Research On Parkinson's Disease Classification Algorithm Based On Transfer Learning And Convolutional Neural Networ

Parkinson’s disease is a common chronic neurodegenerative disease that mostly occurs in middle-aged and elderly people.The disease causes multiple impairments,both motor and nonmotor,that seriously affect the patient’s quality of life.However,the pathogenesis of Parkinson’s disease is still unclear.How to accurately and effectively detect Parkinson’s disease is of great significance for medical intervention and health management.Studies have shown that Parkinson’s disease can cause changes in the brain structure,and magnetic resonance imaging can clearly show changes in the brain structure of patients with Parkinson’s disease,which can help in the early diagnosis of Parkinson’s disease.Therefore,based on the magnetic resonance imaging of Parkinson’s disease,this thesis uses transfer learning technology to classify Parkinson’s disease by improving the convolutional neural network algorithm.The main work as follows:(1)A Parkinson’s disease classification algorithm based on transfer learning and improved VGG16 is proposed.First,the raw data is preprocessed using homomorphic filtering to highlight high-frequency features of brain images.Secondly,a convolutional block attention module is introduced after the convolutional layer of VGG16,and a branch is added in the gap layer of each block to obtain feature information,and all the features obtained by the branch are serially fused to improve the classification performance of the model.Finally,the fine-tuned transfer learning method is used to train the model to improve the generalization ability of the network on medical images.Experiments show that the model has a higher accuracy than other Parkinson’s disease classification algorithms,reaching 95.16%.At the same time,it verifies the importance of transfer learning technology to model training.Using transfer learning can further speed up the convergence speed of the model and significantly improve the classification performance of the model.(2)A Parkinson’s disease classification algorithm based on feature migration and improved Res Net34 is proposed.First,improve Res Net34,embed squeeze-and-excitation modules into each residual unit of the network,and redesign the network structure.Secondly,Google Net is used to extract the initial features of the image,and the improved Res Net34 is used to extract the high-frequency features of the image,and the model is trained through feature migration.Finally,the extracted two sets of features are fused in series,and five machine learning algorithms are used for classification.The results show that the overall classification performance of SVM is better than that of other machine learning classification algorithms,and compared with other studies using SVM classification,the model has a higher accuracy rate.(3)A Parkinson’s disease classification algorithm based on quadratic transfer and dualattention residual network is proposed.First,the residual network is optimized,and the squeezeand-excitation module and the convolutional block attention module are introduced to help the network better capture the correlation between different channels.Secondly,the combination of Center loss and Softmax is used to optimize the loss function and increase the sensitivity of the model to small changes in the input data.Finally,the model is trained using the secondary migration method to solve the problem of low classification accuracy of the training model due to insufficient samples.The experimental results show that compared with the one-time migration model,the classification accuracy rate of the secondary migration model has increased by 3%,and the classification accuracy rate of the proposed model has reached 96.87%.Compared with the advanced classification model,it shows certain advantages.