The thalamus is a central hub that interfaces with widespread cortical and subcortical nodes. Thalamic deep brain stimulation (DBS) offers a principled strategy for distributed cortical modulation: since distinct thalamic nuclei project to spatially segregated cortical territories, stimulation at a single thalamic site can influence multiple cortical nodes. Realizing this potential requires accurate subject-specific estimates of directed thalamocortical effective connectivity (EC) and a computational framework for optimizing stimulation parameters that achieve desired cortical responses. Here, we address both challenges using Neural Perturbational Inference (NPI), a surrogate-brain approach that estimates EC by applying virtual perturbations to a nonlinear dynamical model fitted to resting-state fMRI data. We extend NPI to a high-resolution thalamocortical network comprising 360 cortical regions and 442 thalamic voxels spanning 12 nuclei. We introduce two innovations in training: (i) a temporal signal-to-noise ratio (tSNR)-weighted loss accounting for signal heterogeneity, and (ii) a multi-resolution, cross-scale consistency loss that regularizes model complexity. These strategies yield improved performance in synthetic benchmarks across varying tSNR regimes. Leveraging the inferred subject-specific EC, we further formulate a constrained linear control problem to identify sparse thalamic stimulation targets that achieve desired cortical activation patterns. We validate the inferred EC structure on two independent datasets: the MacStim dataset comprising two macaque monkeys with infrared neural stimulation on medial pulvinar, and the HumanTC resting-state fMRI dataset comprising twelve human subjects. Our results reveal site-specific thalamocortical EC profiles, producing interpretable predictions that align with known ground-truth structures. Together, this work establishes a computationally grounded pathway toward personalized optimization of thalamic DBS in both human and nonhuman primates.
Ahmed, R., Feng, Y., Roe, A. W., Chen, Z. S.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 12
- Comments 0