Online Program Home
My Program

Abstract Details

Activity Number: 297 - SPEED: Food, Environment, Biomedical Imaging and Physical System Visualization/Learning, Part 1
Type: Contributed
Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
Sponsor: Section on Statistics in Imaging
Abstract #305063
Title: Survival Analysis for Medical Imaging Data
Author(s): Samantha Morrison* and Jon Steingrimsson and Constantine Gatsonis
Companies: Brown University and Brown University and Brown University
Keywords: Medical Imaging; Survival analysis; Neural Networks; Deep learning

Recently there has been large growth in the development of deep learning and other methodology to analyze medical images, such as CT and MRI scans. In addition to developments in localization and diagnosis of tumors in imaging scans, there has been work on using these medical scans in the prediction of time to event outcomes. However, with survival outcomes there are issues with bias due to right censoring, and methods to account for this bias with imaging datasets are not as widely studied. To predict survival past a time point for an imaging dataset of histologies of brain tumors, we implemented several neural networks with censoring unbiased loss functions, derived by Steingrimsson et al., 2018. We first used the pre-existing vgg16 neural network to extract features from the images and then trained neural networks with censoring unbiased loss functions (Buckley-James, Doubly Robust). The resulting predictions of these neural networks were compared to a Cox Proportional Hazards model using brier scores, and we found similar performance which motivates the future use of these methods for dealing with censoring bias in medical imaging analyses.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program