Comparing the accuracy of machine learning diagnosis of lung cancer from different data modalities
Principal Investigator
Name
Polina Golland
Degrees
Ph.D.
Institution
Massachusetts Institute of Technology
Position Title
Henry Ellis Warren Professor of Electrical Engineering and Computer Science
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-1349
Initial CDAS Request Approval
Nov 5, 2024
Title
Comparing the accuracy of machine learning diagnosis of lung cancer from different data modalities
Summary
The NLST trial has demonstrated that lung cancer screening performed using CT is more effective than screening with chest X-ray. To further improve the efficiency and accuracy of screening, many CT-based machine learning diagnostic algorithms have been developed in recent years. The performance of top-performing algorithms is only slightly worse than expert chest radiologists and a few of these algorithms have even been granted FDA approval for clinical use. However, given the numerous other advantages of X-ray over CT (such as its portability, accessibility, and low radiation dose), there remains immense interest in building machine learning algorithms to diagnose lung cancer from chest X-ray images. While many X-ray-based diagnostic algorithms have been proposed in the literature, their performance has not been carefully benchmarked relative to existing CT-based algorithms. Therefore, we propose to use the NLST dataset to measure the diagnostic performance gap between X-ray and CT-based machine learning algorithms. Finally, to complete this benchmarking, we also propose to compare these models against diagnostic algorithms built for pathology data.
Aims
1) Retrain open-source implementations of the top-performing machine learning algorithms on the large corpuses of X-ray, CT, and pathology images in the NLST dataset
2) Evaluate the performance of all methods using multiple performance markers and measure the performance gap between X-ray and CT/pathology-based algorithms
Collaborators
Vivek Gopalakrishnan, MIT
Peiqi Wang, MIT
Polina Golland, MIT