关键词: 关键词: 功能磁共振成像, 功能连接, 轻量级神经网络, 脑龄预测, 深度学习

Abstract: Abstract: In view of the low accuracy of functional Magnetic Resonance Imaging (fMRI) brain age prediction and the lack of research on the combination of this problem and deep learning, a fMRI brain age prediction model with Lightweight Multi-scale Convolutional Network (LMCN) was proposed. First, the Pearson correlation coefficient of the region of interest (ROI) in fMRI was calculated to obtain the functional connectivity matrix (FC) of the ROI as input; secondly, the number of FC channels was increased to ensure the number of features while reducing the size of the feature map and the multi-scale dilated convolution module (RFB) with the characteristics of human visual attention was used to extract age features; finally, the brain age was predicted through the output of the fully connected layer, and the ablation prediction results of each brain area were calculated to explore the key brain regions affected by the generation of brain age prediction results. Evaluated on two public data sets, E-NKI and Cam-CAN, the memory required for LMCN parameters is 2.30MB, which is 60.3% and 43.5% less than MobileNetV3 and ShuffleNetV2. In terms of prediction results, in the E-NKI data set, the Mean Absolute Error (MAE) is 5.16, the Pearson correlation coefficient (R) is 0.947, and the Root Mean Square Error (RMSE) is 6.40, which is comparable to network-based feature selection combined with the minimum angle regression model. Compared with the method based on the connectome, MAE decreased by 1.34 and R increased by 0.037; in the Cam-CAN data set, MAE was 5.97, R was 0.904, and RMSE was 7.93. Compared with the connectome-based model, R increased by 0.019 and RMSE decreased by 0.64. The results show that while the LMCN has small parameters and is easy to deploy, it can effectively improve the accuracy of fMRI brain age prediction and provide clues for assessing the brain status of healthy adults.

Key words: Keywords: functional Magnetic Resonance Imaging(fMRI), functional connectivity, lightweight neural network, brain age prediction, deep learning