Fast tissue segmentation based on a 3D and 4D feature map: comparison study

Abstract

The aim of this work was to develop a fast and accurate method for tissue segmentation in MRI based on 4D feature map and compare it with that derived from the 3D feature map. High resolution MR imaging was performed in 5 normals, 6 patients with brain MS, and 6 with malignant brain tumors. Three inputs: proton density, or weighted fast spin-echo, T1-weighted spin echo MR images were routinely utilized. As a fourth input, either magnetization transfer MRI was used in normals and some patients or Tl weighted post contract MRI in other patients. Modified k-Nearest Neighbor segmentation algorithm was optimized for maximum computation speed and high quality segmentation. In that regard: 1) the authors discarded the redundant seed points, 2) discarded the points within 0.5 standard deviation from the cluster center that were non overlapping with other tissue classes, 3) they removed outlying seed points outside 5 times the standard deviation from the cluster center of each tissue class. Their new technique utilizing all 4 MRI inputs provided better segmentation than that based on three inputs. (p<0.001) The tissues were smoother and the delineation of the tissues was increased. Details that were previously blurred or invisible now became apparent. In normals, detailed depiction of deep gray matter nuclei was obtained. In malignant tumors, up to 5 abnormal tissues were identified: 1) solid tumor core, 2) cyst, 3) edema in white matter 4) edema in gray matter and 5) necrosis. Delineation of MS plaque in different stages of demyelination, became much sharper. In conclusion, proposed methodology warrants further development and clinical evaluation.