Identification of nonconvulsive status epilepticus in the ictal-interictal continuum using artificial intelligence: a...

This study aims to investigate differences in functional connectivity between patients on the ictal-interictal continuum (IIC) with nonconvulsive status epilepticus (NCSE) versus those with coma-IIC, and to evaluate whether machine learning based on these connectivity measures was able to distinguish between these two groups.

We prospectively enrolled patients with IIC electroencephalography (EEG) patterns and classified them into NCSE or coma-IIC groups according to the Salzburg criteria and clinical information. We analyzed functional connectivity based on EEG using graph theory. For deep learning, EEG signals were transformed into time-frequency images using short-time Fourier transforms. We investigated differences in functional connectivity between the two groups.

We enrolled 72 patients on the IIC. Of the 72 patients, 53 patients had NCSE, and 19 had coma-IIC. Patients with NCSE had decreased global functional connectivity in all frequency bands compared to patients with coma-IIC. Global efficiency was significantly lower in the NCSE group than the coma-IIC group (e.g., gamma band, 0.258 vs 0.350; p = 0.001), and local efficiency was consistently decreased across all frequency bands in the NCSE group (all p ≤ 0.01). Machine learning based on these measures classified patients with NCSE and those with coma-IIC with an accuracy of 92.8%, while the accuracy of the convolutional neural network model to distinguish between them was 73.9%.

We demonstrated that graph-theoretical functional connectivity derived from EEG data differs significantly between patients with NCSE and those with coma-IIC. Furthermore, our results confirm the feasibility of using machine learning models based on these connectivity measures to effectively distinguish between these two conditions.