IntroductionExtended Reality (XR) technologies offer unprecedented opportunities to redefine physical rehabilitation experiences through embodied telepresence and full-body motion tracking. Realizing meaningful XR in physical rehabilitation requires stakeholder input addressing remote embodied therapy's unique design challenges.MethodsThis study presents one of the first stakeholder co-design workshops on embodied telehealth for physical rehabilitation, engaging participants [N = 24 total including ten clinicians, eight patients, three designers, and three engineers] across five focus groups with stakeholders from clinical, academic, and industry settings. Through structured usability testing and a four-axis framework analysis, we evaluated immersive Virtual Reality (VR) applications supporting synchronous clinician-led embodied appointments and asynchronous clinician-authored patient home exercise programs with full-body tracked avatars and biomechanical assessments.ResultsStakeholder feedback reinforced the need for embodied agency with adaptive input modalities for diverse patient needs, ensuring clinical authenticity through real-time full-body tracking for accurate movement correction and scaling interaction complexity from minimal interfaces for VR novices (78% of participants) to customizable clinical dashboards. Crossstakeholder prioritization identified minimalist XR interface navigation (19 votes) and flexible clinical assessment capabilities (17 votes) as the highest-priority design requirements, while exit surveys suggested high patient comfort (M = 4.50/5) and strong clinician adoption interest (M = 4.14/5).DiscussionWe propose six design principles for meaningful XR telehealth in physical rehabilitation: 1) embodied guidance with real-time feedback, 2) progressive complexity with minimalist defaults, 3) adaptive accessibility through multi-modal input, 4) clinical authenticity via domain-specific assessments, 5) biomechanical precision for trust and safety, and 6) contextual onboarding to improve therapeutic competency. These findings offer design considerations for developing embodied XR telehealth systems that support sustained therapeutic engagement and meaningful rehabilitation outcomes.

The mouse cortex is a canonical model for studying how functional neural networks emerge, yet it remains unclear which topological features arise from intrinsic cellular organization versus sensory input. Mouse forebrain organoids provide a powerful system to investigate these intrinsic mechanisms. We generated dorsal (DF) and ventral (VF) forebrain organoids from mouse pluripotent stem cells and tracked their development using longitudinal electrophysiology. DF organoids showed progressively stronger network-wide correlations, while VF organoids developed more refined activity patterns with enhanced small-world topology and increased modular organization. Both organoid types form small-world networks, but their topological organization differs. These differences emerge without extrinsic inputs and correlate with Pvalb+ interneuron enrichment in VF organoids. Our findings demonstrate how cellular composition influences neural circuit self-organization, establishing mouse forebrain organoids as a tractable platform to study cortical network architecture.

Experimental neuroscience techniques are advancing rapidly, with developments in high-density electrophysiology and targeted electrical stimulation enabling single-cell-resolution recording and stimulation. Cortical organoids derived from pluripotent stem cells show great promise as in vitro models of brain development, function, and disease. In this work, we demonstrate goal-directed learning in brain organoids through feedback-driven neural plasticity. We developed a closed-loop electrophysiology framework to embody mouse cortical organoids into a pole-balancing task ("cartpole") and evaluated performance improvements when delivering high-frequency training signals. We found that, for most organoids, training signals chosen by artificial reinforcement learning yield better performance than randomly chosen training signals or no training signal, yet improvements do not persist after the 45-min rest period. We further show that training-induced plasticity requires intact glutamatergic transmission, as pharmacological blockade of AMPA and NMDA receptors abolished performance improvements. This systematic approach to studying goal-directed neural plasticity mechanisms in vitro opens new possibilities for neural rehabilitation and biological computation.

Immersive Virtual Reality (VR) rehabilitation shows promise for remote movement care, yet implementation requirements differ across healthcare delivery models. This study presents findings from a stakeholder co-design workshop conducted with United States Veterans Affairs (VA) providers, extending previous work with private practice stakeholders. Through structured usability testing and four-axis framework analysis with VA clinicians paired with design researchers, we examined how integrated healthcare networks shape VR telerehabilitation design requirements. Results reveal that VA providers prioritize accessibility and adaptation as foundational requirements, contrasting with private practice stakeholders who emphasized interface simplicity. Key accessibility priorities included wheelchair and seated avatar modes, adaptive range-of-motion mapping, and alternative input methods for veterans with motor impairments. Comparative analysis demonstrates convergent priorities around onboarding, movement feedback, and clinical assessment tools, while divergent priorities reflect differences in patient populations between healthcare contexts. We propose design recommendations for VR rehabilitation platforms serving integrated healthcare networks, emphasizing accessibility-first design, multimodal guidance, and context-aware configuration. These findings contribute a comparative framework for adapting XR rehabilitation across healthcare delivery models.

