Research On Breast Cancer Pathological Diagnosis Method Based On Deep Learnin

Breast cancer is one of the most dangerous female diseases.Its diagnosis process involves complex pathological knowledge and diagnostic experience.By observing the pathological slice images of patients,diagnostic physicians with high diagnostic skills can clearly determine the type of cancer in patients.However,the cultivation of excellent diagnostic physicians will consume enormous human costs and medical resources.At the same time,the main body of artificial diagnosis is human,which determines that the accuracy of diagnosis will be affected by factors such as knowledge accumulation,experience level,fatigue,work mood,and so on.With the development of artificial intelligence,deep learning has gradually become an efficient and reliable diagnostic technology in the field of intelligent breast cancer pathology picture classification.In order to solve the problem of multi classification in breast cancer histopathological images,we designed a convolutional neural network that integrates multiple improved attention mechanisms,and trained and tested it on the Break His dataset.The network focuses on multi-scale feature information and cancerous tissue details in breast cancer histopathological images.To further improve network performance,we also proposed two joint construction strategies for multi convolutional neural networks and constructed multiple joint networks based on these strategies.The main contents and contributions of this article are as follows:(1)Three attention mechanisms have been improved to form four attention modules with channel reduction effects,and a Dense Attention Combined Network(DAUNet)has been proposed(2)Two model combination methods based on professional enhanced classification strategies are proposed,and their principles are elaborated.Their performance advantages are analyzed.Six enhanced convolutional decision tree models are constructed.This combined convolutional network is used to further improve the final classification performance of the network.(3)Aiming at the problem that a single professional reinforcement classifier involves a small amount of training data,a self migration learning method is proposed to further exploit the performance advantages of professional reinforcement classification strategies for enhancing convolutional decision trees.The final test experimental results show that the three-type decision tree network construction has the highest accuracy and high running speed for breast cancer pathology recognition,with 99.81% accuracy in binary classification and 96.01%accuracy in octal classification,and the network has high performance reliability and application potential,and the two types of professional strengthening strategies show strong interpretability and operability.