Research On Detection And Recognition Of Breast Masses Based On Convolutional Neural Network

The improvement of automatic detection and recognition accuracy of medical image is the key to improve the work efficiency of pathologists.As an important part of automatic detection and recognition of medical image,the detection and recognition of breast molybdenum target mass is the focus and difficulty of current research.Improving the accuracy of detection and recognition of breast molybdenum target mass is of great significance to promote the automatic detection of medical image Righteousness.Convolutional neural network,as a working mode of bionic vision mechanism,can learn very small image features by sharing information among its unique characteristic parameters and sparsity of connections between various layers under strong supervised learning,but there are still many problems in the application of medical image,which is also due to the complexity of image information and organization area of medical image itself It is caused by some reasons,such as subtle and special circumstances,which need to be combined with the features of multiple views.This paper makes the following improvements for the model:(1)Based on the classical full convolution neural network,a detection algorithm based on active learning strategy and pyramid pooling is proposed.On the one hand,by introducing active learning training strategy,the workload of data annotation is reduced;on the other hand,because the tissue difference on molybdenum target image is slight and the difference of mass size between different patients is very large,the generalization ability of the algorithm is enhanced by pool splicing of different sizes of features extracted from convolution layer.(2)In the actual work flow of doctors,when we want to judge,we need to combine multi view to propose a tumor matching algorithm based on dual view.Through vgg16 network model,the mass features of different views of the same patient are extracted,high-dimensional splicing is carried out,and then the physiological feature information such as the mass size and distance information is combined to determine whether the two masses are the same mass.Through the improvement of this method,we can simulate the actual work flow of doctors.On the one hand,if there is a false detection in one side view,we can see whether there is a matching mass in the other view to reduce the false detection rate.On the other hand,the subsequent classification of benign and malignant tumors can find the matching mass combination analysis when the characteristics of one side view are not obvious.The experimental results show that the detection accuracy of the algorithm based on active learning and pyramid pooling is more than 90%.The two view mass matching algorithm can also achieve a FPI of 0.14 in matching recognition.