Closed-loop brain-computer interfaces often require both a forecast of upcoming neural population activity and a readout of the animal's behavioral state. A single Mamba forecaster, trained only on next-step spike counts at Neuropixels scale, can deliver both in one forward pass. A lightweight per-session linear head reading the model's predicted rates decodes behavior better than the same linear classifier reading the raw spike counts, under matched temporal context. We test on the Steinmetz visual-discrimination benchmark, which spans 39 sessions, roughly 27,000 neurons, and 1,994 held-out trials. Across three training seeds, Mamba's predicted rates decode mouse choice at 75.7±0.2

Electrophysiology instruments generate continuous, high-volume data streams during experiments. Processing of these data efficiently requires coordination across distributed computing resources. Longitudinal recordings from high-density microelectrode arrays (HD-MEAs) are particularly challenging to manage due to their scale, heterogeneous software dependencies, and limited computational capacity at data acquisition sites. To address these challenges, we present an event-triggered analysis workflow that utilizes an Internet of Things (IoT) messaging protocol as a lightweight control plane to coordinate data transfer, storage, and computation. By treating IoT messaging as a control-plane abstraction, this approach separates workflow coordination from the software systems that execute analysis tasks, enabling coordination without the need for centralized workflow schedulers or heavyweight runtime environments. We integrated cloud-based storage and elastic computing resources to support automated processing while reducing dependence on local software installations. Analysis services and algorithms are containerized to allow workflows to be composed and executed under constrained infrastructure conditions. We examine this approach through case studies including in vitro neural activity recordings from cortical organoids and ex vivo brain slices, demonstrating that IoT-style workflow orchestration can support multiscale electrophysiology analysis in real experimental settings.

Large language models have the potential to transform scientific research and analysis, but without domain-specific structure they produce silent methodological errors, unreported decisions, and irreproducible results. Here we present SpikeLab, a text-to-analysis framework for neural spike data that combines composable data structures with a skill-based agentic system enforcing bounded autonomy: mandatory use of expert-vetted methods, correctness over efficiency, and clarification-seeking on ambiguous requests. In a controlled benchmark on electrophysiology data, Sonnet 4.6 with SpikeLab produced correct and reproducible results across all tasks, outperforming both the unassisted Sonnet and the more capable Opus 4.6, which exhibited deterministic failures including ad hoc method invention, silent data reduction, and inconsistent experimental designs. We demonstrate versatility across in vivo mouse, human, and in vitro brain organoid recordings, and apply the framework to a pharmacological dose-response study spanning single-unit dynamics, pairwise network structure, burst-level temporal sequences, and latent population states, all through natural language prompts without writing analysis code.

Neural population models, which predict the joint firing of many simultaneously recorded neurons forward in time, are typically evaluated by a single aggregate Pearson correlation r between predicted and actual spike counts, a number that masks critical structure. We argue that how we evaluate spike forecasting matters as much as what we build, and introduce SpikeProphecy, the first large-scale benchmark for causal, autoregressive spike-count forecasting on real electrophysiology recordings. Our core contribution is a population metric decomposition that separates aggregate performance into temporal fidelity, spatial pattern accuracy, and magnitude-invariant alignment. The decomposition surfaces aspects of the underlying data that an aggregate scalar collapses together. We apply the protocol to 105 Neuropixels sessions (Steinmetz 2019 + IBL Repeated Site; 89,800 neurons) with seven architecture baselines spanning four structural families: four SSMs (three diagonal and one non-diagonal), a Transformer, an LSTM, and a spiking network. The decomposition surfaces a brain-region predictability ranking that reproduces across all seven baselines and survives ANCOVA correction for firing-statistics constraints (region ΔR^2 = 0.018 above the firing-statistics covariates). It also exposes a sub-Poisson evaluation floor where rigorous metrics combine with genuine biophysical constraints on regular spike trains, and yields a negative result on KL-on-output-rates distillation for ANN-to-SNN transfer in this Poisson count domain.

