Detection of increasing profusion of opacities from a sequence of personal chest radiographs

An effective method for an automated detection of long-term interval changes from a sequence of time-varying medical images is presented. We combine the pattern histogram method with an autoregressive model and novel classification algorithm to distinguish real abnormality variations from chaotic behaviors of background patterns without using an image registration technique. We conducted an experiment using 16 personal chest radiographs of pneumoconiosis acquired at the time span of 33 years. It is confirmed that the proposed method is able to detect suspicious regions and it is advantageous to conventional image comparison techniques particularly when the precise image registration is not available prior to subtraction.