MedCORE: Criteria-Grounded Clinical Reasoning for Interpretable Medical Image Diagnosis

2026-10-06

AI Summary

This study addresses the lack of clinical reasoning structure and low transparency in existing medical image diagnostic models by proposing a novel criteria-based vision-language reasoning architecture. The method decomposes images into clinical criteria, localizes evidential regions, and explicitly models inter-criteria dependencies through multi-scale encoding and graph attention networks. Furthermore, it achieves weighted aggregation via prototype alignment and uncertainty calibration. Evaluated on the ISIC, BUSI, and IDRID datasets, the proposed framework attains up to 96.1% accuracy, achieving state-of-the-art performance and significantly outperforming baseline models. Ultimately, this work delivers highly accurate and interpretable clinical decision support for medical image diagnosis.

Abstract

Clinical diagnosis is inherently a structured reasoning process, yet existing deep learning models often bypass this structure by mapping image features directly to disease labels without explicitly interrogating the morphological and textural criteria that clinicians systematically evaluate. This limits diagnostic transparency and may compromise safe clinical deployment. We present MedCORE (Medical Criteria-Oriented Reasoning and Evidence), a structured diagnostic framework that operationalizes clinical reasoning within a vision-language architecture. For each input image, MedCORE decomposes the diagnostic process into clinically defined criteria, spatially localizes each criterion to diagnostically relevant image regions, encodes evidence through multi-scale representations that capture macro-structural and micro-textural pathological characteristics, and refines criterion representations using a Graph Attention Network that explicitly models inter-criteria dependencies. Criterion representations are further aligned with clinical text descriptors, reinforced through class-wise visual prototypes, and aggregated using uncertainty-calibrated weighting that proportionally discounts low-confidence diagnostic evidence. MedCORE is validated across three clinically heterogeneous imaging modalities, including dermoscopic lesion classification on ISIC 2018, breast ultrasound lesion characterization on BUSI, and diabetic retinopathy grading on IDRiD. Quantitatively, MedCORE achieves 89.2% accuracy, 85.7% macro-F1, and 96.4% AUC on ISIC 2018; 96.1% accuracy, 95.2% macro-F1, and 98.4% AUC on BUSI; and 84.3% accuracy, 80.2% macro-F1, and 92.8% AUC on IDRiD. These results demonstrate consistent improvements over strong CNN, transformer, biomedical vision-language, concept-based, and prototype-based baselines.

Problem

• Medical Image Diagnosis
• Clinical Reasoning
• Interpretability
• Diagnostic Transparency

Innovation

• Clinical Reasoning
• Vision-Language Architecture
• Graph Attention Network
• Uncertainty Calibration
• Interpretable Diagnosis