Abstracts

Rationale: Neonatal seizures are difficult to diagnose and treat due to subtle clinical presentation, ambiguities in electro-clinical correlation and frequently pharmacological intractability. Seizure detection is crucial for timely diagnosis, seizure burden assessment, treatment, and prevention. We aim to assess differences in the electroencephalogram (EEG) and electrocardiogram (ECG) signals between patients with and without prolonged neonatal seizures.


Methods: We studied patients with prolonged neonatal seizures (≥ 3 min long) recorded with simultaneous ECG and EEG (seizure group) and patients without seizures (controls) from an anonymized and publicly available dataset of patients with neonatal seizures¹. From the seizure group, we selected 3 min of data at ictal onset from each recorded seizure. Seizures with muscular or technical artifacts on the EEG or EGC data in the 3-min window were excluded. Similarly, we selected 3-min segments of artifact-free EEG/ECG data from the control group. From the ECG data, we extracted RR intervals using R-DECO software¹, and calculated the following features of heart rate variability (HRV): standard deviation of RR intervals (SDRR), and root mean square of successive differences (RMSSD). From the EEG (bipolar longitudinal Montage, 0.3–30 Hz bandpass), we quantified the power difference between the brain hemispheres using the wavelet-based pairwise derived brain symmetry index (pdBSI)³ and calculated left and right brain hemisphere power differences (Figure 1). Finally, we compared each EEG and ECG feature between the seizure and the control group (Mann–Whitney U test).


Results: We included 40 neonates (gestational age: 35-43 weeks; 19 seizure group; 21 controls) and analyzed 58 control (NC) and 55 seizure (SZ) segments. All extracted features from EEG and ECG signals are significantly different between groups (p< 0.05). Patients with seizures had higher pdBSI than controls (1454.6 vs 1420.8, p< 0.005) suggesting greater hemispheric power differences at seizure onset (Figure 1a). HRV instead was lower in the seizure than control group (SDRR: 0.0134 s vs 0.0217 s, p< 0.05 Figure 1b; RMSSD: 0.003 vs 0.009, p< 10^-5, Figure 1c). The ratio between pdBSI and RMSSD was higher in the seizure than the control group (0.43×10³ vs 0.151×10³, p< 10^-5) demonstrating strong group separation when combining EEG and ECG features.


Conclusions: Neonates with prolonged seizures present with a decrease in EEG and ECG signal variability. Estimation of pdBSI to RMSSD ratio (as an integrated measure of EEG and ECG variability) may serve as a marker for detection of prolonged neonatal seizures. Further validation work is needed to confirm our results in larger patient populations and different age groups.

-Stevenson et al. A dataset of neonatal EEG recordings with seizure annotations. Scientific Data, 2019. - Moeyersons et al. R-DECO: An open-source Matlab based graphical user interface for the detection and correction of R-peaks. Peerj Comput Sci. 2019. - Wilkinson et al. Predicting stroke severity with a 3-min recording from the Muse portable EEG system for rapid diagnosis of stroke, Sci Rep. 2020.


Funding: Funding: This study was in part supported by the Epilepsy Research Fund.