Abstract
Comparison of more than two chest x-ray images and detection of changes among them are very significant in diagnosis and therapy of lung and heart diseases. In this paper, results of fundamental studies on automated comparison of chest x-ray images and detection of changes among them are presented. First, characteristics of variation in the shape of images of organs, and of variation in density values are estimated from a set of several films of the same patient taken at different times. A few methods for correcting variation in the shape of rib images and_ in density values are given together with comparative study of their performance and limitations. Second, effectiveness of a local correlation coefficient as a measure of change is investigated. Experiments are performed on the detection of abnormal shadows using the correlation coefficient between two subpictures cut out from the corresponding locations in right and left lung of the same film. A seventy-five percent detection rate is achieved for eighty pairs of subpictures obtained from chest photofluorograms. From the result achieved the method used here is considered promising.
