Interfacial gradient priors-based geodesic geometric flows for 3D medical image segmentation

Purpose: The 3D medical image segmentation is a really difficult problem. The purpose of this paper is to present a novel segmentation method for cases that some regions of interest to be segmented from 3D medical images have strong similarities such as gradient between adjacent slides. Design/methodology/ approach: This method brings gradient characteristics of the adjacent-segmented slide, called interfacial gradient priors, into the slide waiting for segmentation and to help the contour converge to actual boundary more accurately. Findings: This method will improve the stopping criterion of curve evolution through introduction of adjacent slide's prior information into edge detection function, so that the leakage phenomena that exists in geometric active contour model when discontinuous or weak edges appear is reduced. Originality/value: Introducing adjacent slide's priors improves the precision and stability of geodesic geometric flows in 3D medical image segmentation.