Publikationsdatenbank

Canine mast cell tumours: prediction of the c-Kit exon 11 genotype by phenotype -
human observer vs artificial intelligence (2025)

Art

Poster

Autoren

Puget, C. (WE 12)

Ganz, J.

Kiupel, M.

Breininger, K.

Aubreville, M.

Klopfleisch, R. (WE 12)

Kongress

5th Cutting Edge Pathology Congress 2024

San Lorenzo de El Escorial, 28. – 31.08.2024

Quelle

Journal of comparative pathology

Bandzählung: 220

Seiten: 83

ISSN: 0021-9975

Sprache

Englisch

Verweise

URL (Volltext): https://www.sciencedirect.com/science/article/pii/S0021997525001264?via%3Dihub

DOI: 10.1016/j.jcpa.2025.03.095

Kontakt

Institut für Tierpathologie

Robert-von-Ostertag-Str. 15
14163 Berlin
+49 30 838 62450
pathologie@vetmed.fu-berlin.de

Abstract / Zusammenfassung

Background: Mast cell tumours (MCTs) are frequent neoplasms of dogs, with variable biological behaviour. Internal tandem duplication mutations in c-Kit exon 11 (c-Kit-11-ITD) are associated with poor prognosis but predict therapeutic response to tyrosine kinase inhibitors. In a previous work, a deep learning algorithm managed to predict the presence of c-Kit-11-ITD on digitalized HE-stained histological slides (whole slide images, WSIs) in up to 87% of the cases, suggesting the existence of morphological features characterizing MCTs carrying this mutation.

Materials & Methods: In an ongoing three-stage blinded study, three untrained pathologists were first asked to classify eight WSIs and 200 patches of MCTs as c-Kit-11-ITD positive or negative. Second, they were trained to recognize c-Kit-11-ITD by having access to a set of WSIs with known mutational status and 200 patches of areas highly relevant for algorithmic c-Kit-11-ITD classification. Third, pathologists were asked to classify 16 new WSIs and 200 new patches for c-Kit-11-ITD status. Participants had to report the microscopic features they identified as relevant for their decision.

Results: The participants correctly classified the c-Kit-11-ITD status of 63–88% of the WSIs and 43–55% of the patches without training, but only 25–38% of WSIs and 55–56% of patches after training. Nuclear pleomorphism was commonly named as a potential feature of c-Kit-11-ITD-positive MCTs.

Conclusion: Based on the current results it is assumed that transfer of algorithmic skills to the human observer is difficult. A c-Kit-11-ITD specific morphological feature usable for human observers remains thus to be extracted from the AI model.