| Breast cancer is one of the most common cancers in women,which seriously affects women’s physical and mental health.Computer aided diagnosis(CAD)model can effectively assist pathologists in clinical diagnosis,realize early detection,early diagnosis and early treatment of patients,and lay an important foundation for improving people’s quality of life.The existing research work is faced with the following problem.On the one hand,facing the scaricity of samples,the model is easy to overfit;on the other hand,single category features are often used,but the cross-modal pathological semantics between heterogeneous image features are not fully exploited and utilized,and the importance of different features in breast cancer image recognition is ignored.Therefore,a breast cancer image recognition model based on deep learning and fine-grained feature selection is proposed.The main work is as follows:Breast cancer image recognition model based on convolutional neural network:Aiming at the problem of insufficient feature learning,based on transfer learning method,the pretraining convolutional neural networks(including ResNet,VGGNet and DenseNet)are fine-tuned with breast cancer image dataset to generate high-level semantic features to depict the lesion area in breast cancer image and complete breast cancer image recognition.The experimental results show that: in all convolutional neural networks,ResNet model performs best,especially in the high-quality INbreast dataset.However,due to only using single class features,the overall accuracy of convolutional neural network needs to be improved.Breast cancer image recognition model based on feature optimization algorithm: in view of the problem that there is a lot of noise in the feature and the feature dimension is too high,which leads to overfitting of the model,a multiview effective region gene optimization(MvERGS)algorithm is designed,in which multiview is introduced to remove the noise information from different angles,so as to improve the feature discrimination and complete the recognition task.The experimental results show that: after using MvERGS algorithm to perform feature component optimization,the feature discrimination is further improved,the recognition performance on both datasets is improved,and the overfitting problem of the model is restrained to a certain extent.In addition,due to the reduction of feature dimension,the practicability of the model has been improved.Breast cancer image recognition model based on fine-grained feature optimization algorithm: for the underutilization of cross-modal information contained in heterogeneous features,RCA model is designed: feature optimization(R)is performed based on MvERGS algorithm;cross-modal correlation analysis(C)algorithm is used to mine cross-modal pathological semantics among heterogeneous features to better characterize the lesion area;and GBDT algorithm is used to adaptively select the most concise but efficient feature(A)to complete breast cancer image recognition.The experimental results show that: after the implementation of fine-grained feature selection algorithm,the discriminability of image features is significantly improved,and the overall recognition performance of RCA model is better than that of mainstream baseline.At the same time,the real-time performance of the model is improved,and it has certain practical value. |
