Robust and Accurate Multi-Tumor, Multi-Species, Multi-Labora

Neoplasms are one of the most common causes of death in humans and animals. The decision for appropriate therapy is based in part on the histological examination of tumor samples, capturing various prognostic parameters. One of the most relevant histological parameters for assessing the prognosis of tumor patients is the number of mitotic figures (Mitotic Count) in histological tumor section. In recent years, computer-assisted measurement methods using artificial intelligence have gained significant interest as they are capable of improving the reproducibility and accuracy of the measurements of this prognostic test. However, AI algorithms are highly dependent on the data with which they were trained. Current datasets do not contain the necessary variability (in terms of tumor types, scanners, stains and tissue quality), so a broad application of these algorithms in diagnostic laboratories is currently not possible. Differences in image properties (known as domain shift) between laboratories create a significant drop in the performance of AI algorithms and thus a reliable cross-laboratory application is not possible. This research project has the primary goal of creating a large dataset for mitotic figures in histological tumor preparations, which will include a large number of different domains, i.e., tumor types, species (humans and animals), and laboratories. Innovative methods of database generation are used, which enable maximum quality of labels and an efficient workflow to incorporate as many samples as possible into the database. These data will enable us to develop and validate an algorithm that can be used in the diagnostic workflow of numerous laboratories and thus improve the treatment decision of tumor patients. The developed dataset will be made publicly available, allowing further use by researchers and diagnostic laboratories. The resulting data will also serve as the basis for a learning platform for pathologists, which will provide practical exercise opportunities for proper mitotic figure recognition and the use of image analysis algorithms as decision support in the histological assessment of tumor samples. Finally, we will extensively validate the developed algorithms and investigate the benefits and potential risks for prognostic computer-assisted decision- making.