Feature-Based Classification of Prostate Ultrasound Images using Multiwavelet and Kernel Support Vector Machines

Prostate segmentation has been a challenging task that has long hindered the progress of ultrasound-guided procedures of the prostate. The focus of this paper is on the problem of feature extraction, classification and labeling of prostate tissues and non-prostate tissues for segmentation purposes. Specifically, we propose using multi-wavelet decomposition for prostate feature extraction and support vector machines (SVMs) for prostate differentiation and classification, respectively. Multiwavelets have been shown to possess important properties such as orthogonality, symmetry and compact support which can not be maintained simultaneously with traditional scalar wavelets. Here, multiwavelet features are extracted not from the whole image but rather are collected from several overlapping subwindows represented by square-shaped patches. The extracted features are then used to train an SVM classifier to recognize prostate from non-prostate tissues. Extensive experimentation and comparisons applied to different level of multiwavelet decomposition with different processing algorithms are also presented at the end of this paper.