Skip to Main Content

COVID-19 is an emerging, rapidly evolving situation.

What people with cancer should know: https://www.cancer.gov/coronavirus

Get the latest public health information from CDC: https://www.coronavirus.gov

Get the latest research information from NIH: https://www.nih.gov/coronavirus

Principal Investigator
Name
Matthew Freedman
Degrees
MD, MBA
Institution
Virginia Tech
Position Title
Adjunct Associate Professor Physics (in process)
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-106
Initial CDAS Request Approval
Jan 22, 2015
Title
Computer-Aided System for the Improvement of Nodule Detection and Characterization in Thoracic CT based on a novel vessel suppressed CT Volume
Summary
This project would like to use CT cases from the NLST to blind test two different versions of a new computer-aided detection (CADe) system for lung nodules present on lung CT. The CADe system employs an advanced machine learning techniques capable of separating disease and normal vascular structure based on three dimensional image patterns. Three tests are planned:
1. Expert re-read to establish truth: A panel of expert radiologists to confirm NLST findings and add additional location information such as x-y-z (slice) coordinates and select nodule characteristics. Cases to be requested
a. Cases which contain a NSCLC identified on the T1 or T2 screening exams (300 cases requested).
b. Cases which contain a benign lung nodule (confirmed as benign by NLST procedures) (300 cases)
c. Cases which contain no identified lung nodule greater than 3 mm (300 cases)
d. Preferably, 100 cases or more should be non-solid or semi-solid nodules. (desirable distribution: 50 in 1a and 50 in 1b)
2. A stand alone (machine test) of a CADe system to determine maximum possible sensitivity and its false positives/case (fp/case)
3. Three observer (reader) studies where 20 radiologists interpreting the cases under three conditions
a. Without computer assistance
b. With computer assistance. The computer assistance will be tested in two arms (10 radiologists each)
i. Sequential read where the radiologists first interpret the CTs without CADe immediately followed by their re-interpretation when CADe is made available
ii. Concurrent interpretation where the radiologists are presented with both the standard and CADe images at the same time and are able to look at them in whatever simple or complex sequence that they decide to follow.
4. We will evaluate the new temporal subtraction software to characterize change, if any, in nodule size. For this reason we will need cancer and benign nodule cases identified on the T1 and T2 screens along with their prior CTs so that we can test for change from the preceding studies.
5. Statistical tests will include free response receiver operating characteristic (JAFROC), sensitivity and specificity.
6. Our alternative hypotheses are
a. Observers using the CADe system will detect more lung cancers, significantly greater as measured by JAFROC.
b. Observers using the CADe system will detect more benign nodules, measured as in 6a
c. That the simultaneous presentation of CADe results with the non-processed CT
i. Will be non inferior to the baseline reading performance.
ii. Will take less time than the sequential reading
Aims

This project will test software developed by Riverain Technologies, LLC to support an application to the FDA for clearance. The software is designed to assist radiologists in their detection of lung cancer on lung CTs. The specific aims are:
1. Collect and validate, using a panel of expert radiologists, lung CT cases of patients (subjects) with NSCLC, benign nodules, and cases with no lung nodules larger than 3 mm (to be called “normal”). Determine if changes have occurred between/among sequential studies.
2. Separate cases into two sets. A smaller set for training the radiologists and a larger set for testing lung cancer detection applications.
3. Perform a stand-alone machine test to determine the sensitivity and number of false positives per case of the software alone.
4. Perform three tests of the radiologists’ success in identifying lung cancer vs “normal”cases measuring
a. Performance interpreting the CT cases without the use of software by the radiologists
b. Performance interpreting the CT cases with the computer aided detection software
5.Perform tests of the new temporal subtraction software to characterize change, if any, in nodule size.
6. Perform statistical analysis of the results of these tests. Analyses will include Free-Response Receiver Operating Characteristic, measures of sensitivity and specificity, measures of false positive rate.
7. Report findings.

Collaborators

ShihChung Ben Lo, Ph.D. Arlington Innovation Center: Health Research, Virginia Tech-National Capital Region

Related Publications