Exploring Intrinsic Hierarchical Organization for Medical Diagnosis
Cao, Chengzhi Xu, Min
Cao, Chengzhi
Xu, Min
Author
Supervisor
Department
Computer Vision
Embargo End Date
Type
Conference proceeding
Date
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
We propose a structured knowledge-driven framework that explores intrinsic hierarchical organization to represent highorder dependencies within and across biological scales for interpretable diagnosis. Our approach is motivated by the inherent hierarchical structure of biological systems, where complex physiological functions arise from interactions across multiple organizational levels, including intra-level interactions and inter-level relationships. By integrating them into a unified hypergraph representation, we derive interpretable features that preserve structural dependencies across scales and offer a natural scaffold to integrate heterogeneous medical data in a unified, structured way. Each layer first models group-wise interactions through the hyperedges within the hypergraph, and then fuses information along the bidirectional path formed by the connections between different scales throughout the entire hierarchical structure. In this way, local and hierarchical information can interweave and complement each other. Probabilistic inference over the hypergraph space allows us to uncover latent hierarchical patterns that are predictive of disease states while remaining aligned with biological plausibility. Experimental results on medical report generation and medicalVQA tasks demonstrate improved diagnostic accuracy, suggesting that modeling biological data through hierarchy-based structures offers a principled path for medical diagnosis.
Citation
C. Cao, M. Xu, "Exploring Intrinsic Hierarchical Organization for Medical Diagnosis," 2026, pp. 1-5.
Source
Conference
2026 IEEE 23rd International Symposium on Biomedical Imaging (ISBI)
Keywords
46 Information and Computing Sciences, 4605 Data Management and Data Science
Subjects
Source
2026 IEEE 23rd International Symposium on Biomedical Imaging (ISBI)
Publisher
IEEE