A reliable documentation of seizure frequency is essential for an optimal epilepsy therapy. However, patients are often not aware of their seizures or might document non-epileptic events. Devices for automatic seizure detection would thus not only allow to monitor treatment effects, but may additionally enable an objective seizure quantification, early therapeutic intervention and SUDEP (Sudden Unexpected Death in Epilepsy) prevention.
In the last few years, an increasing number of automatic seizure detection approaches has been developed, using accelerometry, electrodermal activity, or heart rate based algorithms. An informative review on currently available systems and their applicability to different seizure types is given by Ulate-Campos and co-workers1. They point out that most sensors are primarily sensitive to generalized tonic-clonic seizures (GTCS) which are more readily detectable due to their extensive electro-clinical manifestation with vigorous movements, pronounced muscle activation, and considerable changes in autonomic function. Small wearable devices like an ear-EEG detection tool seem to identify GTCS with a high sensitivity based on surface EMG measurement2. In contrast, only few systems are also capable to detect the more subtle focal seizures. De Cooman and co-workers presented an adaptive seizure detection algorithm based on the patient´s heart rate to detect not only GTCS but also focal tonic, clonic, or hyperkinetic seizures3. To improve the diagnostic yield of the seizure detection systems it would be desirable to not only detect seizures but also to differentiate between different seizure subtypes. By implementing Persyst quantitative EEG spectrograms, Goenka and his group identified specific electrographic signatures for focal seizures, focal seizures with secondary generalization, and generalized seizures4. Our study went one step further looking for quantitative parameters which can differentiate between different lobes of seizure origin5. For this purpose we used the audio signal of focal seizures and performed a quantitative sound intensity analysis. This revealed specific sound characteristics differentiating between seizures of frontal, temporal, or parieto-occipital lobe onset. Wearable watch audio recordings were already used by Velez and co-workers to distinguish GTCS from false positive events and this might also be applicable to automatic vocalization analysis6.
Integral system devices, which combine several seizure detection approaches, are thus very promising to address different needs: a reliable detection and prediction of seizures and subsequent optimization of closed-loop approaches may contribute to decrease mobidity and mortality in epilepsy. Further, considering the costs and availability of epilepsy monitoring units, integral seizure detection devices might contribute valuable complementary data in the presurgical assessment of patients with focal epilepsy.
