Machine Learning Classification of Nodules
Principal Investigator
Name
Michal Lada
Degrees
MD
Institution
University of Rochester
Position Title
Assistant Professor
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-484
Initial CDAS Request Approval
Feb 19, 2019
Title
Machine Learning Classification of Nodules
Summary
The goal of our study is to better predict benign vs malignant lung nodules to improve the currently very high false positive rate related to CT scans.
We would like a customized selection of participants in order to have balanced dataset for training the algorithm.
Ideally we would like to have 40% scans of no cancer or abnormalities, 30% scans with a malignant nodule, and 30% scans with a false positive nodule (resulted in more testing but was proven to be benign).
This would allow us to have relatively balanced data during: (1) automated nodule detection, and (2) benign vs malignant prediction.
The actual number of scans would be set by whichever of the true or false positive cohorts has fewer patients available, and then the other cohort numbers determined by the above percentages to produce an overall balanced dataset.
We would like a customized selection of participants in order to have balanced dataset for training the algorithm.
Ideally we would like to have 40% scans of no cancer or abnormalities, 30% scans with a malignant nodule, and 30% scans with a false positive nodule (resulted in more testing but was proven to be benign).
This would allow us to have relatively balanced data during: (1) automated nodule detection, and (2) benign vs malignant prediction.
The actual number of scans would be set by whichever of the true or false positive cohorts has fewer patients available, and then the other cohort numbers determined by the above percentages to produce an overall balanced dataset.
Aims
- Use imaging and patient characteristics to predict benign vs malignant
Collaborators
Brian Ayers, University of Rochester