Objective and Optical Evaluation of Despeckle Filters in Breast Ultrasound Images

Abstract

ABSTRACTSpeckle removal from breast ultrasound images result in blurring of lesion margin, echo pattern, and other sharp details which may carry important diagnostic information. Traditional image quality metrics fail to quantify the losses, dislocation of edges, and other sharp image features. Hence, an image quality metric is needed to measure the edge-retrieval capability of filter so that blurring, loss, and dislocation of edges can be quantified. In this paper, a novel image quality metric called edge retrieval index (in short, ERI) is proposed and utilized to evaluate state-of-art speckle removal techniques in breast ultrasound images along with traditional image quality measures. Besides the objective evaluation, optical evaluation is also carried out by two experts in extracting important diagnostic information from breast ultrasound images. Experiments were conducted on 28 real-time breast ultrasound images on MATLAB® software platform. The results of objective evaluation show that wavelet filter with first-level decomposition and eliminated HH band (WAV (HH/1)) gives the best performance followed by homomorphic filter with wavelet filter function (WAV (HH/1)) and anisotropic diffusion filter (PM2). On the other hand, wavelet filter (WAV (HL/1)) performed best in restoring edges while providing the highest smoothness. Furthermore, anisotropic diffusion filter (PM2) outperformed others, followed by homomorphic filter with wavelet filter function (WAV (HH/1)), in optical evaluation carried out by two experts. It was shown that ERI can be successfully used to measure the edge-preserving ability of the filter and can give significant information when conventional metrics fail to assess the image quality obtained with filtering. Future work is needed to evaluate various filters in clinical practice along with segmentation, feature extraction, and classification.