Computer-aided diagnosis for detection of changes in sequential chest radiographs (in Japanese)

Abstract

The subtraction image of digital chest radiographs taken at different times improved the diagnostic accuracy of radiologists viewing screening chest radiographs. The accurate registration of two chest images using “lung markings” instead of “rib edges” enables us to detect subtle changes in the “difference image.” The “lung markings” are the shadows of blood vessels traversing lung substance. Most of the morbid changes grow in lung substance not in the rib cage. Twenty-nine pairs of posteroanterior chest radiographs with and without temporal morbid changes were selected from approximately 200 sequential cases. The images were 1760×1760 matrix and 1024 gray scale with a computed radiography system (Fuji Medical System, Tokyo). Image registration employed the matching of “lung markings” in the two radiographs. Observer performance tests with and without the use of the “difference image” were carried out by eight radiologists. Six of the eight observers diagnosed with higher sensitivity. The mean for eight observers increased from 43.9% to 55.3% with a statistical significance of [formula omitted] The mean area under the AFROC (Alternative Free-Response Receiver Operating Characteristics) curve improved from 0.596 to 0.647 with a statistical significance of [formula omitted] The false positive response was comparable with and without the use of the subtraction image. We conclude that the use of the subtraction image with the new registration technique improves the diagnostic accuracy for detection of subtle changes between radiographs, such as nodule, infiltrate, and cardiomegaly of radiographic findings. © 1999, American Association of Physicists in Medicine. All rights reserved.