Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Advanced Segmentation in MRI Across Scales and Domains

Advanced Segmentation in MRI Across Scales and Domains

Oral

Analysis Methods

Wednesday, 13 May 2026

Meeting Room 1.40

16:00 - 17:50

Moderators: Shuncong Wang & S Shailja

Session Number: 507-04

No CME/CE Credit

Similar Sessions

This oral session presents methodologically innovative submissions focused on segmentation in MRI, spanning brain and cardiac applications across human, fetal, neonatal, and preclinical imaging. The selected works highlight advances in high-resolution neuroanatomical segmentation, cortical and subcortical structure modeling, and cardiac structure and function segmentation, leveraging physics-guided learning, self-supervision, test-time adaptation, and domain generalization. Collectively, the session emphasizes robust segmentation methodologies that generalize across resolutions, contrasts, developmental stages, and imaging domains, reflecting the current frontier of segmentation research in MRI.

Skill Level: Intermediate

16:00		507-04-001. Physics-Guided Few-Shot Learnable Active Contour Model for Quantitative Body Composition Analysis on MRI PDFF images Lin Yang, Chuanli Cheng, Zhanli Hu, Xin Liu, Hairong Zheng, Chao Zou Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China Impact: This physics-guided learnable active contour model enables high-precision, whole-body tissue segmentation from MRI-PDFF images with minimal annotations. It facilitates scalable body composition analysis, overcoming a key bottleneck in clinical research and population health studies.
16:11		507-04-002. Evaluating Direct Plane Prediction Model and conventional Landmark-based Approach for Robust Multi-View Cardiac MRI Planning Viswanath Pamulakanty Sudarshan, Vineeth VS, Rajat Kumar, Suranjita Ganguly, Jaladhar Neelavalli, Suthambhara Nagaraj, Yogesh k Mariappan Philips Healthcare, Bengaluru, India Impact: By advancing automated cardiac MR view planning, we enhance diagnostic accuracy and workflow efficiency in cardiac imaging, benefiting both expert and non-expert clinicians. Prescribing outflow-tract in addition to standard views sets the stage for complex cardiac evaluations, broadening clinical capabilities.
16:22		507-04-003. BioTTA: Maximizing Domain Generalization in Automatic Fetal Brain Biometry with Test-Time Adaptation YIJIN LI, Mingxuan Liu, Hongjia Yang, Xiaotian Hu, Yi Liao, Kasidit Anmahapong, Ziang Wang, Juncheng Zhu, Yingqi Hao, Yifei Chen, Haoxiang Li, Ziyu Li, Fenglin Jia, Haibo Qu, Qiyuan Tian Beihang University, Beijing, China Impact: The proposed unsupervised test-time adaptation framework BioTTA enables robust cross-domain generalization of fetal brain biometry without requiring manual labels, thereby improving model reliability across heterogeneous MRI scanning protocols and hardware configurations, simplifying clinical workflows, and supporting large-sale, multi-center neuroscience research.
16:33		507-04-004. Open-Access Mouse Cardiac MRI Dataset and Segmentation Model: A Deep Learning Approach for Preclinical Research Wan Shah, Daniel Stuckey, Tina Yao, Mark Wrobel, Ruaraidh Campbell, Vivek Muthurangu, Jennifer Steeden Centre for Translational Cardiovascular Imaging, University College London, London, United Kingdom Impact: This work introduces the first open-access mouse cardiac MRI dataset and deep learning segmentation model, with a user-friendly web app, enabling fast, reproducible cardiac analysis to accelerate post-processing of preclinical imaging.
16:44		507-04-005. Age-Conditioned Neonatal Choroid Plexus Segmentation using Adaptive Conditional Instance Normalization Junghwa Kang, Dayeon Bak, Hyun Gi Kim, Na-Young Shin, Yoonho Nam Hankuk university of Foreign Studies, gyeonggi-do, Korea, Republic of Impact: Neonatal choroid plexus segmentation is challenging due to rapid age-dependent morphological changes. Our age-conditioned automated framework enables improved segmentation across ages using a single model. This method may contribute to developmental studies investigating glymphatic function and neurodevelopmental outcomes.
16:55		507-04-006. HypSeg: A Deep Learning Pipeline for Hypothalamic Subregion Segmentation Trained on High-Resolution 7 Tesla MRI Haihan Zhao, William Salmon, Jinghang Li, Martina Bocchetta, Rebecca Thurston, Sossena Wood, Tales Santini, Tamer Ibrahim Carnegie Mellon University, Pittsburgh, United States of America Impact: This study establishes an automated 7T deep learning method for precise hypothalamic subregion mapping, reducing reliance on manual labeling and supporting large-scale studies of neuroendocrine and metabolic brain mechanisms at unprecedented spatial resolution.
17:06		507-04-007. Deep Learning 4D Segmentation of Short-Axis CMR to Quantify Peak Filling and Ejection Rate in Single Ventricle Patients Tina Yao, Nicole St. Clair, Gabriel Miller, Jennifer Steeden, Rahul Rathod, Vivek Muthurangu University College London, London, United Kingdom Impact: We developed a novel 4D UNet3+ segmentation model that produces spatially and temporally consistent ventricular segmentations in over 1000 single-ventricle short-axis CMR. The model allows rapid calculation of peak ejection rates, which were significantly associated with adverse outcomes.
17:17		507-04-008. Test-time optimization for cortical surface reconstruction across image resolutions/contrasts using untrained neural networks Haoxiang Li, Mingxuan Liu, Divya Varadarajan, Zhangxuan Hu, Qiyuan Tian, Jonathan Polimeni Tsinghua University, Beijing, China Impact: We introduce a pretraining–free, generalizable cortical surface reconstruction method delivering accurate results across imaging resolutions, contrasts, species and brain regions, reducing barriers to cross-domain neuroimaging. This extends morphometric analyses to new datasets, accelerating comparative/translational neuroscience while standardizing analyses.
17:28		507-04-009. Spatio-Temporal Segmentation and Motion Analysis of Uterine Layers in Cine MRI Using Unet-LSTM and FlowNet-Lite Smiti Tripathy, Milauni Desai, Lieselotte Kratzsch, Michael Uder, Matthias May, Jana Hutter Friedrich-Alexander University Erlangen-Nuremberg, Erlangen, Germany Impact: This study lays the foundation for deep-learning based automatic segmentation and motion analysis of the uterine cine MRI, that would in future facilitate the automated quantification of uterine peristalsis and understanding of peristaltic alterations in pathological conditions such as Adenomyosis.
17:39		507-04-010. Anatomy-Aware Dynamic Causal Enhancement: A 3D-Capable Framework for Clinically Compatible Single-Source Medical Segmentation Yixiong Shi, Fuhua Yan The University of Sydney, Sydney, Australia Impact: AADCA overcomes ’black-box‘ 3D spatial feature-based gaps in single-source domain generalization(CT to MRI) tasks, facilitating the innovation of multi-modal domain adaptation and enhancing 3D medical imaging segmentation reliability for clinical translation.