ISMRM25 - Applications of Foundation Models

ISMRM & ISMRT Annual Meeting & Exhibition • 10-15 May 2025 • Honolulu, Hawai'i

Applications of Foundation Models

Digital Poster

AI & Machine Learning

Wednesday, 14 May 2025

Exhibition Hall

16:45 - 17:45

Session Number: D-36

No CME/CE Credit

		Computer Number: 33 4003. Localization of Non-Isocentric Wrist Anatomy with Few-shot Foundation Models View Presentation Video G. R. Madhumani, D. Anand, D. Shanbhag GE HealthCare, Bangalore, India Impact: With as few as 6 labeled images, localization adapter trained with self-supervised visual features performed significantly better than CNN based few-shot model that used as many as 28 image-mask pairs. Demonstrated 3D localization capability from models trained on 2D images.
		Computer Number: 34 4004. Patient Selection for Active Surveillance for Prostate Cancer based on Deep Features from U-Found: An MRI-based Foundation Model of the Prostate View Presentation Video A. Breto, N. Lowry, V. Wallaengen, A. Algohary, S. Punnen, R. Stoyanova University of Miami, Miami, United States Impact: We present an application of novel MRI-based foundation model of the prostate to assess prostate cancer progression risk for AS patients, thereby providing essential data for clinicians that can prospectively improve AS patient selection.
		Computer Number: 35 4005. Few-Shot Contrastive Multilabel Localization of Knee MR Images with Self-Supervised Foundation Model View Presentation Video G. R. Madhumani, D. Anand, D. Shanbhag GE HealthCare, Bangalore, India Impact: Multilabel contrastive model trained by features extracted from FM with few labeled examples showed promising results for localizing multiple RoI's on knee images. Multilabel model showed excellent localization across knee slices by preventing false positives.
		Computer Number: 36 4006. Self-supervised Learning Network on Large Prostate Cancer mpMRI Dataset: Towards A Foundational Model of the Prostate View Presentation Video N. Lowry, A. Breto, V. Wallaengen, A. Algohary, R. Stoyanova University of Miami, Miami, United States Impact: To the best of our knowledge, U-Found is the first foundation-like model developed for prostate mpMRI. The embeddings, combining cancer and overall prostate characteristics features can be used in comprehensive modeling of cancer progression or response to therapy.
		Computer Number: 37 4007. TGD-BO: Task-specific Guidance Design with Bayesian Optimization using unconditional diffusion models for image restoration problems View Presentation Video N. Fujita, Y. Terada University of Tsukuba, Tsukuba, Japan Impact: We proposed a guidance design method adaptable to any image restoration problems and verified its effectiveness. This versatile framework enables diffusion model to handle multiple MRI restoration tasks without task-specific training, potentially serving as a foundation model.
		Computer Number: 38 4008. Using the SAM2 Foundation Model for Zero-shot Delineation of MRI-detected Brain Metastases for Stereotactic Radiosurgery View Presentation Video C-W Chang, R. Qiu, H-K Shu, S. Kahn, L. Sudmeier, J. Fair, M. Giles, H. Mao, Z. Tian, X. Yang Emory University, Atlanta, United States Impact: This brain metastases delineation tool has been shown to enhance the efficiency of treatment planning, potentially improving the effectiveness of stereotactic radiosurgery for brain metastases. Such advancements could lead to better clinical efficiency and patient outcomes.
		Computer Number: 39 4009. Evaluation of effectiveness and fairness of foundation models in multi-organ segmentation View Presentation Video Q. Li, Y. Zhang, Y. Li, Y. Zhang, L. Sun, M. Sun, Q. Li, Z. Wang, M. Liu, X. Hu, S. Wang, C. Wang Human Phenome Institute, Fudan University, Shanghai, China Impact: This study systematically evaluates the variations in segmentation effectiveness of foundation models across different organs and the fairness issues, which finds the shortcomings of the current foundation models and plays an important role in guiding future improvements of foundation models.
