Mathematical Characterization of Lung Nodules Identified by Screening Lung CT

Principal Investigator

Name
Matthew Freedman

Degrees
MD

Institution
Georgetown University

Position Title
Professor of Radiology

Email
freedmmt@georgetown.edu

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
2003-90018

Initial CDAS Request Approval
Jun 16, 2003

Title
Mathematical Characterization of Lung Nodules Identified by Screening Lung CT

Summary
This proposal will attempt to identify key radiologic features of nodules and translate them into mathematical formulations. A trained vision-based neural network system will be used to automatically extract the features. The clinical characteristics of nodules will then be correlated with the nodule features.

Aims

Screening lung CT has a high sensitivity for the detection of lung nodules, but many of the nodules identified cannot be directly classified as benign or malignant. Nodules identified on screening CT have various mathematical features that may help to determine their malignant potential. Our group has defined computational features of nodules that define their mathematical characteristics. The CAD group at Georgetown University, directed by Ben Lo, Ph.D. and Matthew Freedman, MD, MBA, has done a significant amount of work on the computational features that are defined based on image patterns and textures of the nodules. In addition, we have also trained a vision-based neural network system to automatically extract the features. We would like to apply these mathematical features to nodules identified on Georgetowns screening lung CTs from the NLST and LSS protocols. Because histologic diagnoses are not yet available on the majority of screen-detected nodules, we will not be able to make immediate correlations. However, as the cases evolve we will be able to correlate the image features with the histology of the nodules.