Research On Denoising Technology For Breast Ultrasound Images

Improving the quality of breast ultrasound images is of great significance for clinical treatment of breast ultrasound.Low noise and high contrast images can greatly improve the accuracy of doctors in diagnosing the condition.However,due to the principles of ultrasound image acquisition and the influence of imaging equipment,the collected ultrasound images naturally contain a large amount of speckle noise,which appears as granular spots in the image,resulting in low resolution of image details and blurred boundaries,which affects the subsequent feature extraction and lesion segmentation of ultrasound images.Researchers have proposed many methods for removing speckle noise and achieved good results,but there are still some problems,such as insufficient image edge preservation and high algorithm complexity.In this context,this thesis focuses on the problems in speckle noise removal and studies denoising methods for breast ultrasound images,aiming to effectively improve the quality of breast ultrasound images.The specific work content is as follows:(1)This thesis improves the Non-subsampled Shearlet Transform(NSST)and proposes a denoising algorithm for breast ultrasound images based on NSST and improved fuzzy.Firstly,by improving the fuzzy algorithm to enhance image contrast;Then,NSST is used to decompose the image into low-frequency and high-frequency parts,linear transformation is performed on the low-frequency part to adjust the overall contrast of the image,and an adaptive threshold model is used to remove noise in the high-frequency part of the image;Finally,the processed high and low frequency parts are subjected to inverse NSST to obtain denoised images.The experimental results on the self built breast ultrasound image set show that compared to the comparison algorithm selected in the experiment,the algorithm has the best speckle suppression and mean preservation index,and the edge preservation index has increased by at least 3.87%,showing good denoising effect.(2)In traditional Block-matching and 3D Filtering(BM3D)algorithms,the block-matching process requires similarity judgment on all image blocks within a given neighborhood range of the reference block,resulting in low algorithm efficiency.To address this issue,this thesis proposes an improved BM3 D breast ultrasound image denoising algorithm that integrates superpixel segmentation(hereinafter referred to as Superpixel-BM3D).Firstly,a superpixel segmentation method based on Density-based Spatial Clustering of Applications with Noise(DBSCAN)is introduced to perform superpixel segmentation on the original image to obtain the corresponding superpixel label matrix;Then,using the superpixel label matrix to guide the block matching process in the BM3 D algorithm can reduce the search time for similar blocks,and on the other hand,superpixel labels also provide auxiliary information for similar block measurement,improving the accuracy of block matching;Finally,improve the hard threshold filtering in the BM3 D algorithm and further enhance the denoising effect by estimating local noise parameters.The experimental results on a self-built breast ultrasound image set showed that compared to the traditional BM3 D algorithm,the image processed by this algorithm has an average increase of 1.75% in equivalent number of looks,an average increase of 2.56% in edge preservation index,and an average reduction of 51.62% in algorithm processing time,taking into account both denoising effect and algorithm processing time.(3)The Superpixel-BM3 D algorithm proposed in this thesis has improved denoising performance and operational efficiency compared to traditional BM3 D algorithms.However,speckle noise in breast ultrasound images directly affects the accuracy of superpixel segmentation,thereby affecting the selection of similar blocks in the block matching process,resulting in BM3 D still having shortcomings in balancing noise removal and image edge preservation.Therefore,this thesis first uses the Multi-Filter Discrete Fractional Fourier Transform(DFRFT)to enhance the image.Based on the characteristics of the spectrum of breast ultrasound images after DFRFT,two filters are designed.The two filters are used to filter and fuse the images to obtain the enhanced image,and then superpixel segmentation is performed to obtain a more accurate superpixel label matrix.In addition,in order to further reduce the processing time of the algorithm,while ensuring the denoising effect,Manhattan distance is used as a similarity measure.The experimental results on a self-built breast ultrasound dataset show that compared to the traditional BM3 D algorithm,the image processed by this algorithm has an average increase of 5.76% in equivalent number of looks,an average increase of 5.96% in edge preservation index,and an average reduction of 46.27% in algorithm processing time.This reduces the algorithm processing time while balancing denoising effect and image edge preservation.