Shearlet-Based Ultrasound Texture Features for Classification of Breast Tumor

Abstract

Texture features are commonly used in the breast ultrasound computer-aided diagnosis (CAD). Shear let transform provides the spare representation of high dimensional data, and can be used to describe image texture. In this study, shear let-based texture features were extracted as the characterization of breast tumor in ultrasound images. Texture features were also extracted from wavelet and gray-level co-occurrence matrices (GLCM) for comparison. The AdaBoost algorithm was then used to classify breast tumor with the extracted texture features. The experiment result shown that the classification accuracy of shear let-based method was 88.0%, which was much better than those of wavelet- and GLCM-based methods. The results indicated that the texture features extracted by the proposed method could well characterize the properties of breast tumor in ultrasound image. It suggests that the proposed method has the potential to be used in breast CAD.