Extended boundary shift integral

Abstract

The boundary shift integral (BSI) is a widely used method for measuring atrophy rate, dynamic changes of the gray-matter and cerebrospinal fluid boundaries in magnetic resonance images. BSI is based on intensity differences on this boundary region. This work extends the method in two respects: 1) Instead of using only intensity information on the boundary region, a probabilistic approach is proposed in which also other characteristics of the boundary region can be used. 2) The use of the probabilistic model enables to measure changes between any structures or combination of structures in the image. The performance of the extended BSI is verified against standard BSI in the ADNI and AIBL cohorts. The area-under-the-curve is clearly above 90 % in both cohorts when comparing the classification between cognitively normal and Alzheimer's disease groups. The accuracies of the extended BSI were higher than the standard BSI between these groups.