Robert-von-Ostertag-Str. 15
14163 Berlin
+49 30 838 62450
pathologie@vetmed.fu-berlin.de
Introduction: Cutaneous mast cell tumours (MCTs) account for 7–21% of skin tumours in dogs. Numerous prognostic factors are assessed through the histopathological examination of biopsy samples. PCR analysis of the c-Kit exon 11 mutational status is often performed to evaluate the prospect of success of a tyrosine kinase inhibitor therapy. This project aimed to train a deep learning algorithm (DLA) to identify the c-Kit-11 mutational status of MCTs solely based on histological morphology.
Materials and methods: HE-stained slides of 196 c-Kit-11 mutated and 189 non-mutated cutaneous and subcutaneous MCTs were scanned with an Aperio scanner and used as a training database. The sample was then scanned with a 3DHistech Pannoramic scanner to assess the DLA performance with a domain shift.
Results: The DLA correctly classified the HE-stained slides after their c-Kit 11 mutational status in 83% of the cases. Chi2 tests excluded an association between the DLA classification and the tumour grades (Patnaik and Kiupel) as well as the location of the MCTs in the skin. A strong association existed between c-Kit-11 mutated classification and ulcerated epidermis. The DLA reached a classification accuracy of 0.80 on the 3DHistech database.
Conclusions: DLA assisted morphological examination of MCTs can rapidly predict the c-Kit-11 mutational status of MCTs with good precision, potentially saving the time and costs of a PCR analysis. Increasing the number of images as well as including scans originating from different scanners in the training dataset might improve the classification accuracy and robustness of this DLA.