		Computer Number: 40 4010. Data-Efficient Lung Segmentation Using Foundational Models: A Comparative Study of SAM and CNNs for Hyperpolarized Gas MRI View Presentation Video R. Babaeipour, M. Fox, G. Parraga, A. Ouriadov Western University, London, Canada Impact: This study demonstrates the potential of foundational models like Segment Anything Model (SAM) to significantly improve lung segmentation accuracy using less data, opening new avenues for medical imaging in clinical settings where acquiring large, annotated datasets is challenging.
		Computer Number: 41 4011. Optimizing Cardiac MR Image Segmentation: Fine-Tuning the Foundational Segment Anything Model (SAM) View Presentation Video T. Geroski, A. Amini University of Louisville, Louisville, United States Impact: Fine-tuning of the foundational model SAM for cardiac MRI offers enhanced segmentation accuracy, enabling precise assessments of heart structure and function. This advancement supports improved diagnostic workflows and potential early detection in conditions like heart failure and cardiomyopathy.
		Computer Number: 42 4012. Automated Segmentation of the Small Bowel and Body Composition on MR Enterography: initial experience View Presentation Video F. Restrepo, M. Yuce, K. Yasokawa, D. Feldman, A. Geahchan, P. Brachmann, A. Hashmi, C. Lippert, B. Taouli Icahn School of Medicine at Mount Sinai, New York, United States Impact: These results lay a foundation for the application of DL models to CD patients with small bowel involvement, both in automated detection of affected bowel segments and in accurate determination of body composition, potentially providing outcome information in these patients.
		Computer Number: 43 4013. Advancing Parotid Tumor Diagnosis with KAN Transformer: Precise Differentiation of Pleomorphic Adenoma and Warthin Tumor: A Multicenter Study W. Mai, L. Zhang, D. Zhang, J. Zhong, J. Tan, Y. Chen, W. Liu, X. Liu, X. Hua, C. Shi the First Affiliated Hospital of Jinan University, Guangzhou, China Impact: The KAT model provides precise early diagnosis of benign parotid tumors, even with limited MRI data. These findings could extend to other tumor types, improving diagnostic accuracy and supporting individualized treatment.
		Computer Number: 44 4014. Self-supervised Pretraining on OAI data for 3D Knee MRI Analysis View Presentation Video X. Wang, L. Hazan, M. Yang, S. Rabinovici-Cohen, X. Li Cleveland Clinic, Cleveland, United States Impact: Our study has leveraged the OAI database and demonstrated the effectiveness of self-supervised pretraining for 3D knee MRI. Our approach enhances downstream task performance, inspiring further study on advancing automated 3D medical imaging analysis without labeled data.
		Computer Number: 45 4015. Development of an Automated Deep Learning Diagnostic Platform for Medical Imaging View Presentation Video J. Zhang, Y. Song MR Research Collaboration Team, Siemens Healthineers Ltd., shanghai, China Impact: This platform enhances clinical accessibility to deep learning diagnostic tools, supporting high-precision diagnostics and interpretability through an intuitive interface, and reducing the technical barrier to AI in medical imaging.
		Computer Number: 46 4016. Multi-vendor physical intelligent network for accurate quantification of magnetic resonance spectroscopy metabolites View Presentation Video Z. Tu, J. Zhang, Y-H Chu, L. Lin, X. Jiang, J. Wang, Q. Xu, D. Guo, X. Qu Xiamen University, Xiamen, China Impact: The proposed method utilizes a differentiable least squares network layer to achieve precise quantification of low-concentration metabolites, providing more accurate quantification metrics for disease diagnosis.
		Computer Number: 4017. WITHDRAWN
		Computer Number: 47 4018. Towards Standardized BNST Segmentation: A Deep Learning Model for Precise and Reproducible 3D T1 MRI Analysis View Presentation Video O. Ortiz, R. Loke, J. Kramer University of British Columbia, Vancouver, Canada Impact: This model enables automatic, precise BNST segmentation, advancing research on stress-related brain pathways, supporting personalized treatments, and enhancing diagnostic accuracy in psychiatric conditions. Automated BNST analysis facilitates early detection, improves patient outcomes, and accelerates large-scale research possibilities.

Similar Session(s)