关键词: 图像分割, 磁共振图像, 海马结构

Abstract: Objective To overcome the difficulties of delineating hippocampal boundaries in an effective way and benefit the related researches and applications by quantitative information from MR images,the paper presents a new method of computer-aided segmentation.Methods All possible edges at the region of interest were extracted by de-nosing edge detection methods with threshold zero,and they were highlighted.Compared with the original image,the edges which belong to the boundaries of hippocampus were identified and others were removed by a tool.The tool was the computer program developed for aiding to remove edges and modified the boundaries of hippocampal formation.Multiple edges could be removed by marking the two diagonal points of the region at which edges located.With the help of the tool,the magnified boundaries could be further improved by adding the missed pixels and deleting the redundant points,which guarantee the accuracy of the boundaries.Results Most parts of hippocampal boundaries could be well detected by Sobel operators.This made the proposed method more reproducible than manual segmentation.It took lower cost to remove undesired edges by the tool than to delineate hippocampal boundaries by manual tracing.With the guide of the boundaries detected,it was easier to locate the missing reqions of boundaries.Adjusting few pixels at boundaries made it possible to achieve the same accuracy as manual segmentation.Conclusion Although the proposed method cannot free us from manual operation,it is simple and practical.Before reliable automated approaches are fully available,the technique may be a better alternative.

Key words: image segmentation, MR image, hippocampal formation