Discriminating spinal deformity employing centroids difference on the moiré images

Spinal deformities, a condition in which the spine deviates to the left or right from the midline, are on the rise, particularly among elementary and middle-school students. Group diagnoses are being performed at schools for the purpose of early detection, but the burden on physicians is high due to the need to process many images. Automated diagnosis using computers is highly desirable from the standpoint of treatment. In this paper the authors establish a left–right rectangular processing region in a moire image of a human back, then propose a method to automatically detect spinal deformities with the differences in the centroid positions as a feature. Using an approximately symmetrical reference axis which closely follows the midline of a person, and the amount of the feature from the centroid position of the left–right rectangular region as centered on the reference axis, the authors' method discriminates between normal cases and abnormal cases using a linear discrimination function on a two-dimensional surface. If this kind of primary diagnosis of spinal deformities can be automated, group diagnosis in schools would be more efficient and useful. The authors use their method, perform discrimination tests on 120 normal and abnormal cases, and find a recognition rate of 87.9%. © 2001 Scripta Technica, Syst Comp Jpn, 32(7): 20–28, 2001