Reduction of false positives in computer diagnosis of chest X-ray images using interval change detection between two images

We propose a procedure to reduce false positives included in the set of suspicious regions (SR's) automatically extracted by the lung cancer detection system for chest X-ray images. The procedure detects interval changes by referring the previous image taken at several months or years ago and removes SR's which show little changes. We apply the procedure to twenty one pairs of chest X-ray images and find that about sixty percent of the false positives were reduced without any effect on the true positives which overlap the true lung cancer shadows.