Zhu, Hongmei2018-03-012018-03-012017-08-212018-03-01http://hdl.handle.net/10315/34387Manual segmentation is used in the diagnosis, management and evaluation of clinical trials for Multiple Sclerosis (MS), but human error makes manual segmentation variable. Automatic segmentation has been proposed using a Machine Learning algorithm Dictionary Learning (DL). We explored using different feature spaces to automatically segment MS lesions from healthy brain tissue. Methods of image texture analysis quantify the spatial distribution of the voxels in multi-weighted MR scans. We present the results of using a single voxel, single voxel and standard deviation (sigma) of adjacent voxels and a large spatial patch as feature spaces. The single voxel method segments the MS lesions with a Dice Similarity Coefficient (DSC) of 0.985 on simulated Brainweb data, but performed poorly with noise in the image (0.654). The single voxel and sigma performs at a DSC of 0.943 in the presence of 3% noise. The method should be attempted on real patient data.enAuthor owns copyright, except where explicitly noted. Please contact the author directly with licensing requests.Applied mathematicsElectronic Thesis or Dissertation2018-03-01Automatic segmentationMultiple SclerosisMultiple sclerosis lesionMachine learningDictionary learningApplied mathematicsMedical imagingMRIMagnetic resonance imagingMulti-weighted MRISegmentation