Abstracts

Rationale: Patients with Alzheimer's disease (AD) have a 6 to 10-fold increased risk of developing epilepsy compared to their peers. Seizures in patients with AD most likely arise from the mesial temporal lobe (mTL), the highly epileptogenic region of the brain that is most profoundly affected by pathological changes in AD. Activity from the mTL is notoriously difficult to detect on scalp EEG, and entire "scalp-negative" seizures can arise from the mTL without showing a clear ictal correlate on scalp EEG. Currently, the only way to detect this activity is by using invasive intracranial recordings. We recently studied a 67 year old woman with amnestic mild cognitive impairment (aMCI) and CSF biomarkers consistent with AD, who had a history of unusual spells of confusion. She underwent scalp EEG monitoring for 5 days without capturing any seizures. She was then implanted with foramen ovale electrodes which, within 12 hours, captured 3 subclinical, scalp-negative mTL seizures arising from sleep. Whether these subclinical, scalp-negative seizures are common in patients with AD is unknown, because these patients almost never undergo the intracranial recordings needed to detect these events. Here, we present a method to non-invasively detect scalp-negative mTL seizures, using scalp EEG-based network connectivity measures. Methods: We identified 25 scalp-negative mTL seizures from 10 patients, and obtained control records from an additional 13 patients, who underwent simultaneous recordings with foramen ovale electrodes and scalp EEG. We extracted coherence features from these scalp EEGs and trained logistic regression classifiers to recognize scalp EEG coherence patterns that occurred specifically during these seizures. Results: Using a leave-one-patient-out cross-validation scheme (estimates how well the detector performs on new patients), our detector correctly identified scalp-negative seizures in 40% of patients with scalp-negative seizures, and correctly identified the side of seizure onset for all seizures detected. In comparison, routine clinical interpretation of these scalp EEGs failed to identify any of the scalp-negative seizures. 80% of patients in whom the detector raised seizure alarms actually had scalp-negative mTL seizures. The detector had a false alarm rate of only 0.3 per day and a positive predictive value of 75%. Of the 13 control patients, false alarms were raised in only one patient. We applied our detector to the scalp EEG data from our patient above with aMCI and detected all 3 of her seizures. Conclusions: Our scalp-negative mTL seizure detector provides the first opportunity to non-invasively study scalp-negative mTL seizures in patients with epilepsy, as well as in a wide population of patients who may be suspected of having cryptic mTL seizures, but in whom more invasive recordings are difficult to justify. In particular, we believe that this approach has the potential to provide important insights into the role epilepsy in Alzheimer's disease. Funding: NIH R25NS065743 (ADL). NIH NINDS R01-NS062092, K24-NS088568 (SSC).