Multi-Task Deep Learning: Simultaneous Segmentation and Survival Analysis via Cox Proportional Hazards Regression

Abstract

Multi-task learning has taken an important place as a tool for medical image analysis, namely for the development of predictive models of disease. This study aims at developing a new deep learning model for simultaneous segmentation and survival regression, using a version of the Cox model to support the learning process. We use a combination of a 2D U-net and a residual network to minimize a combined loss function for segmentation and survival regression. To validate our method, we created a simple synthetic data set - the model segments circles of different sizes and regresses the area of circles. The main motivation of this work is to create a workflow for segmentation and regression for medical images application: in specific, we use this model to segment lesions or organs and regress clinical outcomes as overall or disease-free survival.

Tasks

Reproductions