Expert and deep learning model identification of iEEG seizures and seizure onset times.

Hundreds of 90-s iEEG records are typically captured from each NeuroPace RNS System patient between clinic visits. While these records provide invaluable information about the patient's electrographic seizure and interictal activity patterns, manually classifying them into electrographic seizure/non-seizure activity, and manually identifying the seizure onset channels and times is an extremely time-consuming process. A convolutional neural network based Electrographic Seizure Classifier (ESC) model was developed in an earlier study. In this study, the classification model is tested against iEEG annotations provided by three expert reviewers board certified in epilepsy. The three experts individually annotated 3,874 iEEG channels from 36, 29, and 35 patients with leads in the mesiotemporal (MTL), neocortical (NEO), and MTL + NEO regions, respectively. The ESC model's seizure/non-seizure classification scores agreed with the three reviewers at 88.7%, 89.6%, and 84.3% which was similar to how reviewers agreed with each other (92.9%-86.4%). On iEEG channels with all 3 experts in agreement (83.2%), the ESC model had an agreement score of 93.2%. Additionally, the ESC model's certainty scores reflected combined reviewer certainty scores. When 0, 1, 2 and 3 (out of 3) reviewers annotated iEEG channels as electrographic seizures, the ESC model's seizure certainty scores were in the range: [0.12-0.19], [0.32-0.42], [0.61-0.70], and [0.92-0.95] respectively. The ESC model was used as a starting-point model for training a second Seizure Onset Detection (SOD) model. For this task, seizure onset times were manually annotated on a relatively small number of iEEG channels (4,859 from 50 patients). Experiments showed that fine-tuning the ESC models with augmented data (30,768 iEEG channels) resulted in a better validation performance (on 20% of the manually annotated data) compared to training with only the original data (3.1s vs 4.4s median absolute error). Similarly, using the ESC model weights as the starting point for fine-tuning instead of other model weight initialization methods provided significant advantage in SOD model validation performance (3.1s vs 4.7s and 3.5s median absolute error). Finally, on iEEG channels where three expert annotations of seizure onset times were within 1.5 s, the SOD model's seizure onset time prediction was within 1.7 s of expert annotation.

Non-linear Embedding Methods for Identifying Similar Brain Activity in 1 Million iEEG Records Captured From 256 RNS System Patients.

Finding electrophysiological features that are similar across patients with epilepsy may facilitate identifying treatment options for one patient that worked in patients with similar brain activity patterns. Three non-linear iEEG (intracranial electroencephalogram) embedding methods of finding similar cross-patient iEEG records in a large iEEG dataset were developed and compared. About 1 million iEEG records from 256 patients with drug-resistant focal onset seizures who were treated in prospective trials of the RNS System were used for analyses. Data from 200, 25, and 31 patients were randomly selected to be in the train, validation, and test datasets. In method 1, ResNet50 convolutional neural network (CNN) model pre-trained on the ImageNet dataset was used for extracting feature maps from spectrogram images (ImageNet-ResNet) of iEEG records. In method 2, ResNet50 custom trained on an iEEG classification task using ~138,000 manually labeled iEEG records was used as the feature extractor (ESC-ResNet). Feature maps were passed through dimensionality reduction and k nearest neighbors were found in the reduced feature space. In method 3, a 256 dimensional iEEG embedding space was learned via contrastive learning by training a ResNet50 model with triplet training sets generated using within-patient iEEG clustering (CL-ResNet). All three methods had comparable performance when identifying iEEG records from the search dataset similar to test iEEG records of baseline (non-seizure) and interictal spiking activity. Epileptic interictal spikes are represented by vertical (broadband) edges in spectrogram images, and hence even generic features extracted using models trained on everyday images appear to be sufficient to represent iEEG records with similar levels of interictal spiking activity in close proximity. In the case of electrographic seizures, however, the ESC-ResNet model, identified cross-patient iEEG records with electrographic seizure morphology features that were most similar to the test iEEG records. For nuanced electrographic seizure iEEG representation learning, domain specific model training with manually generated labels had the advantage. Finally, representative iEEG records were selected from every patient using an unsupervised clustering method which effectively reduced the number of iEEG records in the search dataset from ~750,000 to 2,148, thus substantially reducing the time required for finding similar cross-patient iEEG records.

A High Accuracy Electrographic Seizure Classifier Trained Using Semi-Supervised Labeling Applied to a Large Spectrogram Dataset.

