Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Segmentation for Cardiac Applications

Segmentation for Cardiac Applications

Digital Poster

Analysis Methods

Tuesday, 12 May 2026

Digital Posters Row I

13:40 - 14:35

Session Number: 468-03

No CME/CE Credit

Similar Sessions

This digital poster session showcases recent advances in segmentation for cardiac and vascular MRI, with a strong emphasis on automation, robustness, and clinical translation. The abstracts span ventricular and myocardial segmentation, and pulmonary and aortic vessel analysis. Methodological themes include inline and real-time workflows, anatomy-aware and perfusion-informed learning, foundation and semi-supervised models, cross-center generalization, and visualization frameworks integrating AI and VR. These contributions highlight how modern segmentation approaches are enabling reproducible functional assessment, hemodynamic analysis, and disease characterization across diverse cardiovascular MRI applications.

Skill Level: Intermediate

		468-03-001. Automated Inline Segmentation for Real-Time Evaluation of Ventricular Function in Cardiac MRI Omer Demirel, Spencer Waddle, Enas Ahmed, Dinghui Wang, Tzu Cheng Chao, Joppe Kietselaer, Jouke Smink, Jacinta Browne, Tim Leiner Philips North America Clinical Science, Rochester, United States of America Impact: This work demonstrates a CMR segmentation tool operating directly on the scanner, providing immediate functional analysis during scanning. The approach demonstrated robust performance across both 3T and 0.6T systems, enabling clinical integration of automated ventricular assessment.
		468-03-002. Phasor based segmentation of porcine hearts in hyperpolarized 13C MRI Konstantin Müller, Christoffer Laustsen, Esben Hansen, Josef Mayer, Saar Székely, Pascal P.R. Ruetten Ulm University, Ulm, Germany Impact: Temporal phasor clustering provides a concise descriptor of waveform shape and timing, and is based solely on metabolite dynamics^3,4,5. By mitigating partial-volume effects relative to expert-drawn ROI masks, it provides a standardized metabolic segmentation in 13C-MRI.
		468-03-003. A Comparative Study of Automatic Myocardial Segmentation Methods for Iron Overload Detection in Dark-Blood T2* MR images Ambra Checchetto, Amalia Lupi, Giada Businaro, Simone Perra, Alessandro Giupponi, Valentina Visani, Alessia Pepe, Marco Castellaro University of Padova, Padova, Italy Impact: TotalSegmentator fine-tuned for automatic myocardium segmentation in dark-blood T2* MR images enhances diagnostic accuracy for iron overload assessment while minimizing operator bias and processing time. This approach will improve workflow efficiency and reproducibility, advancing both clinical practice and research applications.
		468-03-004. AutoPAP: Automated prediction of pulmonary artery pressure from cardiac MRI in pulmonary hypertension with deep learning. Ruaraidh Campbell, Tina Yao, Anirudh Raman, Mark Wrobel, Catherine Beattie, Charo Bruce, Niromila Nadarajan, Ruta Virsinskaite, Dan Knight, Jennifer Steeden, Vivek Muthurangu University College London, London, United Kingdom Impact: We demonstrate a pipeline for automating the prediction of mean pulmonary artery pressure (mPAP) in pulmonary hypertension from CMR studies using deep learning - potentially providing a non-invasive method for estimating mPAP and reducing need for right heart catheterisation.
		468-03-005. CNN-based 4D Segmentation Improves Reproducibility of Hemodynamic Parameters in Aortic 4D Flow MRI Hinrich Rahlfs, Julio Garcia, Markus Hüllebrand, Sebastian Schmitter, Sarah Nordmeyer, Titus Kühne, Heiko Stern, Christian Meierhofer, Andreas Harloff, Sebastian Kelle, Peter Bannas, Jeanette Schulz-Menger, Ralf Trauzeddel, Anja Hennemuth Deutsches Herzzentrum der Charité (DHZC), Berlin, Germany Impact: This method enables fast automatic 4D Flow MRI analysis and hemodynamic parameter calculation. By reducing user variability and manual effort, it supports reliable multicenter studies, promotes wider clinical adoption, and advances automated assessment of aortic function and cardiovascular disease.
		468-03-006. A fully automated workflow for acquiring and quantifying 4D flow MRI in the aorta Erik Stein, Marc Vornehm, Jens Wetzl, Florian Knoll, Daniel Giese Research & Clinical Translation, Magnetic Resonance, Siemens Healthineers AG, Erlangen, Germany Impact: Using our workflow, native 4D flow MRI can be automatically planned and quantified in the aorta. This improves standardization and accessibility of 4D flow, noticeably reduces quantification complexity, and consequentially facilitates clinical adoption.
		468-03-007. Domain-Specific Fine-Tuning of SAM2.1 for 3D Segmentation of Cardiac Structures from Cine MRI Tijana Geroski, Amir Amini University of Kragujevac, Kragujevac, Serbia Impact: This work demonstrates the successful adaptation of SAM2.1 for 3D cardiac MRI segmentation, achieving high accuracy with reduced computational cost. The approach enables reliable ejection fraction estimation, supporting efficient and clinically meaningful analysis of cardiac function with minimal manual intervention.
