Computing Optimization Technique in Enhancing Magnetic Resonance Imaging and Brain Segmentation of Hypophysis Cerebri Based on Morphological Based Image Processing

Background: The human brain is the central processing unit of the human body, controlling the actuation of muscles and coordination of limbs movement, hormone release, glands secretion, respiration and body temperature. Of particular interest is the isolation and identification of the hypophysis cerebri that is responsible for hormone control. As such, it is vital to develop a semi-automatic segmentation system for medical experts and radiologists involved in MRI or CT brain image diagnostics. In addition, due to the large image data set, it is important to devise a high-speed processing framework to perform medical image analysis. Methods and results: We proposed the fused Stationary Wave Transform (SWT) and Discrete Wavelet Transform (DWT) to enhance the resolution of the brain scan image. Then, we applied the shareholding and mathematical morphology approach, and also the active contour segmentation approach to determine the boundary of the anatomical structure. Based on visual evaluation by a team of expert radiologists, we derived the manual segmentation results that are utilized as ground truth information. The results from the semi-automatic active contour model based approach are compared with the manually segmented ones using the Dice and Jaccard indices. Good correlation was achieved for the shareholding and mathematical morphology approach whereby >95% results fall within the 95% confidence interval in the Student t-test) and higher Dice and Jaccard indices demonstrate that the proposed segmentation method of hypophysis cerebri using shareholding is more effective than the active contour model. To improve our image processing operations, we utilized CUDA GPU acceleration and proved that this is vital for high-speed image diagnosis. Conclusion: The proposed work shows that the shareholding technique is better comparatively to region growing active contour techniques. Our image segmentation approach that is coupled with GPU acceleration shows promise for the analysis and detection of abnormalities of the brain.