Computerized analysis for automated detection of lung nodules in digital chest radiographs

Abstract

The objective of this study was to develop a computer‐aided diagnosis(CAD) scheme for automated detection of lung nodules in digital chest images in order to help radiologists to improve their performance in detecting lung nodules. Lung nodules in chest radiographs are the important sign of lungcancer which is the leading cause of cancer death in men and women in the United States, with the 5‐year survival rate only about 13%. Early detection and treatment are crucial for the improvement of this survival rate for patients with lungcancer. This dissertation included the following four topics: (1) accurate determination of ribcage boundary in digital chest radiographs; (2) detection of right and left hemidiaphragm edges and delineation of lung fields in digital chest radiographs; (3) development of a CAD scheme for automated detection of lung nodules in digital chest radiographs; and (4) ROC studies of effects of CAD output on radiologists’ performance for detection of lung nodules on chest radiographs. Our CAD scheme for automated detection of lung nodules in digital chest images was based on the segmentation of lung fields, the difference image technique, image feature analysis, and an artificial neural network. The performance of the CAD scheme was evaluated by the FROC methodology. The CAD scheme achieved an improved performance of 70% sensitivity with 1.7 false positives per image, on average. The ROC analysis results indicated that radiologists increased their performance significantly in detecting lung nodules by use of the CAD output as a ‘‘second opinion.’’ Radiologists did not increase their reading time when the CAD output was used. [Thesis copies are available upon request from the author.]