Neuronal subtype generation in the mammalian central nervous system is governed by competing genetic programs. The medial ganglionic eminence (MGE) produces two major cortical interneuron (IN) populations, somatostatin (Sst) and parvalbumin (Pvalb), which develop on different timelines. The extent to which external signals influence these identities remains unclear. Pvalb-positive INs are crucial for cortical circuit regulation but challenging to model in vitro. We grafted mouse MGE progenitors into diverse 2D and 3D co-culture systems, including mouse and human cortical, MGE, and thalamic models. Strikingly, only 3D human corticogenesis models promoted efficient, non-autonomous Pvalb differentiation, characterized by upregulation of Pvalb maturation markers, downregulation of Sst-specific markers, and the formation of perineuronal nets. Additionally, lineage-traced postmitotic Sst-positive INs upregulated Pvalb when grafted onto human cortical models. These findings reveal unexpected fate plasticity in MGE-derived INs, suggesting that their identities can be dynamically shaped by the environment.

The mouse cortex is a canonical model for studying how functional neural networks emerge, yet it remains unclear which topological features arise from intrinsic cellular organization versus external regional cues. Mouse forebrain organoids provide a powerful system to investigate these intrinsic mechanisms. We generated dorsal (DF) and ventral (VF) forebrain organoids from mouse pluripotent stem cells and tracked their development using longitudinal electrophysiology. DF organoids showed progressively stronger network-wide correlations, while VF organoids developed more refined activity patterns, enhanced small-world topology, and increased modular organization. These differences emerged without extrinsic inputs and may be driven by the increased generation of Pvalb+ interneurons in VF organoids. Our findings demonstrate how variations in cellular composition influence the self-organization of neural circuits, establishing mouse forebrain organoids as a tractable platform to study how neuronal populations shape cortical network architecture.

The analysis of tissue cultures requires a sophisticated integration and coordination of multiple technologies for monitoring and measuring. We have developed an automated research platform enabling independent devices to achieve collaborative objectives for feedback-driven cell culture studies. Our approach enables continuous, communicative, non-invasive interactions within an Internet of Things (IoT) architecture among various sensing and actuation devices, achieving precisely timed control of in vitro biological experiments. The framework integrates microfluidics, electrophysiology, and imaging devices to maintain cerebral cortex organoids while measuring their neuronal activity. The organoids are cultured in custom, 3D-printed chambers affixed to commercial microelectrode arrays. Periodic feeding is achieved using programmable microfluidic pumps. We developed a computer vision fluid volume estimator used as feedback to rectify deviations in microfluidic perfusion during media feeding/aspiration cycles. We validated the system with a set of 7-day studies of mouse cerebral cortex organoids, comparing manual and automated protocols. It was shown that the automated protocols maintained robust neural activity throughout the experiment while enabling hourly electrophysiology recordings during the experiments. The median firing rates of neural units increased for each sample, and dynamic patterns of organoid firing rates were revealed by high-frequency recordings. Surprisingly, feeding did not affect the firing rate. Furthermore, media exchange during a recording did not show acute effects on firing rate, enabling the use of this automated platform for reagent screening studies.

Seizures are made up of the coordinated activity of networks of neurons. It follows that control of neurons in the pathologic circuits of epilepsy could allow for control of the disease. In non-human disease models of epilepsy, optogenetics has been effective at stopping seizure-like activity by increasing inhibitory tone or decreasing excitation. However, this has not been shown in human brain tissue. Many of the genetic means for achieving channelrhodopsin expression in non-human models are not possible in humans, and vector-mediated methods are susceptible to species-specific tropism that may affect translational potential. There is currently no platform for testing the effects of these potentially disease-modifying tools on network activity in human brain tissue. Human hippocampus resected from patients with refractory epilepsy were collected, cut to 300um and plated at the air-fluid interface on cell-culture inserts. AAV transduction with channelrhodopsins driven by a glutamatergic promoter took place on the day of collection. Slices were plated on high-density micro-electrode arrays. Hyperactivity was promoted via bicuculline, low-magnesium media and kainic acid. Slices were illuminated by LED fiberoptics positioned over the slice using a custom recording chamber. Neuronal transduction ranged from 12 – 54% (median 23%). Illumination of slices expressing the depolarizing channelrhodopsin HcKCR1 caused reduction in network firing rates in 8/8 slices expressing HcKCR1. Reductions in network firing rates were significant in all conditions, physiologic media, GABAaR blockade, low-magnesium media and low-magnesium media with kainic acid. Here, we demonstrate AAV-mediated, optogenetic reductions in network firing rates of human hippocampal slices recorded on high-density microelectrode arrays under several hyperactivity provoking conditions. This platform can serve to bridge the gap between human and animal studies by exploring genetic interventions on network activity human brain tissue.

