FAST AND AUTOMATIC BRAIN EXTRACTION BY GEODESIC PASSIVE CONTOURS (BEGPC)

Abstract

Brain extraction is pre-processing step for quantitative analysis of biomedical images. Precision at brain extraction is not solved by current techniques especially on T2 MRI towards top of the brain where challenging non-brain tissues appear like coastal band and there is a trade-off between precision and speed. Moreover, full automaticity obligates failure at challenging images by preventing user intervention. This novel technique (BEGPC) is simple, fast and full automatic which utilizes geodesic passive contours (GPC) and imitates edge-sensitiveness ability of an expert. Edges are caught at correct GPC by utilizing morphology of brain. Segments of nearby geodesics exclude non-brain tissues at enhancement stage. It has better precision than existing techniques whereas it gets ahead much at challenging images without sacrificing full automation. Performance of BEGPC is verified on axial and coronal T2 MRI of 40 patients by comparing with manual drawings of radiologists and with known techniques like BET in literature. It has an average correct boundary rate over 90%, similarity over 99% on normal images and performs %10 better than BET at challenging ones.