Research On Parkinson's Disease Classification Algorithm Based On Deep Convolutional Neural Networ

Parkinson’s disease(PD)is the most common neurodegenerative disease after Alzheimer’s disease,and its incidence is increasing significantly,which greatly affects the daily life of patients.Parkinson’s disease classification is one of the research hotspots in the field of computer-aided diagnosis,which aims to improve the accuracy of disease diagnosis through computer technology.However,current Parkinson’s disease classification faces two major challenges.First of all,Parkinson’s disease image data is limited,and the acquisition and labeling work is time-consuming and laborious.Secondly,Parkinson’s disease images contain complex pathological information,which requires guidance from professional doctors,and it is difficult to judge and classify based on visual similarity.With the development of neuroimaging technology,magnetic resonance imaging(MRI)data can clearly display the brain structure.Therefore,based on MRI images,this thesis studies a PD classification algorithm based on depth convolution neural networks.The specific works are as follows:(1)A PD classification algorithm based on multimodal features and fully connected classification networks is proposed.The method solves the problem of insufficient resolution and poor stability of a single feature in image classification by using multimodal features.Firstly,the texture features of T2 MRI slices are extracted using gray-level co-occurrence matrix and local binary pattern;Secondly,the morphological features of T1-weighted MRI images were extracted using voxel-based morphometric measurements;Next,use the Relief F feature selection algorithm to sort according to the size of the feature weights,and select important features for classification;Finally,an end-to-end full convolutional classification network for PD diagnosis is proposed.The experimental results show that this method not only effectively improves the classification performance of PD,but also discovers the pathogenic brain regions associated with PD,providing a biomarker for studying the development of PD diagnosis.(2)A PD classification algorithm based on the combination of texture features and deep convolutional neural network is proposed.This method combines traditional texture features and depth features to solve the classification problem of small sample images.Firstly,a method based on deep convolutional neural network is proposed and used to classify PD;Secondly,traditional texture features of images are extracted using local binary patterns;Next,the texture features and depth features are fused,and the principal component analysis method is used for dimensionality reduction;Finally,a Stacking-based integrated classification method was proposed.The experimental results show that texture features and depth features are complementary to each other,which helps to improve classification performance and obtain better classification accuracy.(3)A PD classification algorithm based on adaptive weighted attention deep convolutional neural network is proposed.This method aims to address the problem of insufficient attention to disease-causing regions in medical images in deep learning network frameworks.Firstly,to achieve effective feature transfer and gradient descent,a deep convolutional neural network framework based on dense block is designed.Secondly,an adaptive attention feature processing algorithm is proposed,which is mainly used to extract features of different scales and enable better fusion of features.Finally,Dropout layer and Soft Max layer are added to the network structure to obtain good classification results and rich and diverse feature information.The experimental results show that the proposed method can make the deep learning network pay more attention to areas with pathogenic information,thereby improving the accuracy of PD classification.