How seizures begin at the level of microscopic neuronal circuits remains unknown. Advancements in high-density CMOS-based microelectrode arrays can be harnessed to study neuronal network activity with unprecedented spatial and temporal resolution. We use high-density electrophysiology recordings to probe the network activity of human hippocampal brain slices from six patients with mesial temporal lobe epilepsy. Two slices from the dentate gyrus exhibited epileptiform activity in the presence of low magnesium media with kainic acid. Both slices exhibit network oscillations indicative of a reciprocally connected circuit, which is unexpected under normal physiological conditions. Future studies may apply this approach to elucidate the network signals that underlie seizure initiation.

Neuronal firing patterning in the dentate gyrus of patients with epilepsy remains unknown at the microcircuit level. Advancements in high-density CMOS-based microelectrode arrays can be harnessed to study network activity with unprecedented spatial and temporal resolution. We use novel computational methods with high-density electrophysiology recordings to spatially map network activity of human hippocampal brain slices from six patients with mesial temporal lobe epilepsy. Two slices from the dentate gyrus exhibited synchronous bursting activity in the presence of low magnesium media with kainic acid, representative of seizure-like behavior. We bridged microscale circuit dynamics with alterations in theta oscillations at the network scale. Future studies may apply this approach to spatially elucidate functional networks and their possible role in seizures.NEW & NOTEWORTHY We apply high-density CMOS-based microelectrode arrays to excised patient brain slices, mapping the communication patterns of hundreds of neurons at unprecedented resolution. We developed novel computational techniques to spatially map neuronal dynamics. In patient slices, our findings suggest that recurrent feedback localized within the dentate gyrus of the hippocampus is linked to a previously unreported phenomenon of theta propagations. This bridges microscale circuit dynamics with alterations in theta oscillations.

Electrophysiology offers a high-resolution method for real-time measurement of neural activity. Longitudinal recordings from high-density microelectrode arrays (HD-MEAs) can be of considerable size for local storage and of substantial complexity for extracting neural features and network dynamics. Analysis is often demanding due to the need for multiple software tools with different runtime dependencies. To address these challenges, we developed an open-source cloud-based pipeline to store, analyze, and visualize neuronal electrophysiology recordings from HD-MEAs. This pipeline is dependency agnostic by utilizing cloud storage, cloud computing resources, and an Internet of Things messaging protocol. We containerized the services and algorithms to serve as scalable and flexible building blocks within the pipeline. In this paper, we applied this pipeline on two types of cultures, cortical organoids and ex vivo brain slice recordings to show that this pipeline simplifies the data analysis process and facilitates understanding neuronal activity.

Seizures are made up of the coordinated activity of networks of neurons, suggesting that control of neurons in the pathologic circuits of epilepsy could allow for control of the disease. Optogenetics has been effective at stopping seizure-like activity in non-human disease models by increasing inhibitory tone or decreasing excitation, although this effect has not been shown in human brain tissue. Many of the genetic means for achieving channelrhodopsin expression in non-human models are not possible in humans, and vector-mediated methods are susceptible to species-specific tropism that may affect translational potential. Here we demonstrate adeno-associated virus-mediated, optogenetic reductions in network firing rates of human hippocampal slices recorded on high-density microelectrode arrays under several hyperactivity-provoking conditions. This platform can serve to bridge the gap between human and animal studies by exploring genetic interventions on network activity in human brain tissue.

