Mix, Align, Distil: Reliable Cross-Domain Atypical Mitosis Classification

2025-08-28

AI Summary

Cross-domain classification of atypical mitotic figures (AMFs) is severely hindered by domain shifts arising from variations in scanners, staining protocols, and image acquisition. To address this, we propose a robust cross-domain classification framework comprising three key components: (i) style-mixing augmentation to enhance feature diversity; (ii) coarse-grained domain label-guided attention to align domain-invariant features; and (iii) an Exponential Moving Average (EMA) teacher model coupled with temperature-scaled KL divergence for stable knowledge distillation. The method operates without fine-grained domain annotations and maintains lightweight inference. Evaluated on the MIDOG 2025 benchmark, it achieves a balanced accuracy of 0.8762, sensitivity of 0.8873, and ROC AUC of 0.9499—demonstrating strong generalization and well-balanced performance across domains.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Adversarial Attacks & Robustness
• Application Domains: Mobility, Driving & Flight

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Atypical mitotic figures (AMFs) are important histopathological markers yet remain challenging to identify consistently, particularly under domain shift stemming from scanner, stain, and acquisition differences. We present a simple training-time recipe for domain-robust AMF classification in MIDOG 2025 Task 2. The approach (i) increases feature diversity via style perturbations inserted at early and mid backbone stages, (ii) aligns attention-refined features across sites using weak domain labels (Scanner, Origin, Species, Tumor) through an auxiliary alignment loss, and (iii) stabilizes predictions by distilling from an exponential moving average (EMA) teacher with temperature-scaled KL divergence. On the organizer-run preliminary leaderboard for atypical mitosis classification, our submission attains balanced accuracy of 0.8762, sensitivity of 0.8873, specificity of 0.8651, and ROC AUC of 0.9499. The method incurs negligible inference-time overhead, relies only on coarse domain metadata, and delivers strong, balanced performance, positioning it as a competitive submission for the MIDOG 2025 challenge.

Problem

• Classifying atypical mitotic figures under domain shift
• Improving cross-domain robustness via style perturbations
• Aligning features across scanners and stains reliably

Innovation

• Style perturbations for feature diversity
• Attention alignment with auxiliary loss
• EMA teacher distillation for stability