A statistical evaluation of the preprocessing medical images impact on a deep learning network’s performance

The aim of this paper is to explore the efficiency of preprocessing medical images before applying a deep learning algorithm to classify the data. The study uses a statistical framework that establishes the fact that depending on the dataset used, image preprocessing indeed decreases the computational time, without having a dropdown in performance. The dataset used in this study regard colon cancer, lung cancer, and fetal brain ultrasound scans. The study proposes a statistical performance that studies the performances of the ResNet50 deep learning network in different preprocessing scenarios.