Semi-supervised one-class transfer learning for heart rate based epileptic seizure detection

Automated epileptic seizure detection in a home environment has been a topic of great interest during the last decade. Normally patient-independent heart rate based seizure detection algorithms are used in practice to avoid the necessity of patient-specific data. They, however, lead to mediocre performance due to the large inter-patient heart rate variability. Therefore these algorithms should be adapted to each patient in an efficient way. In this study, a patient-specific algorithm is constructed with only 1 night of not-annotated patient-specific data by using a transfer learning approach. The algorithm was evaluated on 8 pediatric patients with 25 strong nocturnal convulsive seizures. By using only 1 night of patient-specific data, the false alarm rate dropped by a factor of 4 compared to the patient-independent algorithm, leading to on average 0.76 false alarms per night and 88% sensitivity. The results show that the proposed method can quickly adapt to patient characteristics without the requirement of seizure annotations.