Parameter-efficient fine-tuning (PEFT) of Vision Foundation Models for Atypical Mitotic Figure Classification

TL;DR

Atypical mitotic figures (AMFs) are rare and morphologically subtle prognostic markers challenging to classify across domains. The study investigates parameter-efficient fine-tuning of large vision foundation models (UNI, Virchow, Virchow2) using Low-Rank Adaptation (LoRA) for AMF vs NMF classification within the MIDOG 2025 Track 2 dataset, including both random and domain-aware splits. Across extensive experiments with LoRA ranks, data splits, and ensembling, the best approach—Virchow with LoRA rank 8 and three-fold cross-validation—achieved a balanced accuracy of 88.44% on the final test, ranking 9th. The work demonstrates the promise of foundation models with efficient adaptation for domain-generalized pathology tasks, while underscoring ongoing needs to improve specificity and cross-domain generalization.

Abstract

Atypical mitotic figures (AMFs) are rare abnormal cell divisions associated with tumor aggressiveness and poor prognosis. Their detection remains a significant challenge due to subtle morphological cues, class imbalance, and inter-observer variability among pathologists. The MIDOG 2025 challenge introduced a dedicated track for atypical mitosis classification, enabling systematic evaluation of deep learning methods. In this study, we investigated the use of large vision foundation models, including Virchow, Virchow2, and UNI, with Low-Rank Adaptation (LoRA) for parameter-efficient fine-tuning. We conducted extensive experiments with different LoRA ranks, as well as random and group-based data splits, to analyze robustness under varied conditions. Our best approach, Virchow with LoRA rank 8 and ensemble of three-fold cross-validation, achieved a balanced accuracy of 88.37% on the preliminary test set, ranking joint 9th in the challenge leaderboard. These results highlight the promise of foundation models with efficient adaptation strategies for the classification of atypical mitosis, while underscoring the need for improvements in specificity and domain generalization.