Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition

Segmentation for Neuro Applications

Digital Poster

Analysis Methods

Wednesday, 13 May 2026

Digital Posters Row E

16:55 - 17:50

Session Number: 564-06

No CME/CE Credit

Similar Sessions

This digital poster session features advanced work on segmentation in MRI, with a strong emphasis on neuroanatomical structures, brain pathologies, and methodological robustness. The abstracts cover a wide range of applications including deep brain nuclei, white-matter tracts, cerebellar and thalamic structures, tumors, stroke, vascular anatomy, and functional speech imaging. Methodological themes include multi-sequence learning, physics-informed modeling, contrast translation, failure detection, privacy-preserving preprocessing, and diffusion-based segmentation. These Abstracts highlight the diversity and maturity of contemporary segmentation research in brain and head-and-neck MRI, with a clear trajectory toward reliable clinical and intraoperative deployment.

Skill Level: Basic

		564-06-001. Multi-Sequence Learning Improves Robustness and Accuracy of Substantia Nigra Segmentation in Neuromelanin-Sensitive MRI Oliver Welsh, Kilian Hett, Anna Bosman, Daniel Claassen, Paula Trujillo University of Pretoria, Pretoria, South Africa Impact: Our multi-sequence training and custom preprocessing pipeline significantly improves segmentation robustness on unseen sequence types. This emphasises the importance of heterogeneous data for training deep-learning models and advancing the clinical scalability of NM-MRI-based neurodegenerative disease assessment.
		564-06-002. Fast and Reliable Failure Detection for Image Segmentation Using Pairwise Dice Similarity Tommaso Di Noto, Ian Cherabier, Lina Bacha, Punith Bidarakka Venkategowda, Keerthi Prabhu M, Silvia Pistocchi, Vincent Dunet, Attapon Jantarato, Manuela Vaneckova, Emmanuelle Le Bars, Jeremy Deverdun, Nicolas Menjot de Champfleur, Jonathan Disselhorst, Bénédicte Maréchal Siemens Healthineers International AG, Lausanne, Switzerland Impact: This work helps improve the safety and reliability of AI in medical image segmentation by detecting when results may be inaccurate, supporting better patient care and building trust in automated tools for healthcare professionals and patients.
		564-06-003. UltimateSynth Deep-brain Net (UDN): Deep Brain Segmentation of Any MRI Contrast and Age via Physics-Informed Deep Learning Rhea Adams, Khoi Huynh, Walter Zhao, Zhicheng Wu, Eunate Alzaga Goñi, Hengji Chen, Angela Noecker, Andreas Seas, Benjamin Succop Jr, Stephen Harward II, Cameron McIntyre, Pew-Thian Yap, Dan Ma Case Western Reserve University, Cleveland, United States of America Impact: UDN is the first tool to segment clinically relevant deep-brain structures in one minute across all MR contrast types, and is also the first to do so across the entire human lifespan, including both healthy and pathological data.
		564-06-004. nnU-Net for automatic segmentation of the parasagittal dura in autistic children using 3D T2-FLAIR Francesca Castellotti, Nivedita Agarwal, Ruth O'Gorman Tuura, Raimund Kottke, Chiara Girardi, Tommaso Ciceri Scientific Insitute IRCCS Eugenio Medea, Bosisio Parini (LC), Italy Impact: This study provides a tool for automatic segmentation and quantification of the parasagittal dura in children with autism spectrum disorder to advance the understanding of its role in neurodevelopmental and neuroinflammatory processes.
		564-06-005. OMT and tensor SVD based deep learning model for segmentation and predicting genetic markers of glioma: a multicenter study Zhengyang Zhu, Han Wang, Huiquan Yang, Yang Song, Mengying Xu, Xin Zhang, Bing Zhang Nanjing Drum Tower Hospital, Affiliated Hospital of Medical School, Nanjing University, Nanjing, China Impact: Accurate characterization of glioma is essential for effective clinical decision-making. Most current studies involve a limited number of patients. This research introduces a novel deep learning model based on OMT to predict molecular markers using international multicenter datasets.
		564-06-006. Cross-Domain Transfer Learning and Comparative Analysis of U-Net Architectures for Head and Neck Tumor Segmentation Khushi Singh, Sourav Basak, Sankar Misra, Subhanon Bera, Sourav Bhaduri St. Xavier’s College (Autonomous), Kolkata, India Impact: The study demonstrates that nn-U-Net-architecture can be reliable during the segmentation of head and neck tumor MRI data, and transfer learning may be used to generalize brain tumor segmentation to Head and neck tumor segmentation.
		564-06-007. MRI contrast translation for full-brain segmentation from T2-weighted contrasts Sebastian Rassmann, David Kügler, Martin Reuter German Center for Neurodegenerative Diseases (DZNE e.V.), Bonn, Germany Impact: We establish an image translation predicting high-quality T1w from T2w or FLAIR images. This enables accurate full-brain segmentation using FastSurfer without requiring T1w acquisition.
