Ultrasound Image-Assisted Diagnosis Of Breast Tumors Based On Feature Fusion

Breast disease is one of the most common diseases in women.Effective examination of breast disease plays an irreplaceable role for doctors and patients.At present,breast ultrasonography is the most commonly used method in the diagnosis of breast tumors,but it is difficult to identify the benign and malignant tumors of small tumors.Because of the subjective factors of doctors,it is easy to miss and misdiagnose.Therefore,to achieve the diagnosis of breast cancer,rich clinical experience is needed.With the development of science and technology,assistant diagnosis of breast tumors by computer technology has become a research hotspot.Aiming at the characteristics of ultrasound images of breast tumors,this paper analyses and studies the assistant diagnosis technology of breast tumors from four aspects: image preprocessing,segmentation,feature extraction and selection,classification and recognition.Because of the speckle noise problem in ultrasound image,a new method of ultrasound image denoising for breast cancer is proposed,which combines anisotropic diffusion model to reduce the iteration times of traditional bilateral filtering,and introduces compensation function and exponential function to improve the weighted kernel function to control the shrinkage speed of Fourier coefficient.The denoising effects of wavelet transform,traditional bilateral filtering and improved algorithm are compared and analyzed by experiments.The results show that the improved bilateral filtering algorithm proposed in this paper is superior to the other two algorithms in peak signal-to-noise ratio,structural similarity and running time.In the process of image segmentation,aiming at the problem of over-segmentation of traditional watershed algorithm,the watershed algorithm based on anisotropic diffusion is used to segment the image.In the pre-processing stage,histogram equalization is used to enhance image contrast and morphology is used to pre-process.In the process of processing,Otsu algorithm is used to calculate threshold to distinguish foreground and background.In the post-processing stage,gray mean is used to merge regions.The segmentation effects of Otsu,traditional watershed algorithm and improved algorithm are compared and analyzed by experiments.The results show that,compared with the other two segmentation algorithms,the improved algorithm reduces the problem of over-segmentation.Aiming at the problem of data redundancy caused by feature dimension in the process of feature extraction,optimization and classification of breast tumors,a new method of ultrasound image classification and recognition of breast tumors is proposed.The proportion of samples to be measured is introduced into the weight formula of Relief-F algorithm,and the improved algorithm is combined with the sequential forward floating method to optimize the 63 extracted features and establish feature data.The K-means algorithm is used to combine the selected features and form different feature vectors to classify breast tumors.The classification effects of BP neural network,random forest algorithm and K-means algorithm on breast tumors were compared and analyzed.The results show that compared with other two classification algorithms,K-means algorithm has the highest classification accuracy.