Proceedings of the Annual Meeting of the Cognitive Science Society

Brain functional activity exhibits a distinct temporal hierarchy, characterized by the increase in intrinsic neural timescale (INT) from unimodal to transmodal cortices. However, existing large-scale brain network models struggle to accurately simulate this timescale hierarchy. This study developed a heterogeneous dynamic mean-field (hDMF) model by quantifying cortical microstructural differences using T1-weighted/T2-weighted (T1w/T2w) mapping. For model training, we proposed a novel multi-objective expectation maximization (MOEM) algorithm guided by bifurcation theory to achieve precise optimization in high-dimensional parameter spaces. Results demonstrated that the hDMF model significantly outperformed traditional models in fitting functional connectivity and metastable states, while successfully reproducing the gradient distribution of INT. Further application to ADNI clinical data revealed its ability to effectively simulate the abnormally elevated INT in Alzheimer's patients. In summary, this study provides a high-fidelity computational model for understanding the spatiotemporal brain dynamics and demonstrates potential clinical application value in biomarker exploration for neurodegenerative diseases.