A Probabilistic Adaptive Cerebral Cortex Segmentation Algorithm for Magnetic Resonance Human Head Scan Images

Abstract

The total efficiency of Magnetic Resonance Imaging (MRI) results in the need for human involvement in order to appropriately detect information contained in the image. Currently, there has been a surge in interest in automated algorithms that can more precisely divide medical image structures into substructures than prior attempts. Instant segregation of cerebral cortex width from MRI scanned images is difficult due to noise, Intensity Non-Uniformity (INU), Partial Volume Effects (PVE), MRI's low resolution, and the very complicated architecture of the cortical folds. In this paper, a Probabilistic Adaptive Cerebral Cortex Segmentation (PACCS) approach is proposed for segmenting brain areas of T1 weighted MRI of human head images. Skull Stripping (SS), Brain Hemisphere Segmentation (BHS) and CCS are the three primary processes in the suggested technique. In step 1, Non-Brain Cells (NBC) is eliminated by a Contour-Based Two-Stage Brain Extraction Method (CTS-BEM). Step 2 details a basic BHS technique for Curve Fitting (CF) detection in MRI human head images. The left and right hemispheres are divided using the discovered Mid-Sagittal Plane (MSP). At last, to enhance a probabilistic CCS structure with adjustments such as prior facts change to remove segmentation bias; the creation of express direct extent training; and a segmentation version based on a regionally various Gaussian Mixture Model- Hidden Markov Random Field – Expectation Maximization (GMM-HMRF-EM). The underlying partial extent categorization and its interplay with found image intensities are represented as a spatially correlated HMRF within the GMM-HMRF-EM method. The proposed GMM-HMRF method estimates HMRF parameters using the EM technique. Finally, the outcomes of segmentation are evaluated in terms of precision, recall, specificity, Jaccard Similarity (JS), and Dice Similarity (DS). The proposed method works better and more consistently than the present locally Varying MRF (LV-MRF), according to the experimental findings obtained by using the suggested GMM-HMRF-EM methodology to 18 individuals' brain images.