Bone Segmentation of Magnetic Resonance Images by Gradient Vector Flow Active Contour with Atlas Based Centroid Forces

Abstract

This paper presents a segmentation technique which utilizes atlas based centroid forces coupled with GradientVector Flow (GVF) parametric active contour for the segmentation of femoral cancellous bone. The atlas used in our study provides prior information to constraint contours at regions where edge based forces are missing and to initialize the active contours. GVF external force field is padded with the centroid force derived from the atlas. In our implementation, once the atlas is registered with the target image to be segmented, the segmentation process is fully automatic. Analysis of segmentation accuracy of twenty one slices at the intercondylar location of sagittal slices provides sensitivity of 97.4±1.9%; specificity of 99.6±0.1%; Dice similarity coefficient of 96.7±1.1%. From the inspection of external force fields and the accuracy results, the study suggests that the centroid force formulation is effective in approximating missing boundaries in GVF and in facilitating automatic initialization.