Fine-Grained Hard Negative Mining: Generalizing Mitosis Detection with a Fifth of the MIDOG 2022 Dataset

TL;DR

This paper presents a deep learning approach with hard negative mining and data augmentation to improve mitosis detection in histology images, achieving competitive results in the MIDOG 2022 Challenge.

Contribution

It introduces a rotation-invariant model trained with aggressive hard negative mining and selective training data to enhance generalization across variability in histology images.

Findings

Achieved an F1-score of 0.697 on the challenge test set.

Selected only 19.6% of training patches for optimal performance.

Secured the third place in the MIDOG 2022 Challenge.

Abstract

Making histopathology image classifiers robust to a wide range of real-world variability is a challenging task. Here, we describe a candidate deep learning solution for the Mitosis Domain Generalization Challenge 2022 (MIDOG) to address the problem of generalization for mitosis detection in images of hematoxylin-eosin-stained histology slides under high variability (scanner, tissue type and species variability). Our approach consists in training a rotation-invariant deep learning model using aggressive data augmentation with a training set enriched with hard negative examples and automatically selected negative examples from the unlabeled part of the challenge dataset. To optimize the performance of our models, we investigated a hard negative mining regime search procedure that lead us to train our best model using a subset of image patches representing 19.6% of our training partition of the challenge dataset. Our candidate model ensemble achieved a F1-score of .697 on the final test set after automated evaluation on the challenge platform, achieving the third best overall score in the MIDOG 2022 Challenge.