Edge detection training of a deep learning algorithm using a 3D data set
Principal Investigator
Name
William Randazzo
Degrees
M.D.
Institution
AutoRad, LLC
Position Title
Chief Medical Officer
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-389
Initial CDAS Request Approval
Feb 12, 2018
Title
Edge detection training of a deep learning algorithm using a 3D data set
Summary
Machine learning has become increasingly useful in the field of radiology with a broad spectrum of applications being developed. These algorithms necessitate large data sets to gain the desired sensitivity and specificity of algorithm performance. As such, when building and training a deep learning algorithm, one layer comprises an edge detection model of a 3D image data set. Development of this layer is an important foundation for training subsequent data sets, which ideally are unrelated to data set used to for edge detection training, thereby improving algorithm performance.
The purpose of this project is to train a deep learning algorithm to perform edge detection on a large 3D data set of the chest. The desired goal beyond the scope of this project is to enhance downstream algorithm performance after additional training is performed on other 3D data sets, including anatomic regions outside of the chest.
The purpose of this project is to train a deep learning algorithm to perform edge detection on a large 3D data set of the chest. The desired goal beyond the scope of this project is to enhance downstream algorithm performance after additional training is performed on other 3D data sets, including anatomic regions outside of the chest.
Aims
Aim 1: Train a deep learning algorithm to perform edge detection based upon multi-slice (3D) chest CT images.
Aim 2: Assess the performance of the edge detection model on other 3D data sets, obtained outside of the NLST.
Collaborators
Elliot Stein, A.B., CEO, AutoRad, LLC
Sam Brotherton, B.A., Cairn Labs/AutoRad, LLC