Thank you for visiting nature. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser or turn off compatibility mode in Internet Explorer. In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript. An Author Correction to this article was published on 02 July The prognostic value of mitotic figures in tumor tissue is well-established for many tumor types and automating this task is of high research interest.
However, especially deep learning-based methods face performance deterioration in the presence of domain shifts, which may arise from different tumor types, slide preparation and digitization devices. We provide region of interest images from histological specimens of seven different tumor types with variable morphology with in total labels for 11, mitotic figures: breast carcinoma, lung carcinoma, lymphosarcoma, neuroendocrine tumor, cutaneous mast cell tumor, cutaneous melanoma, and sub cutaneous soft tissue sarcoma.
The specimens were processed in several laboratories utilizing diverse scanners. We evaluated the extent of the domain shift by using state-of-the-art approaches, observing notable differences in single-domain training. In a leave-one-domain-out setting, generalizability improved considerably. This mitotic figure dataset is the first that incorporates a wide domain shift based on different tumor types, laboratories, whole slide image scanners, and species.
Predicting the biological tumor behavior using histopathology is a central requirement for the identification of therapeutic options and the planning of tailored therapy. For this, micrometer-thin sections of tissue are produced from a formalin-fixed and paraffin-embedded tissue block and subsequently stained with histochemical dyes e. Several histological patterns are evaluated in a standardized manner and combined to tumor-type specific grading systems 1 , 2. The density of mitotic figures, i.
Usually, the number of mitotic figures in a standardized region of interest ROI , i. However, the identification of mitotic figures is subject to high intra- and inter-rater variability, resulting in low reproducibility of the MC 8 , 9 , Besides object-level differences, selection of the ROI with the assumed highest mitotic count in the entire histological section s , as requested by the guidelines 1 , 11 , 12 , is prone to significant inter-rater differences Consequently, the computerized identification of mitotic figures in digitized whole slide images WSIs is a relevant topic of ongoing scientific interest, after previous attempts with classical image analysis using special stains Especially since the advent of deep learning, automatized approaches have reached or even exceeded the performance of human experts and have shown a high potential to improve this prognostic task 8 , 10 , All of these challenge datasets used human breast carcinoma images and have since been complemented by two datasets covering two canine tumor types breast carcinoma and mast cell tumors 20 , 21 annotated on the complete WSI.
