Faculty Bibliography

OBJECTIVE:Current brain-computer interface (BCI) studies demonstrate the potential to decode neural signals obtained from structured and trial-based tasks to drive actuators with high performance within the context of these tasks. Ideally, to maximize utility, such systems will be applied to a wide range of behavioral settings or contexts. Thus, we explore the potential to augment such systems with the ability to decode abstract behavioral contextual states from neural activity. APPROACH/METHODS:To demonstrate the feasibility of such context decoding, we used electrocorticography (ECoG) and stereo-electroencephalography (sEEG) data recorded from the cortical surface and deeper brain structures, respectively, continuously across multiple days from three subjects. During this time, the subjects were engaged in a range of naturalistic behaviors in a hospital environment. Behavioral contexts were labeled manually from video and audio recordings; four states were considered: engaging in dialogue, rest, using electronics, and watching television. We decode these behaviors using a factor analysis and support vector machine (SVM) approach. MAIN RESULTS/RESULTS:We demonstrate that these general behaviors can be decoded with high accuracies of 73% for a four-class classifier for one subject and 71% and 62% for a three-class classifier for two subjects. SIGNIFICANCE/CONCLUSIONS:To our knowledge, this is the first demonstration of the potential to disambiguate abstract naturalistic behavioral contexts from neural activity recorded throughout the day from implanted electrodes. This work motivates further study of context decoding for BCI applications using continuously recorded naturalistic activity in the clinical setting.

There are no functional imaging based biomarkers for pharmacological treatment response in temporal lobe epilepsy (TLE). In this study, we investigated whether there is an association between resting state functional brain connectivity (RsFC) and seizure control in TLE. We screened a large database containing resting state functional magnetic resonance imaging (Rs-fMRI) data from 286 epilepsy patients. Patient medical records were screened for seizure characterization, EEG reports for lateralization and location of seizure foci to establish uniformity of seizure localization within patient groups. Rs-fMRI data from patients with well-controlled left TLE, patients with treatment-resistant left TLE, and healthy controls were analyzed. Healthy controls and cTLE showed similar functional connectivity patterns, whereas trTLE exhibited a significant bilateral decrease in thalamo-hippocampal functional connectivity. This work is the first to demonstrate differences in neural network connectivity between well-controlled and treatment-resistant TLE. These differences are spatially highly focused and suggest sites for the etiology and possibly treatment of TLE. Altered thalamo-hippocampal RsFC thus is a potential new biomarker for TLE treatment resistance.

Direct recordings from the human brain have historically involved epilepsy patients undergoing invasive electroencephalography (iEEG) for surgery. However, these measurements are temporally limited and affected by clinical variables. The RNS System (NeuroPace, Inc.) is a chronic, closed-loop electrographic seizure detection and stimulation system. When adapted by investigators for research, it facilitates cognitive testing in a controlled ambulatory setting, with measurements collected over months to years. We utilized an associative learning paradigm in 5 patients with traditional iEEG and 3 patients with chronic iEEG, and found increased hippocampal gamma (60-100 Hz) sustained at 1.3-1.5 seconds during encoding in successful versus failed trials in surgical patients, with similar results in our RNS System patients (1.4-1.6 seconds). Our findings replicate other studies demonstrating that sustained hippocampal gamma supports encoding. Importantly, we have validated the RNS System to make sensitive measurements of hippocampal dynamics during cognitive tasks in a chronic ambulatory research setting.

The running-down phenomenon refers to 2 analogous but distinct entities that may be seen after epilepsy surgery. The first is clinical, and denotes a progressive diminution in seizures after epilepsy surgery in which the epileptogenic zone could not be completely removed (Modern Problems of Psychopharmacology 1970;4:306, Brain 1996:989). The second is electrographic, and refers to a progressive deactivation of a secondary seizure focus after removal of the primary epileptogenic zone. This progressive decrease in epileptiform activity may represent a reversal of secondary epileptogenesis, where a primary epileptogenic zone is postulated to activate epileptiform discharges at a second site and may become independent.³ The electrographic running-down phenomenon has been reported in only limited numbers of patients, using serial postoperative routine scalp electroencephalography (EEG) (Arch Neurol 1985;42:318). We present what is, to our knowledge, the most detailed demonstration of the electrographic running-down phenomenon in humans, made possible by chronic electrocorticography (ECoG). Our patient's left temporal seizure focus overlapped with language areas, limiting the resection to a portion of the epileptogenic zone, followed by implantation of a direct brain-responsive neurostimulator (RNS System, NeuroPace Inc.) to treat residual epileptogenic tissue. Despite the limited extent of the resection, the patient remains seizure-free more than 2 years after surgery, with the RNS System recording ECoG without delivering stimulation. We reviewed the chronic recordings with automated spike detection and inspection of electrographic episodes marked by the neurostimulator. These recordings demonstrate progressive diminution in spiking and rhythmic discharges, consistent with an electrographic running-down phenomenon.

Reliable posture labels in hospital environments can augment research studies on neural correlates to natural behaviors and clinical applications that monitor patient activity. However, many existing pose estimation frameworks are not calibrated for these unpredictable settings. In this paper, we propose a semi-automated approach for improving upper-body pose estimation in noisy clinical environments, whereby we adapt and build around an existing joint tracking framework to improve its robustness to environmental uncertainties. The proposed framework uses subject-specific convolutional neural network models trained on a subset of a patient's RGB video recording chosen to maximize the feature variance of each joint. Furthermore, by compensating for scene lighting changes and by refining the predicted joint trajectories through a Kalman filter with fitted noise parameters, the extended system yields more consistent and accurate posture annotations when compared with the two state-of-the-art generalized pose tracking algorithms for three hospital patients recorded in two research clinics.

Purpose: As part of the Epilepsy Genetics Initiative, we re-evaluated clinically generated exome sequence data from 54 epilepsy patients and their unaffected parents to identify molecular diagnoses not provided in the initial diagnostic interpretation.

In this study, we employed a kernel support vector machine to predict epilepsy localization and lateralization for patients with a diagnosis of epilepsy (n = 228). We assessed the accuracy to which indices of verbal memory, visual memory, verbal fluency, and naming would localize and lateralize seizure focus in comparison to standard electroencephalogram (EEG). Classification accuracy was defined as models that produced the least cross-validated error (CVϵ). In addition, we assessed whether the inclusion of norm-based standard scores, demographics, and emotional functioning data would reduce CVϵ. Finally, we obtained class probabilities (i.e., the probability of a particular classification for each case) and produced receiver operating characteristic (ROC) curves for the primary analyses. We obtained the least error assessing localization data with the Gaussian radial basis kernel function (RBF; support vectors = 157, CVϵ = 0.22). There was no overlap between the localization and lateralization models, such that the poorest localization model (the hyperbolic tangent kernel function; support vectors = 91, CVϵ = 0.36) outperformed the strongest lateralization model (RBF; support vectors = 201, CVϵ = 0.39). Contrary to our hypothesis, the addition of norm, demographics, and emotional functioning data did not improve the accuracy of the models. Receiver operating characteristic curves suggested clinical utility in classifying epilepsy lateralization and localization using neuropsychological indicators, albeit with better discrimination for localizing determinations. This study adds to the existing literature by employing an analytic technique with inherent advantages in generalizability when compared to traditional single-sample, not cross-validated models. In the future, class probabilities extracted from these and similar analyses could supplement neuropsychological practice by offering a quantitative guide to clinical judgements.