Ben Shofty

Distributed brain networks compete with one another in the service of information processing. One dramatic example is antagonism between networks that support internally directed cognition, often collectively referred to as the default network, and networks that orient to external events and drive action. Here we combined within-individual precision functional neuroimaging and intracranial electrophysiology to investigate network competition at high spatiotemporal resolution. In six patients, robust antagonism between networks was established during a target detection task. Network-assigned electrophysiological recordings displayed concordant responses, including in single-neuron firing rates, and further revealed the temporal dynamics of network competition. Networks involved in salience detection and action planning showed early, transient responses that peaked near the participant response. Default network suppression occurred later, reflected in decreases in high-frequency broadband activity and in single-neuron firing rates in response to targets. Examining subcortical regions revealed rapid activity increases in the thalamus and posterior hippocampus aligned to the participant response. Critically, a late evolving, sustained suppression was observed throughout the hippocampus. These findings suggest that, across cortical and subcortical regions, network competition is temporally layered. Furthermore, the hippocampus can be dissociated along its long axis during competition. Such push-pull antagonism may be fundamental to advanced human cognition, and its disruption may underlie mental illness.

Real-world decision-making rarely occurs with perfect information. Instead, individuals must constantly weigh potential rewards against the probability of adverse outcomes.1 Failures of this process can lead to maladaptive decisions associated with reduced lifetime success, and numerous psychiatric disorders such as gambling addictions, bulimia nervosa, and substance use disorder.2,3 The neural computations that facilitate inference about the landscape of potential outcomes remain unclear, but are thought to occur in distributed frontotemporal circuits.4 Here we used deep reinforcement learning agents to predict distinct behavioral strategies and their underlying neural population dynamics during a risky decision-making task. Across a range of training conditions, deep reinforcement learning agents separated into strategies marked by either overly cautious exploration of the reward contingency space or a high-performing, risk-adaptive Bimodal strategy. The internal dynamics of high-performing Bimodal agents formed low-dimensional representations that segregated safe and risky states. In contrast, the cautious exploration agents were associated with more skewed and entangled neural representations. We found remarkably similar dynamical representations and their associated behavioral strategies in neuronal ensemble recordings from human epilepsy patients performing a similar risky decision-making task. These results reveal the structure of dynamical computations that underlie inferences about uncertain outcomes and their associated behavioral strategies.

Time perception and decision-making are interrelated processes that are critical to much of human and animal behavior. Although the behavioral importance of this link between the processes has been established, the neural underpinnings of temporal decision-making are unclear. The current study leverages human intracranial electro-physiology to investigate the neural contribution of temporal processing to decision-making under risk. To probe temporal decision-making, we use a version of the Balloon Analogue Risk Task, in which participants inflate virtual balloons with the goal of reaching maximal balloon size but stopping inflation before the balloon pops. Points are awarded based on inflation duration of successful trials, such that points awarded are linearly related to balloon size. By decoding temporal features of the task on a trial-by-trial basis, we find neural representations of time during balloon inflation. Specifically, we find that both medial and lateral temporal regions of the brain encode the majority of this time-related signal. Importantly, we can decode the progression of time leading up to the outcome of a successful decision (pressing a button to stop inflation before the balloon pops). This encoding of time depends on an active decision to end the trial: we cannot decode time during passive observation of balloon inflation to a known maximal size. To validate these results further, we use a temporal convolution network to predict participants’ response (i.e., time of button press) based on the brain activity leading up to the response. We find that we can predict the moment of response with high accuracy and temporal resolution. The model predicts the timing of successful decisions but not the timing of balloon pops. Together, these findings reveal neural representations that encode the progression of time, specifically in relation to an active decision.

Objective: To address the limitations of current trial-and-error programming strategies in deep brain stimulation (DBS) for refractory obsessive-compulsive disorder (OCD), we implanted patients with sensing-capable DBS devices to identify neural biomarkers that could provide objective feedback to the clinician about therapeutic efficacy. Methods: We conducted an early feasibility study in 10 patients with severe, treatment-resistant OCD. All subjects received bilateral DBS leads targeting the ventral internal capsule (VC) and the second half of the cohort also received strip electrodes over the bilateral orbitofrontal cortex for recording only. All leads were connected to investigational, bidirectional DBS devices. After implantation, participants returned for scheduled programming visits to determine optimal stimulation parameters (Phase 1). In Phase 2, patients completed a course of exposure and response prevention (ERP) psychotherapy, and in Phase 3, patients underwent a double-blind discontinuation of DBS to test true vs. sham response. Phase 4 was an open-label follow-up. We administered standardized symptom scales throughout the study and used non-parametric repeated-measures analyses to analyze neuropsychological data. Results: All patients elected to resume stimulation after a discontinuation phase. At the end of the study, the mean reduction in the Yale-Brown Obsessive-Compulsive Scale (Y-BOCS) was 22 points or 60% across all patients with 8 patients demonstrating full response ( > 35% decrease in Y-BOCS). All participants also experienced reduced depression severity. Conclusions: In patients with refractory OCD, we demonstrate excellent clinical response to VC DBS. We show feasibility of recording neural data both at home and in the clinic on board bidirectional DBS devices.

