Abstracts

Interictal spikes in intracranial EEG (IEEG) may correlate with epileptogenic cortex. However, visual review of IEEG is labor-intensive, and has not consistently correlated with pathology. Understanding the role of interictal spikes in epileptogenesis requires the development of accurate, automated IEEG spike detectors to map their spatiotemporal occurrences., Eight patients, 5-20 years, undergoing IEEG for intractable epilepsy were studied. Ten-minute segments, two channels per patient were analyzed by two human (H1, H2) and two automated (A1, A2) spike detectors. The automated detection algorithms employed a three point template detector (A1), and a prototype detector measuring peak and trough frequencies (A2). In the first of two experiments, identification, automated detection results were calibrated to a consensus set of detections defined by H1 and H2 co-marking each patient[apos]s records. Subsequently, H1 and H2 re-marked each segment independently. Spike identification accuracy was defined from positive predictive value and sensitivity, taking the consensus set of events as groundtruth, and were calculated for each detector for each segment. In a second experiment (verification), H1 and H2 independently reviewed all detections in a blinded fashion, and the resulting percent verification was calculated., Identification accuracy varied by patient between 3.21% to 66.28% for A1/A2, and 0.57% to 51.93% for H1/H2. The accuracy measures were similar between human and automated detectors within patients; however there were large discrepancies between patients. In 3 of 8 patients, automated detectors and humans had lower accuracies in patients with high-amplitude, sharply-contoured background IEEG. Humans verified 84.51[plusmn]4.10% of spikes marked by humans and 76.71[plusmn]6.56% of spikes found by the automated detectors. Again, no difference was found between spikes marked by humans or the detectors on verification., Both automated detectors performed comparably to human evaluators on both the identification and verification experiments. In a subset of patients with almost continuous rhythmic activity, identification was more subjective and variable than verification. These data suggest automated spike detection methods may perform adequately compared to humans on IEEG. Future use of these detectors for mapping spike occurrences in epileptic networks could reveal associations between spike distributions, seizure onset, and pathology., (Supported by CURE to BEP.)