Isolation of Prostate Gland in T1-Weighted Magnetic Resonance Images using Computer Vision

In this work, segmentation of the region of prostate gland from T1-weighted magnetic resonant (MR) images is considered. There are many practical challenged that are encountered while segmentation such as the regions surrounding the prostate gland may be inconsistently defined, which can pose a problem for the algorithm to differentiate the pixels of the gland from the neighborhood pixels. The other problem may be the region around the gland may not be homogenous. Therefore, in this study, we have proposed a semi-automatic approach that can efficiently segment out the prostate gland in a computationally effective and robust manner as compared to the approaches available in the literature. Here, we have used image processing algorithms like anisotropic diffusion, Gaussian filtering, k-means clustering, and region-growing approach. The proposed method has been evaluated on the T1-weighted MR images of 30 subjects. The proposed method has given a Dice coefficient of 86.95% and a Jaccard index of nearly 80.56%.