Multi-Center Validation Study of Automated Classification of Pathological Slowing in Adult Scalp Electroencephalograms Via Frequency Features

Pathological slowing in the electroencephalogram (EEG) is widely investigated for diagnosis of neurological disorders. Currently, gold standard detection visual inspection EEG by experts, which time-consuming and subjective. To address those issues, we propose three automated approaches to detect EEG: Threshold-based Detection System (TDS), Shallow Learning-based (SLDS), Deep (DLDS). These systems are evaluated on channel-, segment-, EEG-level. The perform prediction via detecting at individual channels, detections arranged histograms segment- We evaluate through Leave-One-Subject-Out (LOSO) cross-validation (CV) Leave-One-Institution-Out (LOIO) CV four datasets from US, Singapore, India. DLDS achieved best overall results: LOIO mean balanced accuracy (BAC) 71.9%, 75.5%, 82.0% EEG-level, LOSO BAC 73.6%, 77.2%, 81.8% channel- segment-level performance comparable intra-rater agreement (IRA) an expert 72.4% 82%. can process a 30 min 4 s be deployed assist clinicians interpreting EEGs.