		468-03-008. Foundation Model-Assisted Robust Semi-Supervised Learning for Cardiac MRI Segmentation Rui Zhang, Zijian Zhou, Yuan Feng, Yining Wang, Peng Hu School of Biomedical Engineering, ShanghaiTech University, Shanghai, China Impact: Limited annotated data often leads to suboptimal performance in deep-learning models. Our method overcomes this by effectively generating pseudo-labels for unlabeled data, combining model-based and model-independent strategies. This results in significantly improved segmentation performance and strong potential for clinical applications.
		468-03-009. Robust Cross-Center Vessel Lumen and Wall Segmentation on Intracranial Vessel Wall MRI via Anatomy-Aware Learning Xin Wang, Gador Canton, Zhiwei Tan, Yin Guo, Dan Cheng, Mona Kharaji, Beibei Sun, Jie Sun, Duygu Geleri, David Tirschwell, Thomas Hatsukami, Mahmud Mossa-Basha, Niranjan Balu, Chun Yuan University of Washington, Seattle, United States of America Impact: This work enables reliable intracranial vessel-wall segmentation across clinical sites without new annotations, allowing scalable vessel-wall MRI analysis. By leveraging shared anatomical priors for generalizable vessel segmentation, it improves cross-center consistency and supports multi-institution cerebrovascular studies and clinical translation.
		468-03-010. Evaluating the Role of Anatomical Priors in Deep Learning-Based Myocardial Scar Segmentation Across Multiple Datasets Isabel Margolis, Valery Visser, Stefano Buoso, Sebastian Kozerke ETH Zurich and University of Zurich, Zurich, Switzerland Impact: Two-stage models incrementally improve myocardial scar segmentation but at increased training cost. Our study further demonstrates that segmentation performance varies depending on the dataset and underscores the need for multi-dataset evaluation to advance robust and reproducible DL segmentation of scar.
		468-03-011. Automated Extraction of False Lumen Flow Dynamics in Aortic Dissection Using 4D Flow MRI Sebastian Cohn, Elizabeth Weiss, Christopher Mehta, Charilaos Apostolidis, William Dong, Michael Markl, Bradley Allen Biomedical Engineering, Northwestern University, Chicago, United States of America Impact: In aortic dissection, automated regional peak orthogonal flow (POF) mapping identifies regions of hemodynamically active communication between the true lumen (TL) and false lumen (FL) and complements diameter criteria. If validated longitudinally, POF could enable MRI-based hemodynamic risk scores.
		468-03-012. Automated Segmentation of Multi-Region 4D Flow MRI Using Deep Learning Yuyang Ren, Ruiyu Cao, Zijian Zhou, Chengyan Wang, Hao Li, Peng Hu ShanghaiTech University, Shanghai, China Impact: This segmentation model enables fast, automated segmentation for 4D flow MRI, significantly improving operational efficiency.
		468-03-013. Deep learning and radiomics based disease discrimination using cardiac cine MRI in patients with left ventricular hypertrophy Yifeng Gao, Zhen Zhou, Wenjia Wang, Lei Xu Beijing Anzhen Hospital, Capital Medical University, Beijing, China Impact: This study developed an automated segmentation and radiomics model capable of discriminating among hypertrophic cardiomyopathy (HCM), hypertension heart disease (HHD) and cardiac amyloidosis (CA). The framework provides a robust and non-invasive tool for etiological diagnosis of left ventricular hypertrophy (LVH).
		468-03-014. Perfusion-Informed Deep Learning for Automated Pulmonary Vessel Segmentation Pavlos Panos, Oliver Bieri, Grzegorz Bauman University of Basel, Basel, Switzerland Impact: This work provides a fully automated generalizable solution for vascular signal removal, enabling unbiased and rapid quantitative parenchymal signal analysis in healthy and diseased lungs for accelerating the development of reliable biomarkers of pulmonary pathologies.
		468-03-015. AI & VR for Visualizing & Segmenting of Intracardiac Anatomy and Great Vessels from Cardiovascular Magnetic Resonance Images Arjun Tomer, Maria Chase, Jonathan Awori, Nita Chaudhuri, Dhaval Chauhan, Christopher Mascio, Jai Udassi, Mehdi Hedjazi Moghari West Virginia University, Morgantown, United States of America Impact: Integrating AI-assisted cardiac MRI segmentation with virtual reality offers rapid, accurate 3D model generation and immersive anatomical visualization. This approach will benefit pediatric patients with more precise preoperative assessments, while transforming how cardiologists, radiologists, and surgeons communicate and plan interventions.
		468-03-016. BCA-MT: Boundary-aware, Class-distribution-Aligned Mean-Teacher for Semi-supervised Cardiac MRI Segmentation on ACDC Pengchen Liang, Zhifeng Chen, Haishan Huang, Kaiting Wang, Shiwei Wang, Xiaoyun Liang Neusoft Medical Systems Co. Ltd, Shanghai, China Impact: BCA-MT cuts annotation needs while preserving boundary accuracy, enabling reliable cardiac MRI segmentation with 5–10% labels. Clinicians gain faster, consistent measurements; scientists can explore 2.5D/3D and temporal SSL. This lowers costs, accelerates deployment, and broadens access to timely cardiac care.