Thigh muscle segmentation using a hybrid FRFCM-based multi-atlas method and morphology-based interpolation algorithm

Abstract

The volume of lower extremity muscles is affected by some diseases. Quantification of thigh muscles in medical images can lead to an easier investigation of these diseases. Most of the previous works in thigh muscle segmentation are based on models and atlases that require manually segmented datasets in 3D. As manual segmentation of these muscles is a time-consuming task, in this work, only one initial slice is segmented by a new hybrid FRFCM-based multi-atlas method and other slices are segmented based on this slice. In the proposed method, after noise reduction, the muscle region is extracted from other tissues by the FRFCM method. Then, an initial slice of each dataset is segmented by a multi-atlas method. The segmented muscles in the initial slice are used to segment muscles in the other slices of each dataset. The proposed method was evaluated with 20 CT datasets. The average DSC, Precision, and Sensitivity of the method for individual muscle segmentation were 91 . 20 ± 2 . 37, 91 . 95 ± 3 . 54, and 90 . 71 ± 3 . 89, respectively. The quantitative and intuitive results of the proposed method show the better results of this method in comparison to other state-of-the-art thigh muscle segmentation techniques.