Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Generative Modeling: A Unifying Lens on MRI

Generative Modeling: A Unifying Lens on MRI

Oral

Analysis Methods

Wednesday, 13 May 2026

Meeting Room 1.40

08:20 - 10:10

Moderators: Fan Lam & Onat Dalmaz

Session Number: 507-02

No CME/CE Credit

Similar Sessions

The session showcases advanced generative modeling techniques developed for image/data analysis tasks in MRI.

Skill Level: Basic,Intermediate,Advanced

08:20		507-02-001. Harmonization for a Black-box Model using Disentanglement-based Generator and Bayesian Optimization Minjun Kim, Dong Ju Mun, Hwihun Jeong, Haechang Lee, Se Young Chun, Jongho Lee Seoul National University, Seoul, Korea, Republic of Impact: BboxHarmony advances MRI harmonization by enabling adaptation to black-box models, broadening applicability to real-world settings where data and model access are restricted. This approach paves the way for more robust and generalizable AI deployment across diverse MRI domains.
08:31		507-02-002. Increased Decoding Accuracy Using Adaptive Lag Responses Based on an Attention Mechanism Nguyen Huynh, Gopikrishna Deshpande Auburn University Neuroimaging Center, Auburn University, Auburn, United States of America Impact: By adaptively aligning stimulus timing with brain responses, this approach enhances decoding accuracy and yields more reliable interpretations of neural representations, providing a promising framework for modeling dynamic brain–stimulus relationships in naturalistic and high-resolution fMRI studies.
08:42		507-02-003. Beyond Correlation: Graph Diffusion Autoregression Captures Directional Information Flow in Aging and Alzheimer's Disease Felix Schwock, Daniel Nordgren, Rachel Iritani, Les Atlas, Azadeh Yazdan-Shahmorad, Hesamoddin Jahanian University of Washington, Seattle, United States of America Impact: Our framework introduces a paradigm shift by incorporating structural constraints into functional connectivity estimates and capturing dynamic, directional information flow across brain regions, potentially uncovering novel spatiotemporal biomarkers for neurological diseases.
08:53		507-02-004. A Multi-Task Diffusion Framework for Synthetic Contrast-Free LGE and Simultaneous Myocardial Infarction Segmentation Jing Qi, Xiuzheng Yue, Miao Hu, Yinyin Chen, Hang Jin, Tao Li, Kunlun He Chinese PLA General Hospital, Beijing, China Impact: This study demonstrates the feasibility of virtual LGE generation with simultaneous segmentation, paving the way for contrast-free, fully automated, and reliable myocardial scar quantification.
09:04		507-02-005. Contrastive Radiomics-Aligned Latent Diffusion Model for High-Fidelity CT Synthesis from MRI Bin Zhang, Taofeng Xie, zhuoxv cui, Xingliang Li, Jinxuan Lyu, Haonan Li, Shiyi Zhang, Haifeng Wang, Dong Liang, Yihang Zhou Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China Impact: With physiologically grounded radiomic constraints embedded, CRaLDM generates fast, high-fidelity synthetic CT images with sufficiently accurate CT numbers (HU) required for radiotherapy, which could accelerate the practical implementation of this precision therapy approach.
09:15		507-02-006. Quantitative MRI Mapping using Diffusion Models with Data Consistency on 3D Fast Zero Echo Time Acquisition Shishuai Wang, Florian Wiesinger, Noemi Sgambelluri, Carolin Pirkl, Stefan Klein, Juan Hernandez-Tamames, Dirk H. J. Poot Erasmus MC, Rotterdam, Netherlands Impact: This work introduces a qMRI mapping approach that combines the strengths of data-driven generative AI and physics information. The method achieves high-quality quantitative maps and demonstrates superior performance over traditional dictionary matching.
09:26		507-02-007. FIRE-integrated CineGen: inline conditional flow-matching super-resolution for real-time Cine MRI Changyu Sun, Yu Wang, Lei Jiang, John Grinstead, Kelvin Chow, Xiaoming Bi, Samantha Baxter, Senthil Kumar, Talissa Altes University of Missouri, Columbia, United States of America Impact: This work demonstrates, for the first time, a flow matching generative AI super-resolution model operating inline on the MRI scanner. CineGen overcomes diffusion-model latency barriers, enabling high-quality cardiac images and advancing the translation of AI-driven reconstruction within clinical MRI workflows.
09:37		507-02-008. Flowdiff: Cardiac Cine Frame Interpolation Combining Optical Flow and Diffusion Models Zhaochi Wen, Jing Cheng, Daisong Gan, Yuliang zhu, Dong Liang School of Biomedical Engineering & State Key Laboratory of Advanced Medical Materials and Devices, ShanghaiTech University, Shanghai, China Impact: Flowdiff integrates the advantages of optical flow (stable) and diffusion (realistic) models to achieve accurate cardiac cine interpolation at arbitrary time points within a unified framework, enhancing clinical diagnosis, and benefiting patients.
09:48		507-02-009. Joint Diffusion and Classification to Learn Deep Brain Stimulation Outcomes from Presurgical Targets Alexandra Roberts, Dominick Romano, Mert Sisman, Maneesh John, Sema Akkus, Jip de Bruin, Jinwei Zhang, Ceren Tozlu, Alexey Dimov, Thanh Nguyen, Pascal Spincemaille , Ki Sueng Choi, Brian Kopell, Yi Wang Cornell University, Ithaca, United States of America Impact: The proposed joint diffusion-classification model simultaneously predicts outcomes directly from presurgical imaging and learns the underlying distribution of the dataset. This model employs classifier-free guidance to generate robust samples from noisy, imbalanced labels, improving patient selection in deep brain stimulation.
09:59		507-02-010. Robust Posterior Sampling for MRI Reconstruction by the Preconditioned Unadjusted Langevin Algorithm Tina Holliber, Moritz Blumenthal, Verena Fink, Jon Tamir, Martin Uecker Graz University of Technology, Graz, Austria Impact: The proposed method provides fast and robust diffusion posterior sampling for different MRI reconstruction problems without tuning the step size or regularization parameter, enabling high-quality MRI reconstruction and uncertainty quantification within seconds.