DMDSC: A Dynamic-Margin Deep Simplex Classifier for Open-Set Recognition on Medical Image Datasets

2026-05-01

AI Summary

This work addresses the challenge of open-set recognition under extreme class imbalance in medical imaging by proposing the Dynamic Margin Deep Simplex Classifier (DMDSC). Grounded in neural collapse theory, DMDSC integrates a deep simplex structure with an uncertainty-aware mechanism and introduces a novel class-frequency-adaptive dynamic margin strategy that enforces larger margins for rare classes to encourage compact feature clustering. Evaluated on multiple medical benchmark datasets—including BloodMNIST, OCTMNIST, DermaMNIST, and BreaKHis—DMDSC significantly outperforms current state-of-the-art methods, achieving high accuracy on known classes while effectively rejecting unknown samples, thereby demonstrating its efficacy and robustness.

Abstract

Medical imaging datasets are often characterized by extreme class imbalances, where rare pathologies are significantly underrepresented compared to common conditions. This imbalance poses a dual challenge for Open-Set Recognition (OSR): models must maintain high classification accuracy on known classes while reliably rejecting unknown samples unseen during training in the clinical settings. While recently proposed Deep Simplex Classifier (DSC)~\cite{cevikalp2024reaching} and UnCertainty-aware Deep Simplex Classifier (UCDSC)~\cite{Aditya_2026_WACV} successfully leverage Neural Collapse to ensure maximal inter-class separation, they rely on a uniform margin that does not account for the varying densities of medical classes. In this paper, we propose DMDSC an enhanced framework featuring a dynamic margin approach. Our approach automatically adapts class-specific margins based on label frequency, enforcing a higher penalty and tighter feature clustering for rare pathologies to counteract the effects of data imbalance. Extensive experiments conducted on diverse medical benchmarks on BloodMNIST\cite{medmnistv2}, OCTMNIST\cite{medmnistv2}, DermaMNIST\cite{medmnistv2}, and BreaKHis~\cite{spanhol2015dataset} datasets, demonstrate that our framework outperforms state-of-the-art methods.

Problem

• class imbalance
• open-set recognition
• medical image classification
• rare pathologies
• unknown sample rejection

Innovation

• Dynamic Margin
• Open-Set Recognition
• Class Imbalance
• Deep Simplex Classifier
• Medical Image Classification