AI: Prof. Thomas Walter’s team shines at the MIDOG 2025 mitosis detection challenge

Held in September during the MICCAI 2025 conference, a leading international event in medical image analysis, the MItosis DOmain Generalization (MIDOG) Challenge aims to advance the automatic detection of mitotic figures, key indicators of tumor aggressiveness.

Following two previous editions in 2021 and 2022, the 2025 challenge included two tasks:

• Detection of mitoses across data from diverse scanners, staining protocols and tissue types ;
• Classification of normal and atypical mitoses, the latter being less common but highly relevant for diagnosis and prognosis.

One of the main challenges lay in the diversity of the data. Images came from many laboratories, using different equipment and protocols, which made training robust machine-learning models particularly demanding.

Under the leadership of Prof. Thomas Walter¹, the team stood out among 150 participants from around the world, achieving outstanding results with a second place in mitosis detection and a first place in mitosis classification.

The results were made possible by state-of-the-art AI approaches. The team developed models able to adapt to the diversity of scanners and protocols, a key requirement for making these tools reliable in clinical practice.

This work was carried out by Raphaël Bourgade, a PhD student in Prof. Thomas Walter’s team at Institut Curie and co-supervised by Prof. Anne Vincent-Salomon², who led the mitosis detection component, and by Guillaume Balezo, a PhD student at Mines Paris – PSL, in charge of the classification task.

“We are very proud of these results! They are the outcome of true teamwork and reflect the dynamism and potential of research in AI applied to oncology at Institut Curie and Mines Paris – PSL,” says Prof. Thomas Walter.
 

[1] Professor at Mines Paris – PSL, head of the Statistical Machine Learning and Modelling of Biological Systems team (Inserm U1331) and deputy director of the Computational Oncology unit (Inserm U1331).

[2] Pathologist at Institut Curie and Director of the Institute of Women’s Cancers.