Brain Age Prediction And Brain Disease Classification Based On 3D Convolutional Neural Network

With the increasing aging of the population,the problems of brain aging and brain diseases have attracted increasing attention.The brain age prediction study based on3 D convolutional neural network can quantify and analyze the brain aging of individuals,which can help to detect brain aging and brain disease risk early and propose effective intervention measures.Traditional brain age prediction models suffer from large prediction errors,while the existing 3D convolutional neural network brain age prediction models are complex in structure,insufficient training sample size and prone to overfitting.Therefore,current research is devoted to improving the accuracy of brain age prediction.And disease-specific related studies often face the problems of insufficient sample size and low classification accuracy.Therefore,whether transferring brain age prediction models can be helpful for disease research deserves further investigation.In order to solve these problems,this thesis develops a brain age prediction model with simple parameters and excellent model performance based on 3D neural network using structural MRI images as the dataset.The specific work in this thesis is as follows:(1)Traditional machine learning brain age prediction has a complex workflow that relies on tissue segmentation and feature selection,while existing deep learning has simple models with simple extracted features and poor prediction results,complex models with strong learning ability but large model parameter size and easy overfitting,and two-dimensional convolutional neural networks with loss of three-dimensional neuroimaging contextual information.In this thesis,we propose to build a threedimensional convolutional network model for brain age prediction,and through training and testing on a public dataset,the model achieves good prediction results,and increasing the sample size and introducing integrated learning can further improve the model performance.The test results based on Alzheimer’s disease patients show that brain age prediction can effectively assess brain age bias,and validate the feasibility and effectiveness of the brain age prediction model in brain disease classification studies.(2)Aiming at the problems of small sample size and low accuracy of disease classification in disease-specific research,this thesis proposes to apply the brain age prediction model to two independent research directions,Alzheimer’s disease and schizophrenia,with the help of a transfer learning approach.In Alzheimer’s disease research,the brain age prediction model is first trained based on a public dataset of subjects over 50 years old,and its performance in the test set is evaluated;subsequently,a fine-tuning approach is used to transfer the model to an Alzheimer’s disease classification task based on the public dataset.On the other hand,in the schizophrenia classification study,a brain age prediction model was trained based on a public dataset of subjects aged 10 to 50 years and applied to a schizophrenia classification study with a self-constructed dataset.Statistical comparison of direct classification accuracy and transfer classification accuracy in these two studies reveals that the performance metrics of transfer classification are better than direct classification.This result validates the effectiveness of the transfer learning-based brain age prediction model in improving the performance of brain disease classification and further confirms the intrinsic link between brain age and brain disease.