# Efficient Fine-Tuning of DINOv3 Pretrained on Natural Images for Atypical Mitotic Figure Classification (MIDOG 2025 Task 2 Winner)

## TL;DR

This paper presents a highly effective fine-tuning approach for DINOv3, a vision transformer pretrained on natural images, to classify atypical mitotic figures in histopathology, achieving state-of-the-art results in the MIDOG 2025 challenge.

## Contribution

The authors introduce a low-rank adaptation fine-tuning method for DINOv3 that efficiently transfers to histopathology, handling domain heterogeneity with minimal parameter updates.

## Key findings

- Achieved first place in MIDOG 2025 Task 2 for AMF classification.
- Fine-tuning only ~1.3M parameters with LoRA is effective across domains.
- Demonstrated robustness and efficiency of the proposed fine-tuning strategy.

## Abstract

Atypical mitotic figures (AMFs) represent abnormal cell division associated with poor prognosis. Yet their detection remains difficult due to low prevalence, subtle morphology, and inter-observer variability. The MIDOG 2025 challenge introduces a benchmark for AMF classification across multiple domains. In this work, we fine-tuned the recently published DINOv3-H+ vision transformer, pretrained on natural images, using low-rank adaptation (LoRA), training only ~1.3M parameters in combination with extensive augmentation and a domain-weighted Focal Loss to handle domain heterogeneity. Despite the domain gap, our fine-tuned DINOv3 transfers effectively to histopathology, reaching first place on the final test set. These results highlight the advantages of DINOv3 pretraining and underline the efficiency and robustness of our fine-tuning strategy, yielding state-of-the-art results for the atypical mitosis classification challenge in MIDOG 2025.