Performance Comparison of Segmentation Algorithms for Image Quality Degraded MR Images

Medical image segmentation is one of the most important research areas of clinical diagnosis. Especially, brain is the most critical organ that is tracked, investigated and analyzed mostly by using Magnetic Resonance Imaging (MRI). Developing a highly accurate automated segmentation of brain region is a very difficult task due to involving noise and deviation. In recent years, various image segmentation techniques have been developed in the literature such as clustering, thresholding (intensity-based), active contours (surface-based), expectation maximization (probability-based). In this study, these commonly used algorithms are handled in order to see the performance of the segmentation while degrading the image quality and saving from memory for brain MR images. For this purpose, the level of acceptable degradation is obtained by compressing MR slice images with different quality factors by using JPEG algorithm. Peak signal to noise ratio (PSNR), bits per pixel (BPP), mean, variance parameters of the MR images are used to characterize the corresponding compressed image degradation quality. On the other hand, segmented intracranial area, white matter (WM), gray matter (GM) regions are compared with the non-compressed MR images for various compression ratios. Then, the area overlap ratio for these regions is obtained in order to get segmentation performance results. It is believed that detected optimum parameters can be used as prior indicators to determine which segmentation algorithm (or which group, i.e. intensity or surface-based) should be chosen. Besides, it will be able to occupy less space in memory by compressing image for appropriate parameters.