Abstracts

Seizure Detection by Multi Extracerebral Biosignal Analysis

Abstract number : 3.084
Submission category : 1. Translational Research: 1D. Devices, Technologies, Stem Cells
Year : 2015
Submission ID : 2314012
Source : www.aesnet.org
Presentation date : 12/7/2015 12:00:00 AM
Published date : Nov 13, 2015, 12:43 PM

Authors :
Diana L. Cogan, Mehrdad Nourani, Jay Harvey, Venkatesh Nagaraddi

Rationale: Seizures can be detected by algorithms based on heart rate (HR) changes, but the false positive rate is high [Osorio, Ivan, “Automated Seizure Detection Using EKG,” International Journal of Neural Systems 2014;24(2):1450001]. In many patients, additional extracerebral biosignals change in response to seizures and can be used to improve HR based algorithm accuracy.Methods: Under IRB approval, two commercially available wristworn devices were used to collect HR, arterial oxygenation (SpO2) and electrodermal activity (EDA) from 20 patients electively admitted to a Dallas, Texas Epilepsy Monitoring Unit.Results: Of the 20 patients, 11 provided HR, SpO2 and EDA data on 24 seizures during 355 hours of data collection. Analysis of this data is summarized in Table 1. Seizure occurrence was established through EEG analysis by attending physicians. HR increased 15% or more in all 24 of the captured seizures. In 20 of these seizures, an SpO2 drop of 5% or more immediately followed the HR rise, creating an HR↑ + SpO2↓ combination. This combination is less likely to be found in non-seizure situations than is an increase in HR alone, as seen by the marked decrease in false positives (Table 1). Two CPS show less pronounced SpO2 disturbances which we plan to capture with more advanced (e.g. wavelet based) signal processing. In only 2 CPS is SpO2 undisturbed. Of the 20 seizures with HR↑ + SpO2↓ combinations, 12 show strong EDA responses (increases of at least 60% and 1μS), 6 show weak EDA responses (increases of at least 60% but less than 1μS), and 2 show no EDA response. Researchers have hypothesized that EDA responses to focal seizures are contralateral [Poh, Ming-Zher, “Continuous Assessment of Epileptic Seizures with Wrist-worn Biosensors,” PhD diss., Massachusetts Institute of Technology, 2011]. We plan to test this hypothesis by collecting data from both wrists; if it is correct, we will see stronger EDA responses on the wrist opposite the seizure focus. Fig. 1 illustrates multi extracerebral biosignal responses to a CPS (left) and a GTCS (right). Some HR and SpO2 data is missing because the sensor we are using to collect it is housed in a finger cuff, which can lose contact with or come off of the finger during a seizure. Several companies are working on a wrist based replacement for this device, and we plan to use one of them. We have observed that for some patients the relative timing of biosignal response to seizure is very similar for each seizure. We will present a pattern recognition algorithm later this year which further improves seizure detection accuracy for these patients.Conclusions: Seizure detection using HR, SpO2 and EDA signals is much more accurate than detection by HR changes alone. It will allow development of a comfortable, easy to manage wristworn device capable of recognizing seizures (convulsive or not), alerting a caregiver and creating an electronic diary for use by physicians. The device will classify possible seizure events based on probability tables developed from biosignal responses to known seizures for each patient. This research was partially funded by Texas Medical Research Collaborative (TxMRC).
Translational Research