On the Use of Spectral Clustering for Prostate Tissue Characterization

Prostate cancer is the second most diagnosed malignancy in men over the age of fifty. Trans-rectal ultra-sound (TRUS) is among the routine tests that are used for prostate cancer diagnosis as well as for directing prostate biopsy. To identify suspicious cancerous regions, TRUS images are usually segmented manually by an expert radiologist which can be a laborious process. In this paper, we investigate the use of spectral clustering for identifying suspicious cancerous regions in prostate TRUS images. Our preliminary results confirm that there is a high correlation between the desired segmentation and the second smallest eigenvector of the affinity matrix of the TRUS image