Mass Detection Of Mammography Based On Deep Kernel Hashing

According to the survey of cancer,breast cancer ranks first in the new cases of female malignant tumors with a rising trend,and breast cancer has already become the number one killer for women’s health.Early detection and diagnosis is the most effective way to prevent breast cancer.In the face of the complexity of the large-scale breast cancer census,considering the differences between doctors’ abilities and judging criteria,the computer aided detection(C AD)system is designed and applied.The goal of C AD system is to detect suspicious breast mass area,to assist quickly doctors to make diagnosis decision,to reduce workload and to improve the detection rate of breast mass.Because of the complexity and ambiguity of the mass in mammography,it is very important to design a mass detection system with high real-time and high detection rate.Therefore,according to the pathological characteristics of breast masses,combining with artificial intelligence,machine learning and deep learning,this paper realizes the detection and display of suspicious mass area,and then scientifically assists doctors to make better diagnoses.Based on kernel hashing,the characteristics of edge,texture and gradient information and convolution neural network CNN,this paper deeply studies the suspicious masses detection method for mammography,and the main works are summarized as follows:First of all,a fast Multiple-Feature kernel Hashing for Mass Detection method is proposed.After the research on various hashing algorithms,we integrate histogram of oriented gradient(HO G),hierarchy-weigh Gist and convolution neural network CNN feature into the kernel space of kernel-based supervised hashing(KSH),we design a universal detection method suitable for different mass types and large-scale data.This method has a strong representation of mass information,and make full use of label information to realize supervised learning.Due to the efficiency of hashing,the proposed method make the real-time CAD system come true and improve the practicability of detection systems.Secondly,this paper proposed a multilayer kernel hashing algorithm framework.The data space is reduced dimension by the kernel space in the KSH algorithm,a lot of useful information is lost in the process of dimensionality reduction,which is leading to the reduced coding precision.To solve this problem,this paper constructs a multi-level nonlinear mapping framework based on supervising information and the loss function of the input image pair.Minimizing the loss information during the mapping process which is from the original data or eigenvector to the learning binary encoding.The method proposed in this paper improves the accuracy of classification and t he coding precision while maintaining the similarity of data.Finally,this paper proposed a fast breast mass detection based on deep kernel hashing method.In order to represent mass in a more accurate way,this paper constructs the deep hash supervised learning framework based on the convolution neural network CNN model and the deep kernel hashing algorithm framework.With powerful mass information expression ability,learning ability and classification ability,the framework effectively conquers the nonlinear characteristic of mammographic classification.The overall performance of our detection system will be improved by this way.The experimental results show that the mass detection of mammography based on deep kernel hashing method proposed in this paper can distinguish masses from normal tissues more precisely and improve the mass detection accuracy while maintaining a low false positive rate at the same time.