Abstracts

Rationale: There is growing interest in using responsive neurostimulation to treat seizures in individuals with Lennox-Gastaut Syndrome (LGS). The centromedian nucleus of the thalamus has emerged as a promising target due to its widespread reciprocal connectivity with the cortex and its role in generalized seizure networks. The feasibility of responsive thalamocortical neurostimulation to treat seizures in LGS depends on detection of interictal and ictal activity in these regions. In a clinical trial evaluating thalamocortical responsive neurostimulation in LGS (NCT05339126), two bilaterally implanted neurostimulators record intracranial EEG (iEEG) from thalamic and cortical leads, enabling detailed analysis of electrographic seizure patterns.

Methods: Long episode (LE) iEEG recordings from the RNS® System, characterized by prolonged abnormal activity, were obtained from 15 patients with two independent devices. 25000 LE recordings were analyzed, with each recording consisting of 4 channels of iEEG activity. Within each patient, individual iEEG channels were clustered to identify distinct seizure patterns. For clustering, iEEG signals were first transformed into time-frequency spectrogram representations and then passed through a convolutional neural network (ResNet-50). High-dimensional embeddings corresponding to individual iEEG channels were generated from the network output. These embeddings were subsequently reduced in dimensionality using principal component analysis (PCA) followed by t-distributed stochastic neighbor embedding (t-SNE). Finally, clustering was performed using a Bayesian Gaussian mixture model to identify seizure patterns within each patient.

Results: Distinct clusters were identified within LE iEEG recordings from all patient devices. An average of 8 ± 2 (mean±SD) clusters were observed (Fig 1). Visual inspection revealed that cortical seizure patterns were predominantly characterized by low-voltage fast activity and high-amplitude rhythmic discharges, while thalamic seizure patterns were characterized by low-amplitude sharp/rhythmic activity < 13 Hz (Fig 2).

Conclusions: These findings show that abnormal iEEG activity in patients with LGS can be reliably detected and characterized using LE iEEG recordings from the RNS System. Distinct electrographic seizure patterns were identified within individual patients through unsupervised clustering, with clear differentiation between cortical and thalamic iEEG channel patterns. In the future, AI models trained on such data may enable automated classification of electrographic seizure onset patterns. Longitudinal analysis of these clusters may reveal how seizure dynamics evolve over time within patients, and whether specific seizure patterns are more responsive to certain types of neurostimulation. This could ultimately support more personalized neurostimulation therapies for individuals with LGS.

Funding: Research reported in this publication was supported by the National Institute Of Neurological Disorders And Stroke of the National Institutes of Health under Award Number UH3NS109557. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.