Uncertainty-aware fusion of foundation and task-specific models for cardiac MRI segmentation

Vision foundation models, such as the Segment Anything Model (SAM), demonstrate strong zero-shot generalization but lack precision with anatomically challenging structures. In contrast, convolutional neural network (CNN)-based models achieve high accuracy on domain-specific data but struggle to generalize to unseen data. To address these complementary limitations, we propose an uncertainty-aware fusion framework that integrates the generalizability of foundation models with the anatomical precision of task-specific models for cardiac MRI segmentation. The approach combines Dempster-Shafer Theory (DST) with an entropy-guided fallback mechanism to perform voxel-wise fusion of calibrated probability maps. DST fusion is applied in regions of agreement, while high-conflict regions are handled by selecting predictions from the model with lower uncertainty. Experiments on in-domain and cross-domain datasets show consistent improvements, with larger gains under domain shift. To the best of our knowledge, this is the first voxel-wise uncertainty-based DST fusion of foundation and task-specific models for cardiac MRI segmentation.

Name thesis:degree_name

Master of Science (MSc)

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Ontario Institute of Technology

Year dc:date.issued

2026

Author dc:creator

• Rumman, Mosarrat

Advisors dc:contributor.advisor

• Ebrahimi, Mehran
• Davoudi, Kourosh

Language dc:language.iso

en

Handle dc:identifier.uri

https://hdl.handle.net/10155/2095

OAI identifier oai:identifier

oai:ontariotechu.scholaris.ca:10155/2095