Brain Tissue Segmentation Based On Corrected Gray-Scale Analysis

Image signal-to-noise ratio (SNR) and signal intensity (SI) inhomogeneities are factors that strongly affect the accuracy and precision of brain tissue segmentations in magnetic resonance image (MRI). In this work, SNR and contrast of brain images are optimized by TR and inversion recovery time TI in multi-spectrum MRI data sets. SI inhomogeneities are measured in vivo using a recently developed method allowing improved correction. The three-Gaussian distribution model is used to fit histograms of the images to find the initialization parameters for an expectation-maximization (EM) segmentation algorithm. Compared with other methods, the field map method provides better correction of SI inhomogeneities and excellent segmentation results