Abstract:
Segmentation of tissues in magnetic resonance images is essential especially for a radiologist to be able to identify a disease, tumors, or any tissue. In any magnetic resonance image there exists many different types of tissues each with characteristic T1 and T2 decay times and proton densities. If these parameters of tissues can be calculated from the regular magnetic resonance images, the type of tissue could also be determined on any MR image independent of MR hardware characteristics. One such important hardware limitation is the varying sensitivity of an imaging coil spatially. Segmentation algorithms can not distinguish between an intensity variation caused by the imaging coil sensitivity or a variation by tissue change. Calculated T1, T2, and PD images provide consistent pixel intensity corresponding to the same tissue therefore easier to utilize in conventional segmentation algorithms. To be able to calculate true T1 and PD parameters, a slice of human head were imaged sixteen times by holding TE fixed and changing TR each time. Levenberg-Marquardt Method is applied to the data and T1 and PD values were estimated. The true T1 and true PD images were produced. The maximum likelihood classification is then applied successfully to four MR images of different slices of human head and the robustness of this method in segmenting CSF, WM, and GM is illustrated.
