A Hidden Markov Factor Analysis Framework for Seizure Detection in Epilepsy Patients

Approximately 1% of the world population suffers from epilepsy. Continuous long-term electroencephalographic (EEG) monitoring is the gold-standard for recording epileptic seizures and assisting in the diagnosis and treatment of patients with epilepsy. Detection of seizure from the recorded EEG is a laborious, time consuming and expensive task. In this study, we propose an automated seizure detection framework to assist electroencephalographers and physicians with identification of seizures in recorded EEG signals. In addition, an automated seizure detection algorithm can be used for treatment through automatic intervention during the seizure activity and on time triggering of the injection of a radiotracer to localize the seizure activity. In this study, we developed and tested a hidden Markov factor analysis (HMFA) framework for automated seizure detection based on different features such as total effective inflow which is calculated based on connectivity measures between different sites of the brain. The algorithm was tested on long-term (2.4-7.66 days) continuous sEEG recordings from three patients and a total of 16 seizures, producing a mean sensitivity of 96.3% across all seizures, a mean specificity of 3.47 false positives per hour, and a mean latency of 3.7 seconds form the actual seizure onset. The latency was negative for a few of the seizures which implies the proposed method detects the seizure prior to its onset. This is an indication that with some extension the proposed method is capable of seizure prediction