Graph-based prostate extraction in T2-weighted images for prostate cancer detection

In this paper, our ultimate purpose is to extract a prostate from a T2-weighted image for easily detection of prostate cancer. Therefore, we present an algorithm of the prostate extraction by using a graph-based unsupervised and semi-supervised learning. An image is made up of inhomogeneous regions. Only the homogeneous region of an image can be segmented in image processing technologies. The prostate is also made up of inhomogeneous regions. We cannot segment the prostate from the T2-weighted image by image processing technologies. The prostate is extracted as the following steps. First, entire inhomogeneous regions are detected in the T2-weighted image by a graph-based unsupervised scheme. Secondly, the placement of the stokes are decided by inhomogeneous regions in a semi-supervised learning. Finally, the prostate is extracted based on the stokes by the semi-supervised learning. Detection of prostate cancer is diagnosed by the histogram of the prostate which is extracted from the T2-weighted image by the graph-based unsupervised and semi-supervised learning.