Model-free and Model-based Reinforcement Learning in Patients Undergoing Ventral Capsulotomy for OCD
OCD is a disease of cognitive errors, but its underlying neural mechanisms remain poorly understood. Patients with intractable OCD undergoing MRI-guided ablation of the anterior limb of the internal capsule present a unique opportunity to study prefrontal-striatal circuits involved in reinforcement learning (RL). RL is thought to involve model-free (habitual) and model-based (planning) algorithms. We sought to quantify how these systems are impaired in OCD patients, and to map behavioral impairments to lesioned prefrontal fibers that are involved in the brain's RL circuitry.
By engaging OCD patients in a maze task requiring experiential learning and dynamic planning, and fitting patient behavior to a space of artificial RL agents performing the same task, we elicit precise and rigorous descriptions of model-free and model-based RL in our patients. Notably, OCD patients were found to have poor reward learning in the model-free system and poor planning in the model-based system, and both of these parameters improved postoperatively, suggesting that these are core features of OCD pathophysiology. On DTI, greater lesioning of orbitofrontal-striatal fibers were found to be associated with improvements in model-based planning. These results represent an effort to map prefrontal cortex with artificially intelligent agents. Future studies will examine these computations with more complex behavioral tasks and electrophysiology.
Interpretable Neural Networks Reveal Diverse Spectral Encoding of Parkinsonian Neurophysiology in Subthalamic Nucleus
Better understanding of the neural mechanisms mediating Parkinsonian symptoms could lead to more precise biomarkers for closed-loop DBS. By using a real-time intraoperative motor task, we quantify short-timescale Parkinsonian motor dysfunction while recording from microelectrodes in the subthalamic nucleus (STN). Neural networks were trained to decode tremor and bradykinesia on oscillatory neural activity from these microelectrodes.
By examining the local and global activity of our trained neural networks, we sought to explain the computations used to decode Parkinsonian symptoms, which could suggest causal mechanisms in STN oscillations. By occluding canonical oscillation bands and computing subsequent changes in decoding accuracy, we found that information is primarily in the theta-alpha and beta range, and that this information is segreated between elctrodes. By developing algorithms to examine the internal weights and representations of the networks, we seek to discover latent behavioral states and spectrotemporally precise biomarkers.
Anatomical Modeling of Two-Trajectory Laser Amygdalohippocampotomy
Laser interstitial thermal therapy (LITT) is an increasingly popular alternative to surgical resection for the treatment of drug-refractory temporal lobe epilepsy. However, outcomes of LITT lag behind anterior temporal lobectomy. By using 3D anatomical modeling and by creating search algorithms to find optimal ablation trajectories, we demonstrate that status quo one-laser LITT has poor coverage of epilpetogenic cortical structures such as piriform and entorhinal cortex. Using the same computational modeling techniques, we develop and clinically validate a two-laser LITT trajectory that offers greatly increased coverage of these structures.
Thalamic Neuromodulation for Epilepsy
The thalamus has extensive cortical and subcortical connections, suggesting a natural target for preventing or aborting seizures. We provide a brief review of normal thalamic anatomy and physiology, and then propose general strategies for modulating seizure networks through thalamic nodes. Additionally, we provide specific recommendations for targeting the thalamus under different contexts, motivated by a more detailed discussion of its distinct nuclei and their respective circuits. By describing important principles governing thalamic involvement in seizure networks, we aim to guide patient-specific, network-oriented treatment interventions for epilepsy.