With the use of high-density multi-electrode recording devices, electrophysiological signals resulting from action potentials of individual neurons can now be reliably detected on multiple adjacent recording electrodes. Spike sorting assigns these signals to putative neural sources. However, until now, spike sorting can only be performed after completion of the recording, preventing true real time usage of spike sorting algorithms. Utilizing the unique propagation patterns of action potentials along axons detected as high-fidelity sequential activations on adjacent electrodes, together with a convolutional neural network-based spike detection algorithm, we introduce RT-Sort (Real Time Sorting), a spike sorting algorithm that enables the sorted detection of action potentials within 7.5ms±1.5ms (mean±STD) after the waveform trough while the recording remains ongoing. RT-Sort's true real-time spike sorting capabilities enable closed loop experiments with latencies comparable to synaptic delay times. We show RT-Sort's performance on both Multi-Electrode Arrays as well as Neuropixels probes to exemplify RT-Sort's functionality on different types of recording hardware and electrode configurations.

The introduction of Internet-connected technologies to the classroom has the potential to revolutionize STEM education by allowing students to perform experiments in complex models that are unattainable in traditional teaching laboratories. By connecting laboratory equipment to the cloud, we introduce students to experimentation in pluripotent stem cell (PSC)-derived cortical organoids in two different settings: using microscopy to monitor organoid growth in an introductory tissue culture course and using high-density (HD) multielectrode arrays (MEAs) to perform neuronal stimulation and recording in an advanced neuroscience mathematics course. We demonstrate that this approach develops interest in stem cell and neuroscience in the students of both courses. All together, we propose cloud technologies as an effective and scalable approach for complex project-based university training.

Most physical therapists would agree that physical rehabilitation is difficult to perform remotely. Consequently, the global COVID-19 pandemic has forced many physical therapists and their clients to adapt to telehealth, especially with video conferencing. In this article, we ask: How has telehealth for physical rehabilitation evolved with the global pandemic and what are the largest technological needs, treatment methodologies, and patient barriers? With the increased widespread use of telehealth for physical therapy, we present a qualitative study towards examining the shortcomings of current physical therapy mediums and how to steer future virtual reality technologies to promote remote patient evaluation and rehabilitation. We interviewed 130 physical rehabilitation professionals across the United States through video conferencing during the COVID19 pandemic from July—August 2020. Interviews lasted 30–45 min using a semi-structured template developed from an initial pilot of 20 interviews to examine potential barriers, facilitators, and technological needs. Our findings suggest that physical therapists utilizing existing telehealth solutions have lost their ability to feel their patients’ injuries, easily assess range of motion and strength, and freely move about to examine their movements when using telehealth. This makes it difficult to fully evaluate a patient and many feel that they are more of a “life coach” giving advice to a patient rather than a traditional in-person rehabilitation session. The most common solutions that emerged during the interviews include: immersive technologies which allow physical therapists and clients 1) to remotely walk around each other in 3D, 2) enable evidence-based measures, 3) automate documentation, and 4) provider clinical practice operation through the cloud. We conclude with a discussion on opportunities for immersive virtual reality towards telehealth for physical rehabilitation.

Objective: The adoption of telehealth has rapidly accelerated owing to the global COVID19 pandemic disrupting communities and in-person healthcare practices. While telehealth had initial benefits in enhancing accessibility for remote treatment, physical rehabilitation has been heavily limited owing to the loss of hands-on evaluation tools. This paper presents an immersive virtual reality (iVR) pipeline for replicating physical therapy success metrics through applied machine learning of patient observation. Methods: We demonstrate a method of training gradient boosted decision-trees for kinematic estimation to replicate mobility and strength metrics using an off-the-shelf iVR system. During the two-month study, training data were collected while a group of users completed physical rehabilitation exercises in an iVR game. Utilizing this data, we trained on iVR-based motion capture data and OpenSim biomechanical simulations. Results: Our final model indicates that upper-extremity kinematics from OpenSim can be accurately predicted using the HTC Vive head-mounted display system with a Mean Absolute Error less than 0.78° for joint angles and less than 2.34 Nm for joint torques. Additionally, these predictions are viable for runtime estimation, with approximately a 0.74 ms rate of prediction during exercise sessions. Conclusion: These findings suggest that iVR paired with machine learning can serve as an effective medium for collecting evidence-based patient success metrics for telehealth. Significance: Our approach can help increase the accessibility of physical rehabilitation with off-the-shelf iVR head-mounted display systems by providing therapists with the metrics needed for remote evaluation.