Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition - Segmentation: Automated, Scalable, and Clinically Oriented MRI Analysis

Segmentation: Automated, Scalable, and Clinically Oriented MRI Analysis

Power Pitch

Analysis Methods

Monday, 11 May 2026

Power Pitch Theatre 2

13:50 - 15:26

Moderators: Sevgi Gokce Kafali & Yuki Arita

Session Number: 352-02

No CME/CE Credit

Similar Sessions

This PowerPitch session highlights a broad and methodologically rich collection of abstracts focused on segmentation in MRI across diverse anatomies, patient populations, and clinical workflows. The selected works span brain, cardiac, spine, prostate, fetal, placental, and whole-body MRI, and address key challenges including annotation efficiency, domain adaptation, model compression, anatomical consistency, uncertainty modeling, and inline/real-time deployment. These studies reflect the growing emphasis on robust, reproducible, and clinically translatable segmentation pipelines, ranging from foundational frameworks and benchmarks to disease-specific applications in neurology, oncology, cardiology, and prenatal imaging.

Skill Level: Intermediate

13:50		352-02-001. Automated assessment of internal capsule maturation in neonatal 3D-reconstructed structural T2-weighted MRI at 7T Chiara Casella, Alena Uus, Luke Dedominicis, Benjamin Clayden, Jucha Willers Moore, Philippa Bridgen, Pierluigi Di Cio, Ines Tomazinho, Cidalia Da Costa, Dario Gallo, Sophie Arulkumaran, Sharon Giles, Jonathan O'Muircheartaigh, Maria Deprez, Jo Hajnal, Serena Counsell, MARY RUTHERFORD, Shaihan Malik, Tomoki Arichi King's College London, London, United Kingdom Impact: We present a novel, anatomically detailed method for assessing internal capsule myelination and injury in neonates, supporting objective radiological evaluation and large-scale studies of early brain maturation.
13:52		352-02-002. Automated segmentation of the subthalamic nucleus in Parkinson’s disease for deep brain stimulation using 7T MRI Matthijs de Buck, Matthan Caan, Jip de Bruin, Yarit Wiggerts, Wietske van der Zwaag, P. Richard Schuurman, Maarten Bot Spinoza Centre for Neuroimaging, Amsterdam, Netherlands Impact: The subthalamic nucleus is an effective target for deep brain stimulation in Parkinson’s disease. We present an nnU-Net based framework for automated subthalamic nucleus segmentation from 7T MRI scans, trained on 200 manual segmentations performed for DBS surgery.
13:54		352-02-003. Domain adaptation and Model Compression for Glioma Segmentation in Sub-Saharan African MRI Willem Boonzaier, Farhana Moosa, Udunna Anazodo University of the Free State, Bloemfontein, South Africa Impact: Lightweight, domain-adapted models achieve comparable glioma segmentation performance on Sub-Saharan African MRI scans despite computational constraints. This work advances equitable access to AI-assisted brain tumor diagnosis in resource-limited settings, improving care for underrepresented populations.
13:56		352-02-004. A Semi-Automated Method for Body Fat Quantification Using MRI Harisha C, RAVI CHANDIRA APPAJI RAVI CHANDIRA APPAJI, Yatin sharma, Gregor Thoermer, Rakesh Kumar, Bhairav Mehta, Devasenathipathy Kandasamy Siemens Healthcare Pvt. Ltd., BENGALURU, India Impact: Increased visceral fat is a major risk factor for cardiometabolic diseases. Accurate, quantification is essential for precise patient risk stratification. Current bottleneck is time and labor-intensive segmentation. This approach addresses this concern enabling evaluation of body fat as imaging biomarkers.
13:58		352-02-005. Biopsy-Informed Automatic Labeling for Prostate MRI: Evaluating a Knowledge-Transfer Pipeline Jinyang Yu, Mark Ladd, Tristan Kuder, David Bonekamp German Cancer Research Center (DKFZ), Heidelberg, Germany Impact: A practical, biopsy-aware automatic labeling pipeline enables training and release of shareable lesion-segmentation models from routine data, unlocking unannotated cohorts for continual in-house model improvement.
