Personalized Deep Brain Stimulation: AI-Driven Fusion of Multi-Modal Imaging and Finite Element Analysis for Predictive Electrode Field Modeling

Abstract Precise targeting in deep brain stimulation (DBS) is challenged by individual neuroanatomical variability and postoperative brain shift, often compromising therapeutic efficacy in movement disorders like Parkinson's disease. Conventional atlas-based approaches lack patient-specific models to predict stimulation field interactions with target nuclei (e.g., STN, GPi). Here we present an integrative computational pipeline combining multi-modal imaging with biophysical simulation to enable personalized DBS planning. Our framework leverages: 1) multi-modal registration (advanced normalization tools, ANTs; or statistical parametric mapping, SPM) with subcortical brain shift correction, significantly reducing electrode placement error; 2) AI-driven electrode reconstruction (PaCER) achieving 0.4 ± 0.1 mm contact localization accuracy; and 3) patient-specific finite element modelling (iso2mesh/TetGen) predicting confined stimulation fields. Validated on clinical imaging data (pre-op T1/T2 MRI; post-op CT), the pipeline generated anatomically grounded electrophysiological models in < 35 min per patient, demonstrating computational accessibility. The resulting 1.3 ± 0.4 mm STN targeting precision and field confinement predictions establish a foundation for physics-informed DBS programming. This work bridges surgical planning with adaptive neuromodulation by translating patient anatomy into dynamically queryable stimulation profiles, paving the way for closed-loop systems responsive to individual neuroelectric landscapes.