SD-ResCNN: Brain Age Prediction Model Based on a Combination of Static and Dynamic Functional Connections

Functional connectivity (FC) of the brain is crucial for understanding how the brain processes information and executes cognitive functions through complex neural networks. FC can be categorized into static functional connectivity (sFC), which describes the brain network's topological structure, and dynamic functional connectivity (dFC), which reflects the variability of connectivity over time. Given the growing interest in jointly utilizing both types of FC for brain-related feature analysis, this study proposes a novel brain age prediction model-SD-ResCNN-based on a residual neural network (ResNet) that integrates static and dynamic functional connectivity. For functional magnetic resonance imaging (fMRI) data, sFCs were constructed using Pearson correlation coefficients, while dFCs were generated via the sliding window approach. The model design leverages the superior image processing capabilities of ResNet by incorporating 2D and 3D convolutional blocks based on the ResNet-18 model and residual blocks (ResBlocks) for different types of FC. Additionally, two diffusion connection mechanisms, S2D and D2S, were introduced to enable effective feature fusion in a dual-pathway structure. Comparative experiments demonstrate that the proposed SD-ResCNN model achieves outstanding brain age prediction performance on the FCP Classic dataset of healthy adults, with dual-pathway integration providing a notable advantage (achieving an accuracy of 75.00% within a 3-year error margin and 85.44% within a 5-year error margin). This study not only enhances the accuracy of brain age prediction but also offers a powerful tool for further exploring brain functional changes.