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

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