An Enhanced Fuzzy Segmentation Framework for extracting white matter from T1-weighted MR images

Abstract

BackgroundWhite matter atrophy computed from Magnetic Resonance (MR) images is a clinical indication of a broad spectrum of neurological disorders. Accurate segmentation of white matter from MR images is necessary to estimate the white matter volume. Most of the techniques in literature used for the segmentation of white matter are computationally slow. The repeatability of segmentation results and consistency of performance on different input images are often poor. ObjectivesA computationally simple fuzzy clustering technique termed Enhanced Fuzzy Segmentation Framework (EFSF) for segmenting the white matter from the T1-Weighted MR images is proposed in this paper. MethodologyIN EFSF, the fuzzy membership function and prototype value are derived from the generic objective function of FCM using the method of Lagrange’s multiplier. The membership and prototype values are updated iteratively. The clustered image is obtained by replacing each grey level in the input image with the prototype value of the cluster with the largest membership value in the corresponding row of the fuzzy partition matrix. The pixels in the clustered image whose values are equal to the largest prototype value belong to the white matter region. ResultsOn 100 test images, the Dice Similarity Index (DSI) and the computational time (in sec) shown by EFSF are 0.8051 ± 0.0577 and 0.6522 ± 0.0502, respectively. ConclusionEFSF offers high segmentation accuracy and is computationally fast. Segmentation results offered by EFSF have good repeatability on the same MR slice and consistency on MR slices from various regions of the brain.