Effect of despeckle filtering on classification of breast tumors using ultrasound images

Abstract

Ultrasound is the most widely used imaging modality for screening of breast tumors. However, due to the presence of speckle noise in an ultrasound image, the diagnostic information gets masked and the interpretation of the breast abnormalities becomes difficult for the radiologist. The texture of the tumor region and the shape/margin characteristics are considered to be important parameters for the analysis of the breast tumors. In the present work, exhaustive experimentation has been carried out for the design of CAD systems for classification of breast tumors by considering (a) original images only, (b) despeckled images only and (c) both original and despeckled images together (hybrid approach).Total 100 breast ultrasound images (40 benign and 60 malignant) have been used for the analysis. Initially, these images have been despeckled using six filters namely Lee sigma, BayesShrink, DPAD, FI, FB and HFB filters. Total 162 features (149 texture and 13 morphological features) have been computed from both original and despeckled breast ultrasound images and SVM classifier has been used extensively for the classification.The results of the study indicate that the hybrid approach of CAD system design using texture features computed from original images combined with morphological features computed from images despeckled by DPAD filter yield optimal performance for classification of benign and malignant breast tumors with a classification accuracy of 96.0%. From the promising results of the study it can be concluded that the proposed hybrid CAD system design could be used as a second opinion tool in clinical setting.