Domain-Robust Mitotic Figure Detection with Style Transfer

TL;DR

This paper introduces a style transfer-based training scheme to improve the robustness of mitotic figure detection models across different scanner domains, enhancing generalization to unseen scanner styles.

Contribution

It presents a novel domain generalization approach using style transfer with StarGAN to make mitotic figure detection models more scanner-agnostic.

Findings

Model performs well on unseen scanner data

Style transfer increases domain robustness

Improves detection accuracy across diverse scanners

Abstract

We propose a new training scheme for domain generalization in mitotic figure detection. Mitotic figures show different characteristics for each scanner. We consider each scanner as a 'domain' and the image distribution specified for each domain as 'style'. The goal is to train our network to be robust on scanner types by using various 'style' images. To expand the style variance, we transfer a style of the training image into arbitrary styles, by defining a module based on StarGAN. Our model with the proposed training scheme shows positive performance on MIDOG Preliminary Test-Set containing scanners never seen before.