Program Information
Quantitative Analysis of Temporal Subtraction Chest Radiographs
S Trinkle*, R Engelmann , H MacMahon , S Armato , The University of Chicago, Chicago, IL
Presentations
SU-H2-GePD-I-2 (Sunday, July 30, 2017) 3:30 PM - 4:00 PM Room: Imaging ePoster Lounge
Purpose: To identify a set of quantitative image signatures for regions of pathologic change in temporal subtraction (TS) chest radiographs.
Methods: A set of 120 chest TS images from 59 patients was obtained. A total of 73 regions of pathologic change were outlined by an experienced radiologist and verified with a corresponding computed tomography (CT) scan. The morphologies of these regions were used to generate a total of 163 “false” regions centered at specified locations within the lung regions of the same image set and corresponding to known regions of no pathologic change. For each “true” and “false” region, 22 first-order texture features were calculated based on gray-level histograms. Receiver operating characteristic (ROC) analysis was performed and the area under the ROC curve (AUC) was calculated using all false regions and (a) the entire truth set, (b) the subset of “dark” truth regions (n = 60) corresponding to areas of new or growing abnormality, and (c) the subset of “bright” truth regions (n = 13) corresponding to areas of resolved abnormality.
Results: AUC values exceeded 0.80 for four quantitative signatures for regions of pathologic change in TS images using the entire truth set. Classification performance improved significantly when dark regions and bright regions were analyzed separately: twelve features had AUC > 0.80 for the subset of tissue growth regions, and nine features had AUC > 0.80 for the subset of tissue recession regions.
Conclusion: This work identified up to twelve quantitative image signatures for regions of pathologic change in TS chest radiographs, motivating the development of a rigorous automatic CAD system for TS images. Gray-level thresholding techniques and information from these identified signatures will be used in future investigation of this system along with information from higher-order texture analysis.
Funding Support, Disclosures, and Conflict of Interest: SA receives royalties and licensing fees for computer-aided diagnosis technology.
Contact Email: