Longitudinal Weak Labeling for Lung Cancer Prognosis and Treatment Response Prediction

This project aims at evaluating whether recent results in deep learning models, trained to exploit weak labels (Hwang, 2016) can serve to extract meaningful lesion localizations from image-level labels, either from individual scans or given a (longitudinal) sequence thereof. To this end, we will scale up existing models that have been shown to work on 2D images to a 3D context, studying labeling performance as the dataset size grows. If successful, this work will assert the usefulness of DCNNs to provide a general modeling framework to integrate imaging with other clinical patient data into a predictive system that could help support clinical decisions and ultimately improve patient care. The proposed research project fits within the partner's scientific roadmap, which is to develop deep learning models suitable to processing clinical data that arises in a sequential fashion at the patient level (longitudinal data), wherein the set of available clinical modalities can be highly variable (heteromodality). The industrial partner has an existing team of full-time researchers dedicated to studying these questions; the intern will attack complementary questions with the help of the team.

Michal Drozdzal
Faculty Supervisor: 
Yoshua Bengio
Project Year: