Automated spike and seizure detection: Are we ready for implementation?

Ambulatory seizure detection

PURPOSE OF REVIEW

To review recent advances in the field of seizure detection in ambulatory patients with epilepsy.

RECENT FINDINGS

Recent studies have shown that wrist or arm wearable sensors, using 3D-accelerometry, electrodermal activity or photoplethysmography, in isolation or in combination, can reliably detect focal-to-bilateral and generalized tonic-clonic seizures (GTCS), with a sensitivity over 90%, and false alarm rates varying from 0.1 to 1.2 per day. A headband EEG has also demonstrated a high sensitivity for detecting and help monitoring generalized absence seizures. In contrast, no appropriate solution is yet available to detect focal seizures, though some promising findings were reported using ECG-based heart rate variability biomarkers and subcutaneous EEG.

SUMMARY

Several FDA and/or EU-certified solutions are available to detect GTCS and trigger an alarm with acceptable rates of false alarms. However, data are still missing regarding the impact of such intervention on patients' safety. Noninvasive solutions to reliably detect focal seizures in ambulatory patients, based on either EEG or non-EEG biosignals, remain to be developed. To this end, a number of challenges need to be addressed, including the performance, but also the transparency and interpretability of machine learning algorithms.

Towards trustworthy seizure onset detection using workflow notes

A major barrier to deploying healthcare AI is trustworthiness. One form of trustworthiness is a model's robustness across subgroups: while models may exhibit expert-level performance on aggregate metrics, they often rely on non-causal features, leading to errors in hidden subgroups. To take a step closer towards trustworthy seizure onset detection from EEG, we propose to leverage annotations that are produced by healthcare personnel in routine clinical workflows-which we refer to as workflow notes-that include multiple event descriptions beyond seizures. Using workflow notes, we first show that by scaling training data to 68,920 EEG hours, seizure onset detection performance significantly improves by 12.3 AUROC (Area Under the Receiver Operating Characteristic) points compared to relying on smaller training sets with gold-standard labels. Second, we reveal that our binary seizure onset detection model underperforms on clinically relevant subgroups (e.g., up to a margin of 6.5 AUROC points between pediatrics and adults), while having significantly higher FPRs (False Positive Rates) on EEG clips showing non-epileptiform abnormalities (+19 FPR points). To improve model robustness to hidden subgroups, we train a multilabel model that classifies 26 attributes other than seizures (e.g., spikes and movement artifacts) and significantly improve overall performance (+5.9 AUROC points) while greatly improving performance among subgroups (up to +8.3 AUROC points) and decreasing false positives on non-epileptiform abnormalities (by 8 FPR points). Finally, we find that our multilabel model improves clinical utility (false positives per 24 EEG hours) by a factor of 2×.