Citations

OBJECTIVE Deep brain stimulation (DBS) is emerging as an additional surgical option for refractory, multifocal epilepsy. Four-lead DBS systems can be used when seizure onset is poorly localized, generalized, or involves multiple targetable circuits. The goal of this study was to assess preliminary outcomes and complication rates in 4-lead DBS systems for treating epilepsy. METHODS A consecutive series of 32 patients with 4-lead DBS implants who had at least 6 months of follow-up was reviewed, and demographics, outcomes, and complications were abstracted. RESULTS Thirty-two patients implanted with 4 DBS leads (including off-label use) were included. The median age at surgery was 28.5 years, the median age at epilepsy diagnosis was 7.5 years, and 18 patients (56%) were women. The median epilepsy duration prior to implantation was 21 years. The most common seizure localization was frontotemporal (53%), followed by generalized (28%) and parietoccipital (13%). Prior vagus nerve stimulation had been trialed in 20 patients (63%), and an invasive stereotactic EEG recording was performed in 17 patients (59%). Eight patients had prior epilepsy surgery (25%). DBS targets included the anterior thalamic nucleus in combination with the centromedian thalamic nucleus, hippocampus, pulvinar nucleus, cingulate, insula, and globus pallidus. The median surgical duration was 5 hours 22 minutes, with a median operating room time of 8 hours 21 minutes. The median postoperative follow-up was 27 (interquartile range 17–41) months. Three patients experienced transient neurological deficits and 2 patients required intraoperative lead repositioning. There were no infections or hemorrhages. Most patients benefited from stimulation (22% Engel class I, 22% class II, 28% class III, and 28% class IV). The median patient-reported seizure reduction rate was 70%, with a responder rate of 72%. One patient elected to have the system explanted. CONCLUSIONS Four-lead DBS is safe and can achieve meaningful improvement in challenging cases of multifocal epilepsy with a low complication rate.

Abstract Objective High-frequency oscillations (HFOs) are highly specific biomarkers of epileptogenic tissue, yet their noninvasive localization remains challenging due to their brief duration, low amplitude, and poor signal-to-noise ratio. Here, we introduce TH-DeepSIF, a transformer-based deep learning framework trained on biologically realistic neural mass model simulations, to robustly perform HFO source imaging from scalp EEG. Methods TH-DeepSIF was evaluated in simulated single- and dual-source HFO scenarios, where it was tasked with recovering both the spatial location and temporal dynamics of HFO generators under increasing spatiotemporal complexity. We further validated TH-DeepSIF in 25 patients with drug-resistant epilepsy by comparing EEG source imaged HFO sources against surgical resection regions. Results TH-DeepSIF accurately recovered both the spatial location and temporal dynamics of simulated HFO generators, achieving low localization error and strong waveform correspondence with ground truth. TH-DeepSIF localization for pathological HFOs (pHFOs, or spike ripples) demonstrated strong agreement with surgical resection regions, achieving a median localization error of 11.9 mm and specificity of 0.896. Compared with all HFOs (aHFOs), pHFO-based source imaging showed significantly stronger spatial correspondence with resection regions, significantly smaller localization error, and higher precision, sensitivity, geometric mean, and F1 score. Significance These findings demonstrate that TH-DeepSIF provides a robust, data-driven framework for noninvasive HFO source imaging with improved anatomical specificity and enhanced clinical utility for presurgical evaluation using scalp EEG. Moreover, they show that pathological HFOs (spike ripples )—rather than general HFOs—serve as robust EEG biomarkers for accurate localization of the epileptogenic zone. Key Points A fully data-driven, parameter-free framework for noninvasive HFO source imaging using scalp EEG. Pathological HFO source maps exhibit strong spatial concordance with both the surgical resection region and the seizure onset zone. Source imaging based on pathological HFOs consistently outperforms imaging based on all detected HFOs.

