Implementation and evaluation of an interictal spike detector

The detection of epileptiform activity, such as interictal spikes, in electrical brain recordings has important clinical and research applications. Identification of interictal spikes is often carried out manually by trained neurologists. It is a time-consuming process and can exhibit variability between experts. In this work, we develop and evaluate an automated spike detector. We implement a template-matching approach and improve its accuracy on one set of recordings using a supervised machine-learning algorithm. Evaluation with two independent datasets shows the template-matching detector to perform comparably with experts and the version augmented with a classifier. In one test dataset, variations of the detection threshold partially explain discrepancies between experts. In the other, the detector demonstrates similar behavior to an existing algorithm developed with this dataset.