Adaptive whole-brain dynamics predictive method: relevancy to mental disorders

Neuropsychiatric disorders are highly prevalent worldwide, significantly impairing cognition, emotional regulation, and social functioning. However, their diagnosis still largely relies on subjective evaluations and lacks stable neuroimaging biomarkers. This study centers on the Landau-Stuart (LS) oscillator model, which simulates BOLD signals by modulating global coupling strength (G) and bifurcation parameters (a) to capture the dynamic evolution of brain regions under varying states of consciousness and pathological conditions.

To improve individual-level adaptability and parameter fitting stability, the research team led by Dr. Junjie Jiang and Dr. Zigang Huang from the School of Life Science and Technology at Xi’an Jiaotong University conducted extensive simulations using synthetic networks and data. They proposed a novel, generalizable, and adaptive whole-brain dynamic prediction framework. By incorporating personalized initialization strategies, variable learning rates, feature-specific gradient modulation, and an approximate loss function combined with gradient adjustment mechanisms, the model substantially enhanced its ability to reconstruct subject-specific BOLD dynamics. This approach effectively overcomes the convergence and accuracy limitations inherent in traditional modeling techniques.

Extensive simulations and empirical analyses confirmed that the proposed method delivers robust and generalizable parameter estimation performance across individuals. The method was further applied to two large-scale real-world fMRI datasets—Major Depressive Disorder (MDD) and Autism Spectrum Disorder (ASD)—enabling precise reconstruction of individual-level brain dynamics. Results showed that the bifurcation parameters estimated by the proposed model more accurately captured resting-state BOLD features compared to conventional methods. Furthermore, the approach achieved high classification accuracy in both MDD subtyping and diagnosis, and in distinguishing ASD patients from healthy controls—markedly outperforming traditional functional connectivity-based models.

Subsequent analyses revealed significant regional differences in the hippocampus, supplementary motor area, cingulate cortex, insula, and precuneus between healthy and pathological states. Notably, bifurcation parameters in the thalamus and parietal cortex were significantly correlated with HAMD and ADOS scores, respectively, suggesting these regions may play key roles in emotional and social dysfunction and hold potential as neurobiological biomarkers.

Looking ahead, future work may focus on refining the model's theoretical foundations and integrating it with structural connectomics, time-varying modeling, and graph neural network techniques. Such advancements are expected to enhance its physiological interpretability and predictive capacity, facilitating translation into clinical diagnosis, feedback systems, and personalized neuromodulation strategies.

Sources: https://spj.science.org/doi/10.34133/research.0648