This study introduces a novel multi-headed convolutional neural network (CNN) architecture for the classification of electroencephalographic (EEG) signals. The proposed model effectively distinguishes interictal EEG segments of patients with epileptic seizures (ES), psychogenic non-epileptic seizures (PNES) and healthy controls (HC). To achieve high classification accuracy, the network exploits temporal and connectivity features. Temporal features are extracted through a dedicated CNN branch, while connectivity features derived from the Phase Lag Index (PLI) are processed through a separate network input. This multimodal fusion strategy leads to an improvement in classification performance. The multi-headed CNN achieves an accuracy of 88% (± 2%). These findings underline the strategic role of incorporating connectivity features in the diagnosis and potentially in the treatment of neurological disorders.