Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition

Vision and Language Models in MRI

Traditional Poster

Analysis Methods

Wednesday, 13 May 2026

Traditional Posters | Exhibition Hall

14:35 - 15:30

Session Number: 570-06

No CME/CE Credit

Similar Sessions

The use case of large language models and vision language models in radiological workflows and reducing error rates in reporting.

Skill Level: Basic,Intermediate,Advanced

		570-06-204. Deep Learning Framework for Automated Reporting of Degenerative Changes in the Lumbar Spine from MRI Yao-Wen Liang, Cheng-Ru Yang, Ssu-Ju Li, Ching-Wen Chang, Ting-Chun Lin, Wei-Der Chung, Chao-Hung Kuo, You-Yin Chen National Yang Ming Chiao Tung University, Taipei, Taiwan Impact: This study validates a scalable foundation for an automated radiology reporting system.
		570-06-205. Fine-Tuned Multimodal GPT-4o for Generating Diagnostic Impressions in Breast MRI: Insights into Non-Mass Enhancement Lesions Meihong Sheng, Jiahuan Tang Nantong First People's Hospital Southeast University, Nantong, China Impact: This MRI-based nomogram enables personalized treatment for HR-positive breast cancer by integrating preoperative imaging and clinical features, thus offering a non-invasive, cost-effective approach for recurrence risk stratification and extended endocrine therapy decisions.
		570-06-206. Textual context in prostate MRI: Learning meaningful representations from radiology reports Peter Lais, Lavanya Umapathy, Patricia Johnson, Hersh Chandarana, Daniel Sodickson NYU Grossman School of Medicine, New York, United States of America Impact: Meaningful representations of clinical reports, which may include information regarding study indications, prior results, and assessments, can augment imaging-based models to enhance clinical decision support.
		570-06-207. NeuroRAP: A Retrieval-Assisted Vision-Language Model for Prognosis Prediction in Neurodegenerative Disorders Dongang Wang, Geng Zhan, Yang Ma, Zihao Tang, Heidi Beadnall, Yuanzhao Chen, Linda Ly, Qianyu Wang, Chun-Chien Shieh, Chenyu Wang, Michael Barnett The University of Sydney, Sydney, Australia Impact: This work pioneers the integration of vision-language modeling with automatically generated imaging reports for prognosis prediction in neurodegenerative diseases. The framework enhances both predictive performance and interpretability, providing a scalable and explainable AI approach for clinical decision support.
		570-06-208. A Hybrid Vision Mamba-Transformer Network for Tissue Quantification from MRF Jing Zou, Rui Li Tsinghua University, Beijing, China Impact: We devise a promising network for tissue quantification from MRF, due to its good performance and low computational cost. In addition, the proposed method demonstrates significant potential for future applications aimed at accelerating MRF acquisition.
		570-06-209. Transformer-Based Structural-to-Diffusion MRI Synthesis with Dual-Pruned Attention Boosts Dementia Classification Performance Tamoghna Chattopadhyay, Dhruv Kudalkar, Sophia Thomopoulos, Jose-Luis Ambite, Greg Ver Steeg, Paul Thompson USC Stevens Neuroimaging and Informatics Institute, Keck School of Medicine, University of Southern California, Los Angeles, United States of America Impact: Transformer-based structural-to-diffusion synthesis generates realistic, clinically meaningful DTI maps for individuals lacking diffusion imaging data. These synthetic maps can augment neuroimaging datasets, mitigate data scarcity, and enhance downstream diagnostic models, supporting scalable applications in neurodegenerative disease analysis.
		570-06-210. Multimodal LLMs Can Name It but Struggle to Place It: Spatial Reasoning Gaps for Radiology Workflows Nithya Ramesh, Ashish Saxena, Sanand Sasidharan, Anuradha Kanamarlapudi GE HealthCare, Bengaluru, India Impact: Multimodal Large Language models (MMLLMs), while strong at semantic MRI tasks, fail at spatial localization, a key component of medical image reasoning. These findings redefine expectations for integrating MMLLMs into radiology workflows such as automated interpretation and structured reporting.
		570-06-211. A Diffusion Model with Multi-task Learning for Diagnosing Acute Myocardial Infarction from Non-contrast Cardiac Cine MR YIMING ZHU, Hanxi Liao, Dongaolei An, Lian-Ming Wu, Haikun Qi School of Biomedical Engineering, ShanghaiTech University, Shanghai, China Impact: Previously developed CGE was improved by using a novel spatiotemporal diffusion model and multi-task learning to synthesize LGE-equivalent images from contrast-free cine cardiac MR for diagnosing both scar and microvascular obstruction, providing a contrast-free solution for acute myocardial infarction diagnosis.