Nowadays, machine learning techniques based on deep neural networks are everywhere, from image classification and recognition or language translation to autonomous driving or stock market prediction. One of the most prominent fields of application is medicine, where AI techniques promote and promise the personalized medicine. In this work, we entered in this field to study the prostate cancer prediction from images digitalized from hematoxylin and eosin stained biopsies. We chose this illness since prostate cancer is a very common type of cancer and the second cause of death in men. We did this work in collaboration with Hospital Reina Sofia of Murcia. As newcomers, we faced a lot of problems to start with, and questioned ourselves about many issues. This paper shows our experiences in developing and training two convolutional neural networks from scratch, exposing the importance of both the preprocessing steps (cropping raw images to tiles, labeling, and filtering), and the postprocessing steps (i.e., to obtain results understandable for doctors). Therefore, the paper describes lessons learned in building CNN models for prostate cancer detection from biopsy slides.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.