		564-06-008. Balancing Privacy and Utility in Brain MRI: Effects of Invasive vs. Geometry-Preserving Defacing Methods Yuli Wang, Yuwei Dai, Haoyue Guan, Cheng-Yi Li, Vora Maulik, Danica Cecil, Ayon Nandi, Jing Wu, Paul Zhang, Justin Honce, Chengzhang Zhu, Haris Sair, Leonora Balaj, Zhicheng Jiao, Yaou Liu, Cheng Tin Lin, Ihab Kamel, Li Yang, Harrison Bai Johns Hopkins University School of Medicine, Baltimore, United States of America Impact: Invasive defacing degrades segmentation, Evans-ratio reliability, and VLM reasoning, while geometry-preserving defacing maintains near-baseline performance. These results provide immediate guidance for repositories and labs to satisfy privacy mandates without sacrificing clinical or research utility.
		564-06-009. Segmentation of thalamic nuclei from Local Diffusion MRI Features: a comparison of clustering schemes Debottama Das, Ali Bilgin, manojkumar saranathan University of Arizona, Tucson, United States of America Impact: This study advances diffusion-based thalamic segmentation by systematically comparing MSMT-CSD and NODDI features across clustering algorithms, improving hemispheric consistency and establishing a reproducible framework that enhances the reliability and interpretability of thalamic parcellation for future neuroimaging analyses.
		564-06-010. 2D U-Net Segmentation of Multi-target Deep Brain Stimulation Lead Trajectories on Postoperative Imaging Eric Cito, Jacob Ellison, Anoushka Shah, Samantha Chan, Naomi Gong, Skyler Deutsch, Sarah Wang, Jill Ostrem, Janine Lupo, Melanie Morrison University of California San Francisco, San Francisco, United States of America Impact: No reliable open-source tools exist to segment deep brain stimulation lead trajectories. Leveraging deep learning to achieve fast yet reliable segmentation, our tool can be applied widely to inform surgical targeting strategies and to better understand variance in clinical outcomes.
		564-06-011. Intraoperative fast fibre tract segmentation in paediatric tumour patients Dana Kanel, Fiona Young, Kiran Seunarine, Nikhita Nandi, Annemarie Knill, Enrico De Vita, Kshitij Mankad, Chris Clark, Kristian Aquilina, Jonathan Clayden University College London, London, United Kingdom Impact: Tractfinder offers a clinical alternative to tractography for segmenting white matter tracts in tumour patients. It requires minimal processing time and expertise while accounting for tract displacement from space-occupying lesions, with potential to improve tract segmentation methods in clinical practice.
		564-06-012. Deep Learning-Based Segmentation of Cerebellar Peduncles Using Diffusion MRI Soumen Ghosh, Susmita Saha, Diogo Shiraishi, Thiago Rezende, Ian Harding, TRACK-FA Neuroimaging Consortium The University of Queensland, Brisbane, Australia Impact: This framework provides a robust automated segmentation of the cerebellar peduncles, enabling reproducible imaging biomarkers for disease monitoring and clinical trials. Its strong intra-cohort performance highlights its potential for future generalization across cerebellar degeneration datasets.
		564-06-013. Deep Learning for Automated Meningioma Segmentation: Toward Clinical Integration and Workflow Efficiency Laxmi Muralidharan, James Ruffle, Ebru Fenney, Anand Pandit, Hani Marcus, Parashkev Nachev, Harpreet Hyare University College London, London, United Kingdom Impact: A fully automated deep learning approach enables accurate, reproducible meningioma segmentation, outperforming existing methods. This technology offers substantial potential to standardise volumetric analysis and streamline radiological workflows, improving consistency, efficiency, and decision-making in clinical neuroimaging.
		564-06-014. Development and Validation of Automated Intracranial Vessel Segmentation Based on a Heterogeneous MR Angiography Dataset Pei-Hsuan Kuo, Wei-Chen Chen, Shih-Pin Chen, Chia-Hung Wu, Jiing-Feng Lirng, Shuu-Jiun Wang, Chia-Feng Lu National Yang Ming Chiao Tung University, Taipei, Taiwan Impact: This study developed an automated vessel segmentation model for intracranial arteries with satisfactory generalizability across different scanner types and cerebrovascular conditions. The proposed model could detect longitudinal changes in cerebrovascular diseases.
		564-06-015. High-Resolution 4D Speech MRI using Compressed Sensing and Deep Learning Segmentation Eric Schrauben, Roos Boekema, Lisette van der Molen, Ludwig Smeele, Engelbert AJM Schulten, Aart Nederveen Amsterdam University Medical Center, University of Amsterdam, Amsterdam, Netherlands Impact: Tongue function is critical for speech and swallowing, yet current 2D or low-resolution 3D imaging limits assessment. Our approach enables high-resolution visualization of vocal tract dynamics, with the future goal of identifying functional impairments and guiding targeted speech therapy strategies.
		564-06-016. Automated White Matter Lesion Detection Among People Living with HIV: A Preliminary Analysis Sophia Wang, Yannan Yu, David Saloner, Felicia Chow, Jared Narvid UCSF, San Francisco, United States of America Impact: Patients with HIV have a greater burden and severity of small vessel disease. In these populations, quantitative WMH assessment can provide a reliable representation for lesion severity, offering more granular data on clinical monitoring currently described with the Fazekas score.