Publikationsdatenbank

Automated diagnosis of seven major skin tumors in canines using a convolutional neural network (CNN) on H&E-stained whole slide images (WSI) (2022)

Art

Hochschulschrift

Autor

Fragoso Garcia, Marco Antonio (WE 12)

Quelle

Berlin, 2022 — 86 Seiten

Sprache

Englisch

Verweise

URL (Volltext): https://refubium.fu-berlin.de/handle/fub188/37131

Kontakt

Institut für Tierpathologie

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

Abstract / Zusammenfassung

Microscopic examination of HE-stained slides is the gold standard for a variety of diseases. Specifically in oncology, it is critical not only for accurate diagnosis, but also for staging tumors and evaluating their limits. In recent decades, with the advent of Digital Pathology (DP) and Whole Slide Images (WSIs), image analysis and the development of algorithms to perform specific tasks on WSIs has been at the forefront of research in pathology, with overwhelming results. In this study, we describe a functional algorithm for automatic detection of seven major skin tumors in dogs: Trichoblastoma, squamous cell carcinoma (SCC), peripheral nerve sheath tumor (PNST), melanoma, histiocytoma, mast cell tumor (MCT), and plasmocytoma. We selected, digitized, and annotated 350 H&E-stained slides (70 per tumor type) to create a database divided into training (n=245 WSIs), validation (n=35 WSIs), and test (n=70 WSIs) data. A convolutional neural network (CNN) was then developed and the efficiency of the algorithm was tested on 140 new WSIs (20 per tumor type). The classification accuracy at the slide level reached 95% (133/140 WSIs), and the precision at the patch level was 85%. The same 140 WSIs were submitted to six certified pathologists for diagnosis, achieving similar slide-level accuracy of 98% (137/140 WSIs). Our results demonstrate that the use of artificial intelligence as a tool in diagnostic and research oncologic pathology is feasible and may be applied to other species and other tumor types in the future.