Inter-ictal spike detection using a database of smart templates.

OBJECTIVE: Visual analysis of EEG is time consuming and suffers from inter-observer variability. Assisted automated analysis helps by summarizing key aspects for the reviewer and providing consistent feedback. Our objective is to design an accurate and robust system for the detection of inter-ictal epileptiform discharges (IEDs) in scalp EEG.
METHODS: IED Templates are extracted from the raw data of an EEG training set. By construction, the templates are given the ability to learn by searching for other IEDs within the training set using a time-shifted correlation. True and false detections are remembered and classifiers are trained for improving future predictions. During detection, trained templates search for IEDs in the new EEG. Overlapping detections from all templates are grouped and form one IED. Certainty values are added based on the reliability of the templates involved.
RESULTS: For evaluation, 2160 templates were used on an evaluation dataset of 15 continuous recordings containing 241 IEDs (0.79/min). Sensitivities up to 0.99 (7.24fp/min) were reached. To reduce false detections, higher certainty thresholds led to a mean sensitivity of 0.90 with 2.36fp/min.
CONCLUSION: By using many templates, this technique is less vulnerable to variations in spike morphology. A certainty value for each detection allows the system to present findings in a more efficient manner and simplifies the review process. SIGNIFICANCE: Automated spike detection can assist in visual interpretation of the EEG which may lead to faster review times.