Research On Prediction Method For Children And Adolescents' Brain Age Based On Deep Learning

Brain magnetic resonance imaging provides the possibility to evaluate the maturity of brain development with its high spatial resolution and high density resolution.However,the maturity of children and Adolescents' brain development is difficult to evaluate empirically by imaging physicians and must be calculated based on computer quantitative measurements The development of healthy human brain is an extremely complicated process in children,adolescents,and early adulthood,which is reflected in the heterogeneity of the order and patterns of tissue development in different regions of the brain.Studies have found that the white matter(WM)volume gradually increases with age in children and adolescents,while the gray matter(GM)volume decreases with age,and the development trends and rates of different brain regions vary.The potential pattern changes of children and Adolescents' brain anatomy during development provide scientific theoretical basis for designing artificial intelligence models to predict brain ageThe anatomical structure of the brain changes with age,and these changes may be related to functional deterioration and neurasthenic diseases.To be more effective in treating disease,researchers must fully understand individual differences in brain aging.In recent years,there have been some personalized techniques for predicting brain aging based on machine learning,such as epilepsy and autism.More importantly,recent research suggests that brain aging is associated with physical and cognitive aging and higher risk of death.With the rapid development of deep learning technology,deep learning technology has been widely used in medical image fields,such as medical image registration,medical image segmentation,and super-resolution reconstruction.Similarly,deep learning can be applied to the task of predicting brain ageIn this article,we use three public data sets for children and adolescents' brain MRIs CMI-HBN,ABIDE,ADHD200.The quality control of these data was performed by multiple neuroimaging radiologists.Through strict and effective data preprocessing process,and finally used these high-quality data to train the deep learning brain age prediction model of this study.Experiments show that the deep learning model proposed in this paper has achieved good results in the brain age prediction task,which can help doctors and reduce the workload of doctors.The specific brain age prediction research work includes the following parts1.This paper proposes an end-to-end 3D convolutional neural network cerebral age prediction model with jumper connection.The 3D model can effectively utilize the context information of the T1 brain MRI space.The addition of hopping connection increases the depth of the network.It allows the network to learn deeper features2.This paper proposes two training strategies for deep learning brain age prediction tasks 1)Gender learning,adolescent children in adolescence,men and women have different brain development trends,so this paper proposes a gender learning strategy to strengthen the influence of gender characteristics on the model,which can make the model learn more targeted characteristics;2)Stratified sampling,because the age distribution of data is not uniform,and the amount of data collected in some age groups is relatively small.In order to ensure good prediction results for all age groups,this paper proposes a stratified sampling strategy to ensure data of different ages.The weights drawn by the model are equivalent3.In previous studies,the brain age estimation model basically used mean square error(MSE)as a loss function because it is easy to solve and smooth.Due to the limitations of the mean square error loss itself,singular values are apt to occur.In order to reduce the singular value of the model in the prediction process and enhance the robustness of the model.This study proposes an outlier constraint loss for brain age prediction tasks.The experimental verification shows that outliers of the prediction result of the model trained with this outlier constraint loss are significantly reducedIn conclusion,based on the prior knowledge that children and Adolescents' brain gray matter and white matter develop differently with age,this study used gray matter brain maps and white matter brain maps to build a deep learning model to evaluate brain age.The network architecture and strategies proposed in this article can effectively train deeper networks.We tested the effectiveness of the model on two public datasets,and the experimental results show that compared with the latest method,for the prediction of brain age in normal children,our method has obtained a more accurate prediction of brain age than previous studies result.There was a significant statistical difference in the prediction results of brain age in children with autism or ADHD compared with those in the control group of normal children and Adolescent.In this study,it was found that the brain development of children with mental disorders lagged behind that of the control group of normal children and Adolescent by about 0.5 years.