Seizure information enrichment in ECG through spectral whitening for improving epileptic seizure prediction

Using electrocardiogram (ECG) as an alternative to electroencephalogram (EEG) for seizure prediction is attractive for its ease of use and cost-effectiveness, but it is not yet popular for its poor performance. In this work, we refine the ECG to enrich the seizure-related information to improve the seizure prediction accuracy.

We use spectral whitening (SW) to remove the information related to the normal functioning of the heart, lungs, brain and other organs (heart++) from the ECG. Preictal intervals of the ECG with heart++ signals removed will be rich in information about the seizure and can enhance seizure prediction accuracy.

The prior state-of-the-art system using the Temple University hospital seizure (TUHSZ) dataset for EEG reports an area under the receiver operating characteristics (AUROC) of 0.84, which outperforms the system using ECG with an AUROC of 0.63. Our system using spectral whitening gave an accuracy of 99.94 %, a sensitivity of 99.69 %, a specificity of 99.96 %, and an AUROC of 0.99 which is an improvement of 0.36 in AUROC compared to the existing literature using the TUHSZ database. Similar results were obtained using the Siena scalp database and the post-ictal heart rate oscillations in partial epilepsy (PIHROPE) database.

We have achieved a very high seizure prediction accuracy using ECG analysis. Our results excel any previously published results, showing the superiority in the seizure prediction capability 20 min prior to the seizure.

We proposed a high-accuracy seizure prediction system using ECG as its input. Unlike using EEG, ECG-based algorithms are suitable for developing ambulatory devices for their ease of wearing and low complexity. Wearable ECG gadgets help address the stigma associated with the visibility of the sensors and improve the confidence of the patient.