point-source, Derenzo phantom, and preliminary human PET data, we show that combining double-Gaussian modeling, spatial variation, and LUT interpolation improves resolution recovery compared with conventional PSF models.
Poster #53
Multimodal Quantification of Muscle Involvement in Inclusion Body Myositis Using Electrical Impedance Myography and Quantitative Ultrasound
Wayne Zhong
Yale Neurology
Quantitative ultrasound and electrical impedance myography (EIM) may provide complementary, noninvasive markers of muscle involvement in inclusion body myositis. This study compares ultrasound echogenicity and 300-kHz EIM phase across IBM and control participants and evaluates whether higher echogenicity corresponds to lower EIM phase.
Poster #54
Development of Microfluidic Phantom for Evaluation of High-resolution PET Scanners
Yue Zhuo
Radiology and Biomedical Imaging, Yale Biomedical Imaging Institute, Yale School of Medicine
We developed a microfluidic phantom (YNOT) for benchmarking high-resolution PET scanners. MSLA 3D printing successfully fabricated sub-millimeter channels down to 0.5 mm. Cross-scanner evaluation on Mediso LFER 150 and NeuroEXPLORER demonstrated resolvability of the microfluidic structures, providing insight on minimum observable feature size. The phantom enables quantification of partial volume effects and will guide development of larger brain phantoms.
Poster #55
Reflective Sidewall Photonic Crystals for Inter-Crystal Optical Crosstalk Suppression in BaF₂ PET Detectors
Yue Zhuo
Radiology & Biomedical Imaging
To suppress inter-crystal optical crosstalk in BaF2 PET detectors without adding dead space, reflective sidewall photonic crystals (PhCs) were engineered. A 1D PhC (AlN/SiO2, Bragg mirror, <0.5 µm thick) achieved high reflectivity across the fast UV emission band of BaF2 crystal, yielding a high light collection efficiency boost over bare Fresnel reflection. It also selectively transmits the slow component to reduce pile-up, advancing 10 ps TOF-PET goals.
Session 1: Image Technology