Heart Rate Variability Analysis for Seizure Detection in Neonatal Intensive Care Units

Abstract

In Neonatal Intensive Care Units (NICUs), the early detection of neonatal seizures is of utmost importance for a timely clinical intervention. Over the years, several neonatal seizure detection systems were proposed to detect neonatal seizures automatically and speed up seizure diagnosis, most based on the EEG signal analysis. Recently, research has focused on other possible seizure markers, such as electrocardiography (ECG). This work proposes an ECG-based NSD system to investigate the usefulness of heart rate variability (HRV) analysis to detect neonatal seizures in the NICUs. HRV analysis is performed considering time-domain, frequency-domain, entropy and multiscale entropy features. The performance is evaluated on a dataset of ECG signals from 51 full-term babies, 29 seizure-free. The proposed system gives results comparable to those reported in the literature: Area Under the Receiver Operating Characteristic Curve = 62%, Sensitivity = 47%, Specificity = 67%. Moreover, the system's performance is evaluated in a real clinical environment, inevitably affected by several artefacts. To the best of our knowledge, our study proposes for the first time a multi-feature ECG-based NSD system that also offers a comparative analysis between babies suffering from seizures and seizure-free ones.

Keywords: ECG; HRV; NICU; multiscale entropy; neonatal seizures.

Figure 1

Histogram showing seizure events’ duration and seizure events’ occurrence.

Figure 2

Unbalanced data in the experiments based on the segmentation of the ECG signal into the 60 s (A) and 180 s (B) epochs.

Figure 3

Example of the epoch-based metrics: comparison between the labeling made by the expert (A) and the time-windows classified by the system (B). In this example, the detector finds two false positive (FP), five false negative (FN), three true positive and five true negative (TN) values.

Figure 4

Example of the event-based metrics: comparison between the labeling made by the expert (A) and the time-windows classified by the system (B). An event is considered as correctly identified if the system detects at least one epoch during the event. In this example, the first two events are correctly identified, while the third one is not identified.

Figure 5

Example of the time delay metric: comparison between the labeling made by the expert (A) and the time-windows classified by the system (B). The time delay describes the time interval between the seizure detected by the algorithm and the seizure onset marked by the expert.

Figure 6

The features ranked with the mRMR algorithm for the experiment based on 180 s segmentation.

Figure 7

(A) Schematic representation of the Sensitivity values for the 22 pathological patients, iteratively obtained by the Linear SVM model trained on the full set of features during the LOSO cross-validation. A total of 10 out of 22 pathological patients are characterized by Sensitivity values >0. (B) Schematic representation of the Sensitivity values for the 22 pathological patients, iteratively obtained by the Linear SVM model trained on the subset of two features selected through the mRMR algorithm during the LOSO cross-validation. A total of 7 out of 22 pathological patients are characterized by Sensitivity values >0.

Figure 8

(A) Schematic representation of the Sensitivity values for the 22 pathological patients, iteratively obtained by the Gaussian SVM model trained on the full set of features during the LOSO cross-validation. A total of 15 out of 22 pathological patients are characterized by Sensitivity values >0. (B) Schematic representation of the Sensitivity values for the 22 pathological patients, iteratively obtained by the Gaussian SVM model trained on the subset of two features selected through the mRMR algorithm during the LOSO cross-validation. A total of 17 out of 22 pathological patients are characterized by Sensitivity values >0.

Figure 9

The concatenated ROC evaluated on the SVM Gaussian kernel-based system with highest AUC. All the recordings are linked together into a single recording.