A Deep Learning Model for Improved Cancer Risk Prediction in Sequential Lung Screening X-Rays

Principal Investigator

Name
Regina Barzilay

Degrees
Ph.D

Institution
Massachusetts Institute of Technology

Position Title
Delta Electronics Professor, EECS

Email
regina@csail.mit.edu

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
NLST-885

Initial CDAS Request Approval
Feb 16, 2022

Title
A Deep Learning Model for Improved Cancer Risk Prediction in Sequential Lung Screening X-Rays

Summary
We will analyze the X-Ray examinations of the chest obtained in participants in the National Lung Screening Trial prior to the tissue diagnosis of lung cancer. The X-Ray examinations of patients diagnosed with lung cancer will be compared with X-Ray examinations of patients who did not develop lung cancer during the trial matched for age, gender and smoking exposure. We will also leverage the sequential screenings per patient to capture changes in the imaging features that are predictive of lung cancer. We postulate that a deep learning algorithm can be trained to estimate the risk for developing a clinically active lung cancer within the next 12 months (1-year risk) and within the next 2 to 6 years.

Aims

- Build a state of the art deep learning approach for lung cancer risk prediction
- Model the temporal changes in X-Ray images for individual patients
- Leverage sequential images over time to improve cancer risk prediction over non-sequential approaches

Collaborators

Regina Barzilay, PhD, MIT