14:00		352-02-006. Automated ESCC Segmentation on Free-Breathing 3D-GRE: A Comparison of nnUNet and UMamba Funing Chu, Bingmei Bai, MENGZHU WANG, Yue Wu, Jinrong Qu The Affiliated Cancer Hospital of Zhengzhou University & Henan Cancer Hospital, Zhengzhou, China Impact: The UMamba model offers a robust, high-performance tool for automated segmentation of esophageal squamous cell carcinoma (ESCC) on high-resolution MRI. This can significantly reduce manual segmentation time and inter-observer variability in clinical practice, supporting more precise, personalized treatment planning.
14:02		352-02-007. Towards reproducible perivascular space quantification: an open-source perivascular space segmentation benchmark Merel van der Thiel, Eva van Heese, Britt TJ van den Heuvel, Jacobus Jansen, Brady Williams, Jose Bernal Maastricht University Medical Center, Maastricht, Netherlands Impact: The efforts of the PVS repository team will establish an open-source platform for software code related to perivascular space quantification, minimizing duplicate development, enhancing reproducibility, and providing a benchmark for future development and comparison of segmentation methods.
14:04		352-02-008. SpineLabelNet: Vertebrae Labelling on Spine MRI 2D Localizers Ashish Saxena, Chitresh Bhushan, Dattesh Dayanand Shanbhag GE HealthCare, Bengaluru, India Impact: SpineLabelNet enhances clinical workflows by automating vertebral labelling with high accuracy, reducing manual effort and radiologist workload. It improves diagnostic precision, minimizes errors, and accelerates patient evaluation, contributing to more efficient and reliable spine imaging in clinical practice.
14:06		352-02-009. Anatomically consistent 3D connectivity framework for medical images Chitresh Bhushan, Ashish Saxena, Dattesh Dayanand Shanbhag GE HealthCare Technology and Innovation Center, Niskayuna, United States of America Impact: Our novel anatomy-aware connectivity framework improves accuracy of spine disk plane orientation estimates, as compared to cartesian pixel-connectivity, allowing robust workflow improvements and increasing sensitivity of automated tasks such as disease or abnormality detection.
14:08		352-02-010. Start Smart: reducing annotation effort in fetal MRI via provenance-aware active learning Rémi HATTAT, Marine Beaumont, Charline Bertholdt, Gabriela Hossu, Olivier Morel, Bailiang CHEN IADI U1254, INSERM and Université de Lorraine, Nancy, France Impact: This framework accelerates local deployment of fetal MRI segmentation models in resource-constrained clinical environments, reducing expert workload and supporting privacy-preserving multi-site collaboration for routine clinical workflows.
14:10		352-02-011. BrainSegNet: A Robust Framework for Whole-Brain MRI Segmentation Enhanced by Large Models Yucheng Li, Xiaofan Wang, Junyi Wang, Yijie Li, Xi Zhu, Mubai Du, Dian Sheng, Wei Zhang, Fan Zhang University of Electronic Science and Technology of China, Chengdu, China Impact: By coupling SAM with a U-Net encoder and decoder refinements (including Atrous Spatial Pyramid Pooling, Channel and Spatial Attention Module and boundary refinement), BrainSegNet achieves state-of-the-art whole-brain multi-label accuracy while improving boundary fidelity and scalability for practical, automatic neuroanatomy.
14:12		352-02-012. Accurate Segmentation of Placenta Accreta Spectrum: An Uncertainty-Guided Fusion Network for Multi-sequence MRI Le Fu, Haima Yang, Peicheng li, Jie Shi, Jianli Yu, Jiejun Cheng Shanghai first maternity and infant hospital, Shanghai, China Impact: This uncertainty-aware fusion framework enhances MRI-based PAS segmentation accuracy, providing a reliable foundation for preoperative evaluation and inspiring uncertainty-driven approaches in clinical multi-sequence imaging.
