Differentiation between epileptic and functional/dissociative seizures using density spectral array of ictal single-channel EEG with deep learning.

Abstract

Differentiating epileptic from non-epileptic seizures is the first step in the diagnosis of epilepsy. We investigated whether the density spectral array (DSA) of ictal EEG could differentiate between these seizures using a deep learning technique. We retrospectively reviewed consecutive patients with mesial temporal lobe epilepsy (mTLE) and functional/dissociative seizures (FDS) and analyzed seizures recorded using long-term video-EEG monitoring. The time of clinical onset was defined as zero, and the EEG recordings were clipped from -3 to + 3 min. Frequency analyses of Cz as well as means of C3 and C4, Fp1 and Fp2, O1 and O2, and all electrodes were performed to generate DSA with a linked-ear reference. ResNet34, which is a convolutional neural network (CNN) model, was trained and tested on these datasets. This study included 48 patients with mTLE and 51 with FDS. The CNN architecture was created using 40 patients with mTLE (91 seizures) and 40 with FDS (82 seizures) as training data, while eight patients with mTLE (15 seizures) and 11 with FDS (18 seizures) were evaluated using the model as test data. Exploratory analyses revealed that the Cz electrode and the Middle 1/3 interval yielded the highest area under the curve among the reduced-electrode settings (0.941), which was statistically confirmed by pre-specified DeLong tests after Bonferroni correction. The DSA of a single-channel EEG (Cz) successfully differentiated between epileptic and non-epileptic seizures using deep learning. These results highlight the potential of this approach as a practical adjunct to early screening and triage, particularly in resource-limited settings.