Comparative Analysis of Feature Extraction Methods for the Classification of Prostate Cancer from TRUS Medical Images
Radhakrishnan Manavalan, Kuttiannan Thangavel · 2012
Diagnosing Prostate cancer is a challenging task for Urologists, Radiologists, and Oncologists. Ultrasound imaging is one of the hopeful techniques used for early detection of prostate cancer. The Region of interest (ROI) is identified by different methods after preprocessing. In this paper, DBSCAN clustering with morphological operators is used to extort the prostate region. The evaluation of texture features is important for several image processing applications. The performance of the features extracted from the various texture methods such as histogram, Gray Level Cooccurrence Matrix (GLCM), Gray-Level Run-Length Matrix (GRLM), are analyzed separately. In this paper, it is proposed to combine histogram, GLRLM and GLCM in order to study the performance. The Support Vector Machine (SVM) is adopted to classify the extracted features into benign or malignant. The performance of texture methods are evaluated using various statistical parameters such as sensitivity, specificity and accuracy. The comparative analysis has been performed over 5500 digitized TRUS images of prostate.