Research And Application Of Convolutional Neural Networks In The Classification Of Breast Masses

For women,Breast cancer is one of the most common malignancies and the incidence of breast cancer has been at the first place in all female malignancies with possibility of around 10%.While its lethality is more than 40%.Today there is no positive prevention of breast cancer.So early diagnosis and timely treatment are the only ways to improve the survival rate.However,problems such as breast mass size and contrast are quite different and highly sensitive to Artifacts and surrounding glands greatly affecting the accuracy of the diagnostic system.So breast lumps research still has some difficulties.Based on the research of Breast cancer computer-aided diagnostic system,this paper has deeply studied application problems of convolution neural network model used in breast mass classification.The main research work are as follows:(1)Breast mass image denoising.In order to eliminate the effect of noise on mammography as much as possible and ensure the edge of the breast mass information meanwhile,this paper analyzes the characteristics of noise in mammography,compares a variety of image denoising algorithms and tested on mammography.According to the experimental results,we finally determine to use wavelet algorithm to remove image noise.(2)Regional division of tumor and morphological treatment.In the segmentation of breast masses,We used Otsu algorithm to segment mammography and extract the regions of interest.Then we use morphological knowledge to do expansion and opening and closing operations to the regions of interest.Through the above operations we can save the edge information of the lump image,eliminate the holes inside the mass and get the final lump image.(3)Mass classification.The paper chooses to use the method of migration learning and we use large-scale deep convolution neural network in Breast mass distinction between benign and malignant.We do analysis and research to convolution neural network model structure and its training process.We try to use the method of migration learning to fine-tune GoogLeNet and AlexNet that trained done on a natural image set on the mammogram image.The fine-tuned model achieved a distinction between benign and malignant breast masses based on convoluted neural networks.In this paper,we also compare several models like shallow convolution neural network directly trained from breast images and manual setting features modles.The results also show that depth convolution neural network model based on migration learning has great advantages in breast mass distinction between benign and malignant.