Analysis of Histogram Distance Measures for Change Detection In Brain Symmetry

It is important to analyze and quantify the spatial relationship between the brain symmetry planes (BSP) on radiological images. The unsupervised change detection problem is taken to detect the abnormalities based on brain symmetry assuming the left and right part of brain is roughly symmetric. The input to the problem is symmetrical axial MR slices, and its output is to place axis-parallel boxes that circumscribe the tumor part. This change detection process uses a score function based on histogram distance measure computed with gray level intensity histograms. In this work histogram distance measures is analyzed for the change detection problem. Several image databases are utilized to test the performance of Bhattacharyya, Chi-Square, Correlation, Intersection distance measures. From the experimental results, Chi-Square distance metric seems effective in localizing brain abnormalities than the other distance metrics for the change detection problem.