Detail-preserving Despeckling Algorithm Based On Phase Asymmetry For Breast Ultrasound Segmentation

Breast cancer is the leading cause of cancer death among women,and the key to its treatment is early diagnosis.Breast ultrasound examination is one of the most ideal diagnostic methods for screening breast cancer.However,the diagnosis is highly dependent on the doctor’s reading of breast ultrasound images,which takes a long time and requires a lot of work.In order to improve the diagnostic efficiency,a computer aided diagnosis(CAD)system for breast ultrasound has emerged.Because the unclear edges and the irregular shapes of tissues are important features of breast cancer,therefore,the edge segmentation of breast ultrasound is the most important part of the CAD system.However,due to the extremely unclear characteristics of breast ultrasound tissue signals,and the inherent multiplicative noise of ultrasound imaging,breast ultrasound segmentation is still most challenging engineering problem.Aiming at this problem,this paper designs a speckle noise removal and tumor segmentation algorithm for breast ultrasound images.Owing to the speckle noise of breast ultrasound image seriously reducing the segmentation accuracy,this paper first proposes a detail-preserving despeckling algorithm based on phase asymmetry to remove speckle noise.Specifically,phase asymmetry based on local phase information is used to detect edges in ultrasound image,and the edge significance of each pixel is obtained,ranging from 0(no edge significance)to 1(predicting this is a very significant edge point).The fractional-order anisotropic diffusion(FAD)model can effectively remove noise in the smoothing regions,while the fractional-order fractional total variation(FTV)model can protect image edges well.The FAD model and the FTV model are combined to achieve a balance between noise reduction and edge protection.Overcoming the shortcoming of the traditional FAD model using only gradient information to adjust its diffusion coefficient,causing low contrast edges to be damaged,this paper proposes to adopt both gradient information and the edge significance to adjust the diffusion coefficient,which can protect low contrast edges.In addition,overcoming the shortcoming of traditional FAD noise reduction model reducing edge contrast.The proposed despeckling algorithm further adopts adaptive fractional order to specifically enhance edge features.Based on the above design,the proposed despeckling algorithm protects image edges while removing speckle noise.Other details of the image features are also satisfactorily protected due to the advantage of edge preservation by our despeckling algorithm.This paper further proposes an improved breast ultrasound segmentation algorithm,which first performs preprocessing steps such as noise reduction,contrast enhancement and grayscale attenuation.The noise reduction method uses the above-mentioned ultrasound despeckling algorithm to remove speckle noise,the contrast enhancement is to improve the contrast of tumor regions by adopting contrast-limited adaptive histogram equalization,and the gray scale attenuation uses a Gaussian function to attenuate the gray of non-tumor area pixels.Then,this paper improves a marker-controlled watershed algorithm to effectively segment tumor regions in breast ultrasound images in obtaining a series of potential tumor edges.Finally,the proposed segmentation method uses the average radial derivative to evaluate the potential edges for the final tumor edge selection.The performance comparison experiments between the proposed ultrasound despeckling algorithm and other state-of-the-art despeckling algorithm are carried out.The experimental results show that the proposed ultrasound despeckling algorithm can not only remove speckle noise but also protect ultrasound image features,especially the detail features.Compared with other breast ultrasound segmentation algorithms,the breast ultrasound segmentation algorithm has the highest tumor segmentation accuracy.