Analysis And Design Of Breast Cancer Image Recognition System Based On Deep Learning

The digital technology of breast cancer images is widely used in pathological section diagnosis nowadays.Various digital medical imaging models of breast cancer have promoted the development of modern pathological diagnosis technology.It has also been used as an important basis for clinical medical diagnosis,postoperative tracking,disease prediction,teaching and research.After a series of processing and analysis of breast cancer medical images,the feature information of images' deep essence is extracted to assist the diagnosis of pathological slice images by using machine learning and deep learning.It reduces the error and workload of doctors.In addition,it also has an extremely important impact on academic as a result of the wide range of application scenarios.The characteristics of breast cancer medical image are high resolution,high dimension,many kinds and uneven sampling periods.The diagnosis of medical images has been a complex problem.In computer vision methods,researchers have proposed many strong supervised learning algorithms based on object detection and weak supervised learning algorithms based on classification combined with machine learning.On the basis of weak supervised learning methods,researchers have proposed many combination algorithms with high accuracy on fixed data sets,including clustering algorithm,neural network plus support vector machine algorithm,and neural network plus conditional random field model.Model needs to divide the big picture into the small pictures and makes a prediction according to the correlation between the small pictures by using the weak supervised classification.The conditional random field model can be used to correlate the location information of each small pictures effectively.Because of the fixed formula,each picture only takes advantage of the pictures' feature information which is next to them.In view of this shortcoming,this paper uses bidirectional cyclic neural network instead of conditional random field.The algorithm model can learn to find the location and feature relationship of all small pictures by using the Long Short-Term Memory(LSTM)computing units.The main research results are as follows:1.Using the deep learning model plus machine learning model can effectively improve the speed and accuracy of breast cancer medical image recognition and localization.2.The accuracy of the classification is improved and the speed is dropped a little by using the bidirectional cyclic neural network instead of the conditional random field.