Shedding light on the black box of a neural network used to detect prostate cancer in whole slide images by occlusion-based explainability

Warning

This publication doesn't include Faculty of Arts. It includes Faculty of Informatics. Official publication website can be found on muni.cz.

Authors	GALLO Matej KRAJŇANSKÝ Vojtěch NENUTIL Rudolf HOLUB Petr BRÁZDIL Tomáš
Year of publication	2023
Type	Article in Periodical
Magazine / Source	NEW BIOTECHNOLOGY
MU Faculty or unit	Faculty of Informatics
Citation
web	https://www.sciencedirect.com/science/article/pii/S1871678423000511
Doi	https://doi.org/10.1016/j.nbt.2023.09.008
Keywords	Artificial intelligence; Digital histopathology; Explainable AI; Machine learning; Occlusion sensitivity analysis; Prostate cancer
Description	• Saliency maps identified histomorphological features characterizing cancer. • VGG16 model utilized all the structures that are observable by the pathologist. • The method can identify standard patterns not used by the model. • The method can also identify new patterns not yet used by human pathologists.
Related projects:	Rozsáhlé výpočetní systémy: modely, aplikace a verifikace VII. e-Infrastruktura CZ Aplikovaný výzkum v oblastech vyhledávání, analýz a vizualizací rozsáhlých dat, zpracování přirozeného jazyka a aplikované umělé inteligence Development of data processing techniques to support search, analysis and visualization of large datasets using artificial intelligence