Macaque Brain Age Prediction Based On Functional Brain Characteristics Using Deep Learning Methods

Brain-predicted age,which is predicted by regression methods based on the features contained in brain image data,is called brain age.Previous studies have shown that establishing a stable and reliable prediction model of brain age based on brain image data of normal individuals and calculating the difference between brain age and chronological age can be used as an effective biomarker to distinguish different types of neurodegenerative diseases.Most of the existing studies have been conducted on the modeling and prediction of brain age based on human brain data.However,as one of the most important animal models,there are few studies on the prediction of brain age of macaque.Therefore,it is not clear whether the research on brain diseases based on the difference between brain age and chronological age can be carried out on macaque.In addition,the existing brain age research methods only focus on the single scale or single modality functional characteristics,ignoring the brain information contained in the characteristic data of different scales and different models,so the accuracy of brain age prediction needs to be further improved.To address the limitations of the above studies,the resting-state functional magnetic resonance imaging data of 450 macaques provided by the publicly released PRIMat E Data Exchange(PRIME-DE)database are used to perform two studies on brain age prediction from two perspectives including multi-scale features fusion and multimodality features fusion in this thesis.The first study is to predict macaque brain age based on multi-scale resting-state functional connectivity patterns using the proposed Multi-Branch Vision Transformer(MB-Vi T)to fusion different scale functional connectivity characteristics.The second work is to predict brain age of macaque based on multi-modality features using the proposed Multi-Head Attention Graph Convolutional Neural Network(MH-AGCNN)and Multi-Branch Guided Attention Neural Network(MB-GANN)to identify multiple meaningful spatial patterns of resting-state network and effectively integrate spatial and temporal patterns to predict macaque brain age.The experimental results show that compared with the baseline models,our proposed models have better predictive ability and stability of macaque brain age.In addition,we conduct extensive ablation experiments and permutation tests on the proposed deep learning models,which confirm the effectiveness of the proposed methods and statistical significance of the predicted results.We further identify the brain functional connectivities and brain regions that contribute most to the age prediction.In summary,our proposed deep learning methods in this thesis provide new insights and perspectives for the study of macaque brain age prediction,and also lay a foundation for the further study of brain diseases based on the difference between brain age and chronological age in animal models.