Abstract Despite promising results, it remains unclear how to optimally target and personalize closed-loop stimulation to ameliorate deficits in memory and other cognitive functions. We hypothesized that evoked connectivity – the measurement of neural pathway activation using single pulses of electrical stimulation – can guide patient-specific selection of stimulation location and parameters for memory. We characterized brain-wide evoked connectivity profiles of stimulation in memory-related brain regions recorded from 81 patients undergoing intracranial monitoring for epilepsy, showing that greater evoked connectivity between the lateral temporal cortex and the broader memory network (including the mesial temporal lobe, limbic regions and prefrontal cortex) corroborates observations of memory improvement by lateral temporal cortex stimulation. We first found that the lateral temporal cortex, compared to other stimulated regions, evokes the greatest and most distributed connectivity response throughout other memory-related regions. Evoked connectivity in downstream regions is greatest when stimulating a previously identified optimal target for memory improvement, bordering white matter at the rostrocaudal center of the middle temporal gyrus. Evoked connectivity corroborates other biomarkers of memory improvement by closed-loop stimulation, including increased resting-state functional connectivity between stimulated and recorded sites and increased modulation of oscillatory power. These results provide insight into the network mechanisms of stimulation for memory and suggest that evoked connectivity can more broadly predict the functional effects of closed-loop stimulation to prospectively guide targeting and parameter selection.

Miller et al. maintain that thalamic SEEG can now be considered a cornerstone of standard clinical practice in individualized epilepsy care, helping to define seizure onset and spread, assess therapeutic stimulation, map connectivity, and identify target-specific biomarkers for responsive neurostimulation.

Noninvasive electrophysiological source imaging (ESI) is a valuable tool for localizing and imaging brain activity, with significant potential to aid presurgical planning in focal drug-resistant epilepsy (fDRE) patients. Scalp electroencephalography (EEG) biomarkers, including interictal spikes, high-frequency oscillations (HFOs), and seizures, each offer unique capabilities in estimating the epileptogenic zone (EZ). However, there is a limited quantitative understanding of how these biomarkers differ in source-imaging precision, requiring distinct processing pipelines. Here, we developed a spatial–temporal–spectral imaging (STSI) framework for precision source imaging, and quantitatively evaluated various epilepsy biomarkers for source imaging in 2,081 individual events (spikes, HFOs, and seizures) from a cohort of 42 fDRE patients, comparing results to clinical ground truth such as surgical resection outcomes and intracranial EEG-defined seizure onset zones. The STSI enabled quantitative comparisons across key EEG epilepsy-related biomarkers, with averaged localization errors of 6.67 mm for seizures, 8.73 mm for HFOs overlapping with spikes (pHFO), 10.28 mm for HFO-riding spikes (pSpike), 19.59 mm for general spikes (aSpike), and 36.53 mm for general HFOs (aHFO), respectively, for seizure-free patients. These findings indicate that HFOs overlapping with spikes is the most spatially accurate interictal biomarker for mapping the EZ. The proposed STSI framework not only establishes a unified analysis approach for epileptic biomarkers to enhance presurgical planning in focal drug-resistant epilepsy, but could also generalize as a versatile tool for mapping event-related potentials, neural oscillations, and dynamic brain states, within a single framework to advance cognitive neuroscience research and clinical management of neurological and psychiatric disorders.

Objective The effects of deep brain stimulation (DBS) manifest across multiple timescales, spanning seconds to months, and involve direct electrical effects, neuroplasticity, and network reorganization. In epilepsy, the delayed impact of DBS on seizures presents challenges for optimization. Single‐pulse stimulation and resulting brain stimulation evoked potentials (BSEPs) provide a means to assess effective connectivity and network excitability. This study integrates BSEPs and short trials of DBS during stereoelectroencephalography (sEEG) to map seizure network engagement, modulate network dynamics, and monitor excitability and interictal abnormalities for biomarker informed neuromodulation. Methods Ten individuals with drug resistant epilepsy undergoing clinical sEEG were enrolled in this retrospective cohort study of epilepsy neuromodulation biomarkers. Each patient underwent a trial of high frequency (145Hz) thalamic DBS. BSEPs were acquired before and after DBS trials. Baseline BSEP amplitude assessed seizure network engagement, and modulation of amplitude (pre vs post DBS) assessed change in network excitability. Interictal epileptiform discharges were tracked by an automated classifier. Results Baseline BSEPs delineated distinct patterns of network engagement between thalamic subfields with maximal frontotemporal engagement achieved with stimulation of the anterior nucleus of the thalamus‐ventral anterior nucleus junction. DBS delivered for >1.5 hours reduced BSEP amplitudes compared to baseline, and the degree of modulation correlated with baseline connectivity strength. Shorter DBS trials did not induce reliable BSEP amplitude suppression, but did immediately suppress interictal epileptiform discharge rates in well‐connected seizure networks. Interpretation BSEPs and trials of DBS during sEEG provide novel network biomarkers to evaluate the modulation of large‐scale networks across multiple timescales, advancing biomarker informed neuromodulation. ANN NEUROL 2026;99:748–760

