Automatic segmentation of facial soft tissue in MRI data based on non-rigid normalization in application to soft tissue thickness measurement

Abstract For measuring the thickness of soft tissue in magnetic resonance (MRI) images, precise borders between skull and face surfaces should be known. We present an algorithm for segmentation of the human head in T1-weighted MRI images that generates smooth, complete segments of head tissues for further landmarks definition and measurements of the soft tissue thickness of the human head. As a segmentation tool we use an algorithm based on nonlinear normalization of the MRI template to MRI data and application of transform matrix to the head model. The algorithm uses preprocessed subject MRI data and a head model with separate tissue segments. The head model is obtained using a hybrid algorithm and consists of four segments: soft tissue, skull, brain and air. To assess the precision of segmentation, specificity, sensitivity, Dice and Jaccard Similarity Coefficients were computed. The algorithm was tested on MRI images from 10 Caucasian adults from free public database IXI. Specificity of 93% and 98% and sensitivity of 87% and 93% was achieved for soft tissue and brain segment, respectively. Specificity of 67% and 72% and sensitivity of 83% and 62% was achieved for the skull and air segments, respectively.