Neonatal Seizures Detection using Stationary Wavelet Transform and Deep Neural Networks: Preliminary Results

Abstract

The increasing use of Electroencephalography (EEG) in the field of pediatric neurology allows more accurate and precise diagnosis of several cerebral pathologies, mainly in Neonatal Intensive Care Units (NICUs), where it represents the gold-standard for the diagnosis of neonatal epileptic seizures. However, EEG interpretation is time consuming and requires highly specialized staff. For this reason, in the last years there was a growing interest in the development of systems for automatic and fast detection of neonatal epileptic seizures. To this aim, we propose here hybrid systems that combines techniques related to the Stationary Wavelet Transform (SWT) as a support to deep-learning algorithms such as Convolutional Neural Networks and Fully Convolutional Networks. The proposed methods are validated on a public dataset of NICUs seizures recorded at the Helsinki University Hospital. Results are encouraging both in terms of Area Under the receiveroperating Curve, AUC (81%), Good Detection Rate, GDR (77%) and False Detection per hour, FD/h (1.6). Actually, the SWT step increases the performance of the proposed methods of about 5% for the AUC as compared to considering the raw EEG time-series only. These results, though preliminary, represent a significant step forward in solving the problem of neonatal seizure detection.

Automatic Detection of Epileptic Seizures in Neonatal Intensive Care Units Through EEG, ECG and Video Recordings: A Survey

“…A dataset of 55 newborns was considered for the training step, coming from 3 centers: Montreal Children's Hospital, Montreal, Canada; Sydney Children's Hospital, Sydney, Australia; Texas Children's Hospital, Houston, Texas. The testing dataset was composed of EEG signals from 9 newborns at the Montreal Children's Hospital, Montreal, Canada; 14 newborns at the Sydney Children's Hospital, Sydney, Australia; 18 newborns at the Texas Children's Hospital, Houston, Texas [44]. The EEG signals were segmented into 10 s epochs with 75% overlap.…”

“…The choice of the features in data-driven methods is a crucial operation as it determines the classifiers' performances. The need for feature extraction can be overcome by introducing deep-learning algorithms that do not require hand-designed features [44] into 90 s epochs, with 60 s overlap. A CNN algorithm is proposed that does not need hand-designed features.…”

Analysis of Brain-Heart Interactions in newborns with and without seizures using the Convergent Cross Mapping approach

“…For 22 patients, the experts did not find any seizures, and we considered them seizure-free patients. In this study, we analyzed only the patients with unanimous consensus [19,20]. Thus, the remaining 18 patients were excluded.…”

MS4PS: A Mentor-Student Architecture for Patient-Specific Seizure Detection With Combination of Transfer Learning and Active Learning

“…In this paper, the raw data format is chosen for the input of model, although there are other commonly used formats such as frequency features [26] and spectrograms [27]. For each EEG segment, the 18 channels of 1-D time-domain signals are stacked in a specific order to form a 2-D matrix (Fig.…”