Segmentation of muscles in CT volumes using a continuous convex relaxation approach

Abstract

Segmentation of muscles in CT volumes is a complicated task even for experts due to their Hounsfield values, which are close to those of surrounding organs. Surgeons and physicians usually prefer to segment them manually due to the lack of efficient tools in implementing a correct segmentation. This is a very time-consuming task, as CT volumes can contain even hundreds of 2D slices. In this paper an algorithm to implement the segmentation of skeletal muscle structures have been developed. 30 CT images belonging to different patients have been used to assess the tool. Results in terms of efficiency and computational times are promising providing DICE and sensitivity coefficients higher than 0.87.