Abnormality Detection
Principal Investigator
Name
Ryan Chamberlain
Degrees
PhD
Institution
Invenshure LLC
Position Title
Director of Imaging
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-168
Initial CDAS Request Approval
Oct 8, 2015
Title
Abnormality Detection
Summary
Automatically detecting abnormalities in thoracic CT images is difficult with traditional computer vision methods that rely on manual feature engineering due to the large variation in abnormalities. The sensitivity and specificity of such methods are usually too low to provide clinical value.
Convolutional neural networks (CNNs) provide an alternative path to computer vision problems by learning the features of interest automatically when trained with large amounts of labeled data. The NLST data is an ideal dataset for training CNNs because of the consistency and quality of the labeled data.
This project will investigate the ability of CNNs to automatically identify areas of a thoracic CT that a Radiologist would have identified an abnormality.
Convolutional neural networks (CNNs) provide an alternative path to computer vision problems by learning the features of interest automatically when trained with large amounts of labeled data. The NLST data is an ideal dataset for training CNNs because of the consistency and quality of the labeled data.
This project will investigate the ability of CNNs to automatically identify areas of a thoracic CT that a Radiologist would have identified an abnormality.
Aims
SA 1: Build a framework for training a CNN on NLST data
SA 2: Determine the sensitivity and specificity of a variety of CNN architectures in identifying abnormalities in NLST images
Collaborators
Jeremy Friese, MD, Invenshure