Domain Adversarial RetinaNet as a Reference Algorithm for the MItosis DOmain Generalization Challenge

TL;DR

This paper introduces a domain adversarial RetinaNet algorithm for mitotic figure detection, addressing domain shift issues in histopathology, and provides a baseline for the MItoisis DOmain GEneralization challenge with publicly available code.

Contribution

The work presents a novel domain adversarial training approach for mitosis detection, specifically designed as a baseline for the domain generalization challenge.

Findings

Achieved an F1 score of 0.7183 on the challenge test set.

Developed a domain adversarial RetinaNet model for histopathology.

Made the model weights and code publicly available.

Abstract

Assessing the Mitotic Count has a known high degree of intra- and inter-rater variability. Computer-aided systems have proven to decrease this variability and reduce labeling time. These systems, however, are generally highly dependent on their training domain and show poor applicability to unseen domains. In histopathology, these domain shifts can result from various sources, including different slide scanning systems used to digitize histologic samples. The MItosis DOmain Generalization challenge focused on this specific domain shift for the task of mitotic figure detection. This work presents a mitotic figure detection algorithm developed as a baseline for the challenge, based on domain adversarial training. On the challenge's test set, the algorithm scored an F$_1$ score of 0.7183. The corresponding network weights and code for implementing the network are made publicly available.