14:14		352-02-013. An Open Innovation Emitter–Modulator–Injector Framework for Inline MRI Reconstruction (PRIME) Omer Demirel, Spencer Waddle, Sandeep Ganji, Melvyn Ooi, Ryan Robison Philips North America Clinical Science, Rochester, United States of America Impact: PRIME enables researchers to flexibly integrate external algorithms directly into clinical MRI reconstruction workflows, bridging the gap between open-source innovation and restricted vendor environments.
14:16		352-02-014. AI-based Fully Automated Whole-Body MRI Analysis for Bone Metastasis in Prostate Cancer Akira Kudo, Katsuyuki Nakanishi, Yasuhiko Yamane, Takuya Yuzawa, Yoshiro Kitamura, Yuki Suzuki, noriyuki tomiyama, Masatoshi Hori Fujifilm Corporation, Minato, Japan Impact: This study proposes a novel fully automated deep learning pipeline that integrates T1WI and DWI for whole-body MRI bone metastasis analysis. Automated ADC colormap generation significantly reduces radiologist workload and improves reproducibility.
14:18		352-02-015. Deep learning spatial prediction of longitudinal WMH progression in ~ 1 year with pCASL, FLAIR, and MPRAGE Sang Hun Chung, Elizabeth Joe, Tianrui Zhao, Yining He, Vasilis Marmarelis, Helena Chui, Lirong Yan Northwestern University Feinberg School of Medicine, Chicago, United States of America Impact: We demonstrate the feasibility of predicting spatial evolution of white matter hyperintensity approximately one year after baseline scans with readily available tools. This work advances the state of WMH prediction and offers an accessible framework for future research.
14:20		352-02-016. Deep Learning for Detection of Intracranial Aneurysms on TOF-MRA: A Multi-Center Study with External Validation Yuxi Hou, Caixia Fu, Hang Pan, Nan Lan, Huiyang Li, tianxiao zhang, Zhongyi Yao, Yunpeng Wang, Yubo Ji, Jing Lu, Bing Tian The 1st affiliated hospital of Navy Medical University(Changhai hospital of Shanghai), Shanghai, China Impact: Based on multi-center TOF-MRA data, this AI model demonstrates robust aneurysm detection capability with 90% external test sensitivity. It shows potential as a clinical assistive tool to support radiologists in screening workflows, particularly for reducing oversight in routine interpretations.
14:22		352-02-017. Volumetric MRI Analysis of Dexamethasone Response in Peritumoral Edema in Patients with Glioma Adel Azghadi, Alexander Wang, Robin Ghotra, Sree Gongala, Deepthi Yammani, Michael Staudt, Tiffany Hodges, Tyler Miller, Chaitra Badve Case Western Reserve University and University Hospitals of Cleveland, Cleveland, Ohio, United States of America Impact: This study establishes MRI-based volumetric biomarkers linking corticosteroid to edema reduction, enabling clinicians to tailor dexamethasone dosing by cumulative exposure rather than daily intensity. Future studies can integrate immunophenotyping to define dosing that preserve antitumor immunity while optimizing edema control.
14:24		352-02-018. Automated organ segmentation for fetal body anomalies in 3D T2-weighted MRI: pilot study Charline Bradshaw, Anangsha Kumar, Alexia Egloff Collado, Megan Hall, Jacqueline Matthew, Vanessa Kyriakopoulou, Sara Neves Silva, Jordina Aviles Verdera, Aysha Luis, Tomas Woodgate, David Lloyd, Jo Hajnal, Jana Hutter, MARY RUTHERFORD, Lisa Story, Alena Uus King's College London, London, United Kingdom Impact: This work establishes a foundation for reproducible quantitative assessment of fetal body anatomy in normal and pathological cases, improving 3D visualisation, volumetric phenotyping, and multidisciplinary planning. It supports a shift from subjective interpretation toward standardised, data-driven assessment of fetal MRI.