Enhancement of MRI human thigh muscle segmentation by template-based framework

Image segmentation of anatomic structures is often an essential step in medical image analysis. A variety of segmentation methods have been proposed, but none provides automatic segmentation of the thigh. In magnetic resonance images of the thigh, the segmentation is complicated by factors, such as artifacts (e.g. intensity inhomogeneity and echo) and inconsistency of soft and hard tissue compositions, especially in muscle from older people, where accumulation of intermuscular fat is greater than in young muscles. In this paper, the combination framework that leads to a segmentation enhancement method for region of interest segmentation are demonstrated. Appropriate methods of image pre-processing, thresholding, manual interaction of muscle border, template conversion and deformable contours in combination with image filters are applied. Prior geometrical information in an initial template image is used to automatically derive the muscle outlines by application of snake active contours, in serial images within a single MRI dataset. Our approach has an average segmented output accuracy of 93.34% by Jaccard Similarity Index, and reduced the processing time by 97.73% per image compared to manual segmentation.