Prostate Tissue Classification Methods and Critiques


The work in this paper examines and criticizes different methods used for prostate tissue classification using trans-rectal ultra-sound (TRUS) images. The suspicious regions are first identified by an accurate region of interest (ROI) identification algorithm. The identified ROIs' are then analyzed using statistical based features as well as spectral based features. For the statistical based features each ROI is treated as an image and different statistical features are constructed from the ROI image. The statistical feature set is constructed from the grey level difference vector (GLDV), as well as the grey level dependence matrix (GLDM). While for the spectral based features, all of the ROIs' pixels are aligned to form a ROI I-D signal. Different spectral features are then constructed from the I-D ROI signals. The spectral feature set is constructed using geometrical features extracted from the estimated power spectrum density (PSD) as well as the estimation of signal parameters via rotational invariance technique (ESPRIT) features. A classifier based feature selection algorithm using ants colony optimization (ACO), a recently proposed optimization technique is adopted and used to select an optimal subset from each of the above extracted features. The obtained accuracy ranges from 72.2% to 93.75% using a Support Vector Machine classifier.

7 Figures and Tables

Cite this paper

@article{Mohamed2007ProstateTC, title={Prostate Tissue Classification Methods and Critiques}, author={S. S. Mohamed and M. M. A. Salama}, journal={2007 IEEE International Symposium on Signal Processing and Information Technology}, year={2007}, pages={144-149} }