Introduction: Responsive electrical stimulation of the centromedian nucleus of the thalamus (CMN) is an emerging therapy for reducing seizures in individuals with multifocal or generalized drug-resistant epilepsy (DRE). Prior studies have reported significant neuropsychiatric side effects associated with thalamic stimulation, particularly in the anterior nucleus of the thalamus. Despite its growing popularity, little is known about the effects of CMN stimulation on neurocognitive and mood outcomes. To address this knowledge gap, we performed a retrospective analysis of patients who underwent responsive neurostimulation (RNS) of the CMN at our institution. Methods: We identified 8 patients with multifocal or unresectable focal DRE who participated in a comprehensive evaluation before and ∼12 months after surgery to implant an RNS device. Neuropsychological testing spanned multiple functional domains (i.e., attention and working memory, learning, retrieval, phonemic fluency, semantic fluency, motor dexterity, mood, and anxiety), and 9 distinct functional metrics were analyzed. Results: In this small retrospective cohort, closed-loop stimulation of the CMN was associated with modest improvements in verbal learning and retrieval and a slight decline in phonemic verbal fluency. Although there was no robust group-level effect on mood, the proportion of patients with clinically significant depression decreased from 43% before stimulation to 0% at last follow-up and from 71% to 57% for clinically significant anxiety. Our analyses did not identify changes in attention and working memory, semantic fluency, or motor dexterity. Conclusion: Our results suggest that closed-loop CMN stimulation was not associated with major adverse cognitive or affective changes over 12 months.

Introduction: Deep brain stimulation (DBS) is a well-defined therapeutic intervention for various neurological and psychiatric disorders; however, there are limited data comparing the costs associated with the various advanced stereotactic techniques employed in DBS surgery. In this study, we assessed the specific factors driving the variability in costs at a single institution across stereotactic approaches for DBS and aimed to highlight potential areas for cost optimization. Methods: Inflation-adjusted cost data for patients undergoing DBS from 2012 to 2023 were retrospectively accessed using the University of Utah’s Value-Driven Outcomes tool. Patients who underwent bilateral DBS surgery with or without same-stage internal pulse generator implantation were included. Surgical approaches to DBS implantation were separated into four techniques by the type or lack of frame and intraoperative imaging used: (1) frame with C-arm fluoroscopy; (2) frame with intraoperative computed tomography (iCT); (3) robotic stereotaxy with iCT; and (4) frameless with intraoperative magnetic resonance imaging (iMRI [ClearPoint™]). Total reported costs for each technique were also split into four subcategories (imaging, facility, supply, services) and expressed as a percentage relative to the total cost for the staged frame with the C-arm method. Results: Our study included 256 patients who underwent surgical procedures for DBS. The median (interquartile range) cost percentages were 171.0% (±29.0%) for frameless with iMRI, 121.7% (±14.9%) for frame with iCT, and 117.5% (±14.0%) for robotic stereotaxy with iCT procedures. Only the iMRI-based method demonstrated a significantly increased cost (p < 0.05) when compared with all other methods. Similarly, the iMRI-based method demonstrated notably higher costs than all other methods across imaging and supply subcategories. Conclusion: The iMRI method’s imaging likely costs significantly more because iMRI procedure times are longer than other methods. Supply costs for the iMRI-based method were not statistically correlated with procedure length, highlighting a potential area for cost-saving measures. Future inclusion of technical accuracy, revision and readmission rates, perioperative complications, and other postoperative outcomes may more fully evaluate the most cost-effective approach.

Human decisions under uncertainty are shaped by experience, but the computations that translate expectation and experience into choice remain debated in neural and cognitive science. Prior studies highlight reinforcement learning (RL) as a unifying framework, yet it is unclear whether human behavior under risk is better captured by symmetric updating from outcomes or by asymmetric learning that weights reward and loss differently. This work examines which learning strategies better explain trial-by-trial choices given contextual uncertainty and manipulations of outcome distributions. Our results show that a Risk Sensitive (RS) model with asymmetric learning rates best explains human behavior in our novel decision-making task. Fitting candidate models to individual trial histories yielded value signals that predicted both choice and response time. These results highlight that RS model, as an asymmetric learning provides a concise and identifiable account of behavior in decision-making under risk tasks.