Highly personalized modeling of Stimulation-driven brain network responses for next-generation neuromodulation

We present an advanced computational pipeline designed to assess the physiological effects of non-invasive brain stimulation (NIBS) using highly personalized models of neuronal network dynamics tuned to replicate the functional characteristics of individual brains. Starting with structural magnetic resonance imaging (MRI) and diffusion tensor imaging (DTI) data, the pipeline leverages an advanced convolutional neural network to generate subject-specific 3D models of the head and brain with tissue-specific dielectric properties and DTI-based heterogeneous maps of anisotropic brain conductivity. These models are used for personalized dosimetric NIBS exposure assessments performed in Sim4Life, wherein electromagnetic (EM) field distributions are calculated for specific stimulation configurations and also used to compute lead field matrices for a 10-10 electroencephalography (EEG) system. Subsequently, the dynamic brain network response to stimulation is simulated using a spatially co-registered graph of neural mass models (NMMs) implemented in The Virtual Brain open-source brain modeling platform, with connection strengths inferred from DTI-derived structural connectivity, and EM field-brain network coupling achieved via the “λ·E” model. The pipeline uses Bayesian optimization techniques to fit NMM parameters against subject-specific EEG recordings, enabling the individualized prediction of neural responses. Initial applications of the pipeline replicate the expected stimulation-driven spectral shifts in EEG activity and brain network synchronization. This methodology shifts the focus from mere EM exposure optimization to the analysis of large-scale network dynamics, opening the door to the development of advanced, closed-loop stimulation paradigms that drive the brain towards activity patterns associated with health. The pipeline is implemented on o2S2PARC, an open, cloud-based infrastructure for collaborative, FAIR (findable, accessible, interoperable, reusable), and reproducible computational life science.