| Brain age is a biomarker to measure the degree of brain aging.Abnormal brain development and aging will result in the deviation between brain age and individual’s real age,which can objectively reflect the degree of brain aging.Magnetic resonance imaging of brain structure is a high-resolution three-dimensional image.Using imaging technology to predict brain age from magnetic resonance imaging is helpful to discover brain diseases such as brain injury,schizophrenia and Alzheimer’s disease.Therefore,the study of brain age based on brain MRI data has great value in medical diagnosis and treatment.Traditional brain age prediction methods extract manual features from brain magnetic resonance images to predict brain age by classification or regression.Its process depends on complicated workflow such as organization segmentation and feature selection.The prediction results are affected by manual features,and the generalization ability of the model is not strong,and the prediction error is large.Some of the existing deep learning brain age prediction models only contain simple convolution layer and pooling layer,so the prediction accuracy is poor and the fitting ability is weak.In order to solve these problems,this paper proposes several brain age prediction models based on deep learning using structural magnetic resonance images as data sets.The specific work of this paper is as follows:(1)The traditional machine learning brain age prediction workflow is complex,and the existing deep learning model has simple feature extraction and poor prediction effect.This paper proposes a three-dimensional convolutional network model based on attention mechanism and bilinear fusion.Through comparative experiments on existing data sets,the proposed model has achieved good prediction effect,which verifies the feasibility and effectiveness of integrating bilinear fusion and attention mechanism into the model.(2)The gray matter(white matter)is used as the model data set alone,which results in the loss of the characteristic information of brain age carried by the white matter(gray matter),and the spatial attention cannot model the dependence relationship between channels.In this paper,a prediction model of SE-Res Net brain age based on two-channel input is proposed.The senet is embedded into the residual module,and the attention mechanism technology in channel dimension is introduced to obtain the key information affecting brain age and suppress invalid information.At the same time,the feature fusion between channels is better to improve the prediction accuracy.Through comparative experiments,the feasibility and effectiveness of SE-Res Net model in brain age prediction task were verified.(3)This paper implements a medical assistant diagnosis system,and applies the brain age prediction model proposed in this paper to the back-end system of the system.The specific work includes: system requirements analysis,logical architecture design,database design and related testing work of the medical auxiliary diagnosis system. |