The objective of this study was to explore using ECoG spectrogram images for training reliable cross-patient electrographic seizure classifiers, and to characterize the classifiers' test accuracy as a function of amount of training data. ECoG channels in ∼138,000 time-series ECoG records from 113 patients were converted to RGB spectrogram images. Using an unsupervised spectrogram image clustering technique, manual labeling of 138,000 ECoG records (each with up to 4 ECoG channels) was completed in 320 h, which is an estimated 5 times faster than manual labeling without ECoG clustering. For training supervised classifier models, five random folds of data were created; with each fold containing 72, 18, and 23 patients' data for model training, validation and testing respectively. Five convolutional neural network (CNN) architectures, including two with residual connections, were trained. Cross-patient classification accuracies and F1 scores improved with model complexity, with the shallowest 6-layer model (with ∼1.5 million trainable parameters) producing a class-balanced seizure/non-seizure classification accuracy of 87.9% on ECoG channels and the deepest ResNet50-based model (with ∼23.5 million trainable parameters) producing a classification accuracy of 95.7%. The trained ResNet50-based model additionally had 93.5% agreement in scores with an independent expert labeller. Visual inspection of gradient-based saliency maps confirmed that the models' classifications were based on relevant portions of the spectrogram images. Further, by repeating training experiments with data from varying number of patients, it was found that ECoG spectrogram images from just 10 patients were sufficient to train ResNet50-based models with 88% cross-patient accuracy, while at least 30 patients' data was required to produce cross-patient classification accuracies of >90%.

Acute effects of brain-responsive neurostimulation in drug-resistant partial onset epilepsy.

OBJECTIVE: Understanding the acute effects of responsive stimulation (AERS) based on intracranial EEG (iEEG) recordings in ambulatory patients with drug-resistant partial epilepsy, and correlating these with changes in clinical seizure frequency, may help clinicians more efficiently optimize responsive stimulation settings. METHODS: In patients implanted with the NeuroPace® RNS® System, acute changes in iEEG spectral power following active and sham stimulation periods were quantified and compared within individual iEEG channels. Additionally, acute stimulation-induced acute iEEG changes were compared within iEEG channels before and after patients experienced substantial reductions in clinical seizure frequency. RESULTS: Responsive stimulation resulted in a 20.7% relative decrease in spectral power in the 2-4 second window following active stimulation, compared to sham stimulation. On several detection channels, the AERS features changed when clinical outcomes improved but were relatively stable otherwise. AERS change direction associated with clinical improvement was generally consistent within detection channels. CONCLUSIONS: In this retrospective analysis, patients with drug-resistant partial epilepsy treated with direct brain-responsive neurostimulation showed an acute stimulation related reduction in iEEG spectral power that was associated with reductions in clinical seizure frequency. SIGNIFICANCE: Identifying favorable stimulation related changes in iEEG activity could help physicians to more rapidly optimize stimulation settings for each patient.

Quantitative electrocorticographic biomarkers of clinical outcomes in mesial temporal lobe epileptic patients treated with the RNS® system.

OBJECTIVES: Find interictal electrocorticographic (ECoG) biomarkers of clinical outcomes in mesiotemporal lobe (MTL) epilepsy patients. METHODS: In the NeuroPace® RNS® System clinical trials with 256 patients, 20 MTL patients with the most reduction in clinical seizures at Year 7 compared to baseline (upper response quartile; -96.5% median change) and 20 with the least reduction in clinical seizures (lower response quartile; -17.4% median change) were evaluated. Clinical and interictal ECoG features from the two response quartiles were compared. RESULTS: Demographic and clinical features were similar in the upper and lower response quartiles. Interictal spike rate (ISR) was substantially lower (p < 0.0001) in the upper quartile patients, while normalized theta (4-8 Hz) and normalized gamma (>25 Hz) were also different (p < 0.05) between the two response quartiles. ISR was positively correlated (p < 0.05) with clinical seizure rates in 71% of the channels analyzed. ECoG records captured during months with no clinical seizures had the lowest ISR. CONCLUSIONS: ISR is a strong differentiator of clinical response in MTL patients. Normalized theta and gamma also differentiates clinical response. SIGNIFICANCE: In MTL patients, the interictal spike rate along with spectral power computed from chronic ambulatory baseline ECoGs may serve as biomarkers of clinical outcomes and maybe used as treatment endpoints.

Treatment of drug-resistant epilepsy in patients with periventricular nodular heterotopia using RNS® System: Efficacy and description of chronic electrophysiological recordings.

