Statistical bilateral asymmetry measurement in brain images

We present an improvement of an automated generic methodology for symmetry identification, asymmetry quantification, and segmentation of brain pathologies, utilizing the inherent bi-fold mirror symmetry in brain imagery. In the pipeline of operations starting with detection of the symmetry axis, hemisphere-wise cross registration, statistical correlation and quantification of asymmetries, we segment a target brain pathology. The detection of pathological difference left to right in brain imagery is complicated by normal variations as well as geometric misalignment in anatomical structures between two hemispheres. Introducing hemisphere-wise registration and spatial correlation makes our approach perform robustly in the presence of normal asymmetries and systematic artifacts such as bias field and acquisition noise.