Skip to content

Code for the paper 'Benchmarking Deep Learning and Vision Foundation Models for Atypical vs. Normal Mitosis Classification with Cross-Dataset Evaluation'

License

Notifications You must be signed in to change notification settings

DeepMicroscopy/AMi-Br_Benchmark

Repository files navigation

Benchmarking Deep Learning and Vision Foundation Models for Atypical vs. Normal Mitosis Classification with Cross-Dataset Evaluation

arXiv

This repository contains the datasets and training and inference scripts for our paper:

Benchmarking Deep Learning and Vision Foundation Models for Atypical vs. Normal Mitosis Classification with Cross-Dataset Evaluation
Sweta Banerjee, Viktoria Weiss, Taryn A. Donovan, Rutger A. Fick, Thomas Conrad, Jonas Ammeling, Nils Porsche, Robert Klopfleisch, Christopher Kaltenecker, Katharina Breininger, Marc Aubreville, Christof A. Bertram
arXiv:2506.21444v1


Abstract:

Atypical mitosis marks a deviation in the cell division process that has been shown be an independent
prognostic marker for tumor malignancy. However, atypical mitosis classification remains challenging
due to low prevalence, at times subtle morphological differences from normal mitotic figures, low
inter-rater agreement among pathologists, and class imbalance in datasets. Building on the Atypical
Mitosis dataset for Breast Cancer (AMi-Br), this study presents a comprehensive benchmark comparing
deep learning approaches for automated atypical mitotic figure (AMF) classification, including
end-to-end trained deep learning models, foundation models with linear probing, and foundation models
fine-tuned with low-rank adaptation (LoRA). For rigorous evaluation, we further introduce two new
held-out AMF datasets - AtNorM-Br, a dataset of mitotic figures from the TCGA breast cancer cohort,
and AtNorM-MD, a multi-domain dataset of mitotic figures from a subset of the MIDOG++ training set.
We found average balanced accuracy values of up to 0.8135, 0.7788, and 0.7723 on the in-domain AMi-Br
and the out-of-domain AtNorm-Br and AtNorM-MD datasets, respectively. Our work shows that atypical
mitotic figure classification, while being a challenging problem, can be effectively addressed through
the use of recent advances in transfer learning and model fine-tuning techniques. We make all code and
data used in this paper available in this github repository.


Contents

  • AtNorM-Br/: Dataset (TCGA-Breast atypical and normal mitoses)
  • AtNorM-MD/: Dataset (Multi-domain dataset from MIDOG++ featuring atypical and normal mitoses)
  • End-to-End_DL_Baselines/: Fully trained baselines
  • Foundation_Models/: Linear probing and Low Rank Adaptation of foundation models

About

Code for the paper 'Benchmarking Deep Learning and Vision Foundation Models for Atypical vs. Normal Mitosis Classification with Cross-Dataset Evaluation'

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published