Paul Caron, Julien Veron Vialard
In this project, we study the possibility to apply Convolutional Neural Networks for Computer Vision to tissue analysis for early stage cancer detection. It is inspired by the contest of PCam dataset. Using different CNN architectures, we create various binary classifiers and evaluate their quality with Receiver Operating Characteristic curves. Using deep residual neural networks and ensemble learning techniques, we aim at competing state-of-the-art classifiers for Histopathologic Cancer Detection. Finally, we analyze our results using specific metrics for medical purpose. The goal is to optimize our model for real-life applications and to consider model interpretation to help human intervention for model prediction confirmation.
This project was part of the course "CS231n: Convolutional Neural Networks for Visual Recognition" taught at Stanford University. http://cs231n.stanford.edu/. June 2020.