Abstract Objective This study reports development and validation of a deep learning seizure detection algorithm for two‐channel subscalp electroencephalographic (EEG) recordings. Ultra‐long‐term monitoring of people with epilepsy may produce new insights into the timing and pattern of their seizures and may pave the way for novel therapeutic options. Although EEG is the accepted standard for epilepsy monitoring, ultra‐long‐term EEG recordings generate a massive quantity of data and are not reasonably reviewable in full by human readers. Methods The convolutional neural network‐bidirectional long short‐term memory (CNN‐BiLSTM) hybrid algorithm uses nine layers, operating on channel spectrograms 5 min in length with 50% overlap. Retrospective subscalp EEG data from 16 patients recorded at three centers were available for algorithm development and testing. EEG was recorded for a median of 63 days (range = 9–508), and a median of 17 seizures (range = 0–96) were recorded. Training data were augmented with scalp EEG seizures, and performance was benchmarked against a conventional spectral power classifier algorithm. We also evaluated an intrapatient training–testing approach where the algorithm was trained on the first 45% of data. Results The CNN‐BiLSTM trained on scalp EEG and subscalp EEG achieved an area under the receiver operating characteristic curve (AUROC) of .98 and an area under the precision–recall curve (AUPRC) of .50, which correspond to 94% sensitivity with 1.11 false detections per day. The same model trained on iEEG achieved only AUROC = .94 and AUPRC = .36. The conventional spectral band power detector achieved AUROC = .93 and AUPRC = .38. The CNN‐BiLSTM detector trained on the earliest half of the subscalp EEG data achieved AUROC = .93 and AUPRC = .37, corresponding to 87% sensitivity and 5.9 false detections per day. Significance High sensitivity and specificity are possible in automated seizure detection in two‐channel subscalp EEG data using a CNN‐BiLSTM framework. Performance of the detector is superior using subcutaneous EEG data for training rather than intracranial EEG, but addition of scalp EEG seizures for training was helpful.

Objective: To evaluate the feasibility and effectiveness of wrist-worn wearable devices combined with machine learning (ML) approaches for detecting a diverse array of seizure types beyond generalized tonic–clonic (GTC), including focal, generalized, and subclinical seizures. Materials and Methods: Twenty-eight patients undergoing inpatient video-EEG monitoring at Mayo Clinic were concurrently monitored using Empatica E4 wrist-worn devices. These devices captured accelerometry, blood volume pulse, electrodermal activity, skin temperature, and heart rate. Seizures were annotated by neurologists. The data were preprocessed to experiment with various segment lengths (10 s and 60 s) and multiple feature sets. Three ML strategies, XGBoost, deep learning models (LSTM, CNN, Transformer), and ROCKET, were evaluated using leave-one-patient-out cross-validation. Performance was assessed using area under the receiver operating characteristic curve (AUROC), seizure-wise recall (SW-Recall), and false alarms per hour (FA/h). Results: Detection performance varied by seizure type and model. GTC seizures were detected most reliably (AUROC = 0.86, SW-Recall = 0.81, FA/h = 3.03). Hyperkinetic and tonic seizures showed high SW-Recall but also high FA/h. Subclinical and aware-dyscognitive seizures exhibited the lowest SW-Recall and highest FA/h. MultiROCKET and XGBoost performed best overall, though no single model was optimal for all seizure types. Longer segments (60 s) generally reduced FA/h. Feature set effectiveness varied, with multi-biosignal sets improving performance across seizure types. Conclusions: Wrist-worn wearables combined with ML can extend seizure detection beyond GTC seizures, though performance remains limited for non-motor types. Optimizing model selection, feature sets, and segment lengths, and minimizing false alarms, are key to clinical utility and real-world adoption.