Cape Town - 2026 ISMRM-ISMRT Annual Meeting and Exhibition

Segmentation for Body Applications

Digital Poster

Analysis Methods

Wednesday, 13 May 2026

Digital Posters Row F

16:00 - 16:55

Session Number: 565-05

No CME/CE Credit

Similar Sessions

Automated and deep-learning based approaches for segmentation in body applications

Skill Level: Intermediate

		565-05-001. Are MRI-based deep learning algorithms for kidney volume assessment in polycystic kidney disease ready for deployment? Enrique Almar-Munoz, Emil Colliander, Sebastian Tupper, Agnes Mayr, Rebeca Miron Mombiela Medical University of Innsbruck, Innsbruck, Austria Impact: This study establishes deep learning kidney segmentation as highly accurate yet not clinically implemented, guiding future validation and regulatory efforts. It enables standardized, efficient TKV assessment in ADPKD, advancing imaging biomarkers and accelerating AI-integration into nephrology research and patient management.
		565-05-002. Comparative Study of Manual Versus Automated MRI-Based Tumor Segmentation and Staging in Rectal Cancer Yen-Chun Chen, Chi-Feng Hsieh, Chia-Ching Chang, Chun-Jung Juan, Yi-Jui Liu, Hsu-Hsia Peng National Tsing Hua University, Hsinchu, Taiwan Impact: We constructed a Swin UNet model for automatically segmenting RC tumors. The automated ROI overestimated the tumor area, enabled the inclusion of peritumoral information, and resulted in higher T-staging performance than manual ROI.
		565-05-003. Automated Prostate Segmentation with AI: Bridging the Gap Between MRI and TRUS Volumetry Nader Gharbia, Yasmine Saad, Kays Cheker, Aymen Kammoun, Yassine Jomli, Wasim Frikha, Yassine Nouira Faculty of medicine of Sfax, Tunisia Impact: AI-based prostate segmentation provides precise and reproducible volume estimation demonstrating superior agreement with manual planimetry, enhancing PSA density accuracy and standardizing prostate cancer assessment across MRI and TRUS modalities for improved diagnostic confidence and clinical decision-making.
		565-05-004. nnU-Net-based Deep Learning for Automated Segmentation and Detection of Non-Mass Enhancement Lesions in Breast MRI Yijiang Huang, Wenjie Xu, Neng Wang, Sikai Wu, Yongyu AN, Cui Zhang, Lei Lv, Zhede Zhao, Zhiwen Yang, Yimin Huang, Changyu Zhou, Yunzhu Wu, Guoqun Mao Tongde Hospital of Zhejiang Province, Hangzhou, China Impact: The developed nnU-Net model provides reliable automated segmentation of NME lesions, assisting clinicians in accurate diagnosis and reducing workload. It encourages future multi-center studies and enhances standardization in breast MRI, benefiting patients through improved screening outcomes.
		565-05-005. Robust Breast Tumor Segmentation Across DCE and DWI MRI with nnU-Net: A Multi-Center Study Xiaobin Jiang, Darong Zhu, Qian Yu, Yantao Yang, Nianshi Song, Sicong Huang, Shiwei Wang The First Affiliated Hospital of Zhejiang Chinese Medical University, Hangzhou, China Impact: Automated nnU-Net–based segmentation provides robust, cross-center breast MRI tumor delineation, supporting efficient and reproducible analysis for clinical and research applications.
		565-05-006. Improving automatic segmentation of 7T MRI ex-vivo human traumatic spinal cord injury: A deep learning approach Kyle Vavasour, Nikolai Lesack, Sarah Morris, Andrew Yung, Kirsten Bale, Andrew Bauman, Piotr Kozlowski, Zahra Samadi-Bahrami, Caron Fournier, Pushwant Mattu, Kevin Dong, Femke Streijger, G. R. Wayne Moore, Adam Velenosi, Veronica Hirsch-Reinshagen, Brian Kwon, Cornelia Laule University of British Columbia, Vancouver, Canada Impact: Accurate, automated segmentation of ex vivo traumatic spinal cord injury MRI enables high-throughput, reproducible MRI–histology analyses. Our model improves accuracy over the standard tool, enabling larger, more reproducible studies and faster validation of potentially clinically relevant MRI biomarkers.
