A quantitative approach to evaluating interictal epileptiform discharges based on interpretable quantitative criteria

Area of Science:

• Computational neuroscience
• Clinical neurophysiology
• Machine learning in medicine

Background:

• Accurate identification of interictal epileptiform discharges (IEDs) is crucial for epilepsy diagnosis.
• Existing methods for IED identification rely on subjective interpretation of electroencephalogram (EEG) waveforms.
• Quantifying established criteria for IEDs can improve diagnostic consistency.

Purpose of the Study:

• To develop a quantitative algorithm for the six International Federation of Clinical Neurophysiology (IFCN) criteria for IED identification.
• To estimate the probability of a candidate waveform being epileptiform based on quantified features.
• To create a user-friendly interface for clinical application.

Main Methods:

• An algorithm was designed to identify five fiducial landmarks of candidate IEDs.
• The algorithm quantifies six IFCN features from these landmarks.
• A machine learning model was trained using these features to predict epileptiform probability.

Main Results:

• The developed model achieved excellent performance (AUROC = 0.88, calibration error = 0.03).
• Waveform asymmetry was the most discriminative feature (coefficient = 0.64), while duration was least (coefficient = 0.09).
• Model performance was lower than human experts and a deep neural network (SpikeNet, AUCROC = 0.97).

Conclusions:

• The approach provides quantifiable measures of IFCN criteria for IED identification.
• The model offers an interpretable and accessible method to assess the epileptiform likelihood of waveforms.
• This tool can assist electroencephalographers and trainees in IED identification.