BUSI Classification Based on Singularity-Exponent-Domain Image Feature Transform and Deep Neural Network

A novel breast ultrasonic image classification is proposed based on singularity-exponent-domain image feature transform (SIFT) and deep neural networks. As an extension of the conventional singularity power spectrum (SPS), the SIFT of the ultrasonic image is deduced, to extract the statistical feature image of the power distribution in the singularity exponent domain. The point-wise SPS is derived from the pseudo Wigner-Ville distribution (PWVD) of the breast ultrasound images (BUSI). Furthermore, the feature images of the BUSI are extracted to discriminate the specificity of the tumor regions. Then, a novel BUSI classification method based on the SIFT and deep learning network is proposed and tested on the public BUSI datasets. The experiment results indicate that the classification accuracy of the original deep network is considerably improved by about 6% with the proposed method.