基于射线距离误差的三维医学图像分割的新型评价方法

三维医学图像的分割是计算机辅助诊断与治疗中至关重要的技术。由于现有的三维医学图像分割算法种类繁多且性能各异,对各种分割算法性能公正直观的评价显得尤为重要。我们在对现有常用评价方法做出综合的比较后,提出了一种基于局域距离误差的综合、直观的三维图像分割的新型评价方法,并详细描述了该评价方法中局域误差和全局精确度的评价法。最后,利用这种新型的评价方法对三种常用的分割算法(Fast Marching,Morpholo-gical Reconstruction and Watershed)及一种笔者提出的新型Hybrid算法进行评价,此新型分割算法也在文中作了简要介绍。通过实验,可以验证新的评价方法能够更加科学、直观地对各种三维医学图像分割算法进行评价。

A Novel Validation Method Based on Radial Distance Error for 3D Medical Image Segmentation

3D medical image segmentation is a fundamental technique for computer assisted surgery and therapy. There exist various segmentation algorithms, but still we hope to have a more efficient and intuitive evaluation method for validation. In this paper, after reviewing the existing validation methods, we propose a new approach for 3D image segmentation validation. The details of the method are introduced in two parts: local error validation and global accuracy validation. Then, we apply it to validate three typical segmentation algorithms and a user proposed hybrid segmentation method in our experiment. The hybrid approach is also briefly introduced. The experimental results reveal that the proposed validation method can evaluate the employed segmentation methods more accurately and intuitionally.