		565-05-007. In and out-of-distribution deep learning models for mesorectum and rectal cancer automatic segmentation Simone Perra, Filippo Crimì, Valentina Visani, Niccolò Sion, Matteo Prezioso, Francesco Celotto, Claudio Coco, Giuditta Chiloiro, Marco Scarpa, Emilio Quaia, Mauro Mattace Raso, Annalisa Marteddu, Daniela Rega, Simona Deidda, Gaya Spolverato, Marco Castellaro University of Padova, Padova, Italy Impact: An ensemble of complementary AI models enhances segmentation of rectal cancer and mesorectum on MRI across scanner types. This work supports development of robust, generalizable, and standardized imaging tools to improve clinical decision-making and outcome assessment in rectal cancer management.
		565-05-008. Adaptation of an Open-Source Deep Learning Segmentation Framework for Efficient Cyst Volume Quantification in ADPKD Joseph Rancitelli, Neeraja Mahalingam, Rima Kang, Caitlin Hackett, Christopher Crabtree, Ashwini Chebbi, Jeff Volek, Orlando Simonetti The Ohio State University Wexner Medical Center, Columbus, United States of America Impact: A probabilistic soft-edge approach enables accurate, efficient measurement of total cyst volume in ADPKD. This adapted workflow reduces computation and manual effort, supporting large-scale MRI studies and consistent disease monitoring across varying stages of disease progression.
		565-05-009. A Multi-Class CNN-Based Segmentation of Spinal Structures to Facilitate the Workflow of Spine MR Examinations Axel Saalbach, Carole Lazarus, Martin Bergtholdt, Xinyu Wang, Suthambhara Nagaraj, Tejas Shah, Viswanath Pamulakanty Sudarshan, Julien Senegas Philips GmbH Innovative Technologies, Hamburg, Germany Impact: The presented work shows the feasibility of obtaining all segmentation results that are relevant for automation of spine MR examination as the results of the single, multi-class segmentation model running in less than 20 sec on a standard console PC.
		565-05-010. Methods and Reliability of using nnU-Net to Automatically Segment the Large Bowel Abi Spicer, Stephen Lloyd-Brown, Neele Dellschaft, Caroline Hoad, Luca Marciani, Robin Spiller, Penny Gowland University of Nottingham, Nottingham, United Kingdom Impact: Assessment of large bowel volume is valuable, though time-consuming, in GI MRI. Here, a model is shown to have excellent agreement with ground truth masks with little need for user correction to gain accurate total colon volumes.
		565-05-011. Segmental hepatic volumes studied in patients with suspected liver iron overload Arthur Wunderlich, Holger Cario, Backhus Johanna, Emrullah Birgin, Michael Götz, Meinrad Beer, Stefan Schmidt Ulm University, Medical Center, Ulm, Germany Impact: Significant differences in relative segmental volume (rSV) between groups split by total hepatic volume were found in segment 3 and 8. The concept of rSV revealed significant differences in disease groups which were not found when analyzing absolute volumes.
		565-05-012. Deep Prompt Initialization and Fine Tuning of SAM2 for Automatic 2D Lung MR Image Segmentation Muduo Xu, Haoyang Pei, Hersh Chandarana, Yao Wang, Li Feng NYU Tandon School of Engineering, New York, United States of America Impact: This fully automatic method provides accurate lung segmentation and consistent functional measurements (e.g., lung-area curves), greatly reducing manual annotation effort. The proposed zero-click initialization and decoder-level SAM2 adaptation notably improve temporal accuracy over conventional prompting or CNN approaches.
		565-05-013. 3D nnU-Net-Based Automated Segmentation of Abdominal Adipose Tissue in Children using Free-Breathing Dixon MRI Wenwen Zhang, Sevgi Gokce Kafali, Shu-Fu Shih, Timothy Adamos, Kelsey Kuwahara, Ashley Dong, Jessica li, Timoteo Delgado, Shahnaz Ghahremani, Kara Calkins, Holden Wu David Geffen School of Medicine at UCLA, Los Angeles, United States of America Impact: The proposed 3D neural network achieved accurate automated segmentation of abdominal adipose tissue in children over a wide age range (6-18 years old) using free-breathing Dixon MRI. This technology provides a scalable tool for investigating cardiometabolic risk factors in children.