MedXplore: Towards Reliable and Unbiased Generalized Category Discovery in Medical Imaging

2026-07-29

AI Summary

This study addresses the limitations of existing medical image analysis methods, which rely heavily on extensive annotations and operate under closed-world assumptions, thereby struggling to identify unknown pathologies encountered in clinical practice. To overcome this, the work introduces generalized category discovery to medical imaging for the first time and proposes the MedXplore framework. At the perceptual level, it designs a Frequency-domain Adaptive Attention Mechanism (FAAC) to enhance local anomalous signals and establish semantic anchors. At the decision level, it incorporates an Adaptive Cosine Angular Margin (ACAM) to balance intra-class compactness and inter-class separability. The approach substantially mitigates bias toward known classes, achieving an average accuracy improvement of 8.5% across multiple medical imaging benchmarks and reducing the misclassification rate of known classes from 14.50% to 0.80% on the Kvasir dataset, significantly enhancing model sensitivity and robustness to novel lesions.

Abstract

Deep learning has shown strong potential in medical image analysis, but most existing methods rely on large-scale annotations and a closed-world assumption that rarely holds in clinical practice. Although Generalized Category Discovery (GCD) has advanced rapidly on natural images, it remains underexplored in medical imaging. To address this issue, we propose MedXplore, a unified framework for reliable and unbiased medical GCD, optimizing from both perceptual and decision levels. Specifically, at the perceptual level, taking a frequency domain perspective, Frequency-SNR Adaptive Attention and Consistency (FAAC) performs learnable full-spectrum filtering and global-local energy contrast activation to not only highlight local abnormal signals relative to the global context, but also provide reliable semantic anchors for patch consistency learning. At the decision level, Adaptive Cosine-Angular Margin (ACAM) adjusts angular margins using semantic difficulty and feature confidence to balance intra-class compactness and inter-class separability. Together, the two modules improve lesion-sensitive representation learning and mitigate old-class bias. Experiments on multiple benchmarks show an average \textbf{8.5\%} gain in \textit{All} accuracy over the strongest competing methods. On Kvasir, MedXplore reduces false-old errors from 14.50\% to 0.80\%, demonstrating strong robustness under severe old-new ambiguity.

Problem

• Generalized Category Discovery
• Medical Imaging
• Unbiased Learning
• Old-New Ambiguity
• Closed-World Assumption

Innovation

• Generalized Category Discovery
• Frequency Domain Attention
• Adaptive Angular Margin
• Medical Image Analysis
• Unbiased Representation Learning