OBJECTIVES: Describe changes in clinical seizure frequency and electrophysiological data recorded in patients with medically-intractable seizures and periventricular nodular heterotopias (PVNH) treated with the RNS® System (NeuroPace, Inc., Mountain View, CA). METHODS: Clinical seizures from eight patients (mean follow-up of 10.1 years) were analyzed pre- and post-treatment. Chronic ambulatory electrocorticograms (ECoGs) recorded from PVNHs, hippocampus and neocortex were evaluated to identify the earliest electrographic seizure onset type, pattern of spread, and interictal characteristics. RESULTS: Mean reduction in disabling seizures was 85.7 % (n = 8); seven patients had >50% seizure reduction and two were seizure-free in the final year of analysis. Seizure rate showed a progressive reduction over the course of the study with the highest rate of improvement in the first two to three years after implantation. Four of seven patients with one PVNH lead and a second lead in the hippocampus or neocortex had some electrographic seizures first recorded at either lead location, suggesting two foci or seizure propagation patterns. Low voltage fast type activity was the prominent seizure onset pattern. Interictal ECoG power was lower in PVNH than hippocampus. CONCLUSIONS: RNS® System treatment substantially reduced clinical seizure frequency in patients with PVNH. Analysis of ictal ECoG records suggests PVNH may be involved in seizure generation. SIGNIFICANCE: Chronic ECoG recordings suggest PVNH tissue can actively participate in epileptogenic networks. Direct brain-responsive neurostimulation is a safe and effective treatment option in such patients, progressively reducing seizure rate over a period of years.

Changes in the electrocorticogram after implantation of intracranial electrodes in humans: The implant effect.

OBJECTIVE: Subacute and long-term electrocorticographic (ECoG) changes in ambulatory patients with depth and cortical strip electrodes were evaluated in order to determine the length of the implant effect. METHODS: ECoG records were assessed in patients with medically intractable epilepsy who had depth and/or strip leads implanted in order to be treated with brain-responsive stimulation. Changes in total spectral power, band-limited spectral power, and spike rate were assessed. RESULTS: 121 patients participating in trials of the RNS® System had a total of 93994 ECoG records analyzed. Significant changes in total spectral power occurred from the first to second months after implantation, involving 55% of all ECoG channels (68% of strip and 47% of depth lead channels). Significant, but less pronounced, changes continued over the 2nd to 5th post-implant months, after which total power became more stable. Similar patterns of changes were observed within frequency bands and spike rate. CONCLUSIONS: ECoG spectral power and spike rates are not stable in the first 5 months after implantation, presumably due to neurophysiological and electrode-tissue interface changes. SIGNIFICANCE: ECoG data collected in the first 5 months after implantation of intracranial electrodes may not be fully representative of chronic cortical electrophysiology.

Deep brain stimulation macroelectrodes compared to multiple microelectrodes in rat hippocampus.

Microelectrode arrays (wire diameter <50 µm) were compared to traditional macroelectrodes for deep brain stimulation (DBS). Understanding the neuronal activation volume may help solve some of the mysteries associated with DBS, e.g., its mechanisms of action. We used c-fos immunohistochemistry to investigate neuronal activation in the rat hippocampus caused by multi-micro- and macroelectrode stimulation. At ± 1V stimulation at 25 Hz, microelectrodes (33 µm diameter) had a radius of activation of 100 µm, which is 50% of that seen with 150 µm diameter macroelectrode stimulation. Macroelectrodes activated about 5.8 times more neurons than a single microelectrode, but displaced ~20 times more neural tissue. The sphere of influence of stimulating electrodes can be significantly increased by reducing their impedance. By ultrasonic electroplating (sonicoplating) the microelectrodes with platinum to increase their surface area and reduce their impedance by an order of magnitude, the radius of activation increased by 50 µm and more than twice the number of neurons were activated within this increased radius compared to unplated microelectrodes. We suggest that a new approach to DBS, one that uses multiple high-surface area microelectrodes, may be more therapeutically effective due to increased neuronal activation.

Closed-Loop, Multichannel Experimentation Using the Open-Source NeuroRighter Electrophysiology Platform.

Single neuron feedback control techniques, such as voltage clamp and dynamic clamp, have enabled numerous advances in our understanding of ion channels, electrochemical signaling, and neural dynamics. Although commercially available multichannel recording and stimulation systems are commonly used for studying neural processing at the network level, they provide little native support for real-time feedback. We developed the open-source NeuroRighter multichannel electrophysiology hardware and software platform for closed-loop multichannel control with a focus on accessibility and low cost. NeuroRighter allows 64 channels of stimulation and recording for around US $10,000, along with the ability to integrate with other software and hardware. Here, we present substantial enhancements to the NeuroRighter platform, including a redesigned desktop application, a new stimulation subsystem allowing arbitrary stimulation patterns, low-latency data servers for accessing data streams, and a new application programming interface (API) for creating closed-loop protocols that can be inserted into NeuroRighter as plugin programs. This greatly simplifies the design of sophisticated real-time experiments without sacrificing the power and speed of a compiled programming language. Here we present a detailed description of NeuroRighter as a stand-alone application, its plugin API, and an extensive set of case studies that highlight the system's abilities for conducting closed-loop, multichannel interfacing experiments.