A Unified Riemannian-Geometric Framework for SARS-CoV-2 Detection from CT Scans
Keywords: SARS-CoV-2, Transfer learning, Medical image identification
TL;DR: We present a novel, theoretically grounded framework for automated SARS-CoV-2 detection from CT
Abstract: We present a novel, theoretically grounded framework for automated SARS-CoV-2 detection from pulmonary Computed Tomography (CT) scans, integrating cutting-edge concepts from statistical learning theory, optimal transport, and information geometry. Our approach begins with a submodular optimization-based image selection protocol, utilizing a continuous greedy algorithm. The feature extraction process employs a Riemannian geometry-inspired attention mechanism, where feature integration is formulated as geodesic interpolation on a manifold induced by the Fisher Information Metric. We introduce a unified decision-making framework based on proper scoring rules and Bregman divergences, encompassing multiple voting schemes with proven consistency and asymptotic normality properties. To address domain shift, we develop an adversarial domain adaptation technique using the Wasserstein-Fisher-Rao distance, complemented by a graph-based regularization term derived from Gromov-Wasserstein theory. Theoretical analysis provides convergence guarantees for the adversarial training process and establishes generalization bounds in terms of optimal transport distances. Empirical evaluation demonstrates the superiority of our approach over existing methods, achieving state-of-the-art performance on benchmark datasets. This work not only advances the field of automated medical image analysis but also contributes fundamental theoretical insights to the broader domains of machine learning and optimal transport theory.
Primary Area: transfer learning, meta learning, and lifelong learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 12958
Loading