Medicine (RMH) - Theses
Permanent URI for this collection
Search Results
1 - 1 of 1
-
ItemMobile and non-invasive devices as a diagnostic tool for physicians in detecting and classifying different seizure types( 2021)Introduction Epilepsy is one of the most common neurological disorders, affecting approximately 1 to 2% of the population. Epileptic seizures are caused by abnormal electrophysiological changes in the brain. In contrast, psychogenic non-epileptic seizures (PNES) are a class of seizures that are involuntary events, not caused by abnormal electrical discharges, but are thought to be a manifestation of psychological distress. The similarities in the observed behaviour between convulsive epileptic and convulsive psychogenic non-epileptic seizures (PNES) pose diagnostic difficulties. 20% of convulsive epileptic seizures are still misdiagnosed as PNES, delaying appropriate treatment for psychogenic seizures by an average of 5 to 7 years. This causes a poor prognosis and quality of life, as well as significant financial and social consequences. The current gold standard method to differentiate between the two is long-term Video Electroencephalography Monitoring (VEM), which is expensive and confined to few specialist centres. Therefore, a timely and accurate method of diagnosis outside the hospital setting is needed. Non-invasive and ambulatory devices are one such option being explored in this field, to not only detect convulsive epileptic seizures and PNES accurately, but also non-convulsive epileptic seizures. These devices typically measure physiological parameters such as 3-dimensional (3D) accelerometry, surface electromyographic signals (sEMG), heart rate, either separately or together, depending on the device. This approach is being recognised as automated seizure detection. This thesis aimed to investigate one such device, using 3D accelerometry, to differentiate between convulsive epileptic seizures and PNES. A further meta-analysis was conducted to quantify the accuracy of the devices, parameters and algorithms used in published studies for detecting all seizure types, including focal, generalised and psychogenic non-epileptic seizures Method In the first study, an off-the-shelf wireless, ambulatory device (Apple iPod) measuring 3D accelerometry was tested to investigate its sensitivity and false alarm rates for generalised tonic-clonic seizures (GTCS) and PNES. For this, consenting patients in the video-EEG monitoring (VEM) unit during their admission, were recruited, with the wireless device attached to their upper limbs. The data from the accelerometers went through an automated process using a machine learning algorithm, whereby convulsive activity detected was classified as a GTCS or PNES. The automated output for each seizure detected was compared to the corresponding VEM diagnosis to determine the device’s reliability using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and Cohen’s Kappa test. In the second study, a systematic review and meta-analysis was conducted involving two independent reviewers, to identify studies reporting the performance of mobile, non-invasive devices for the automated seizure detection of various seizure types. Limitations of these studies are acknowledged, and the directions for future studies are proposed. Results The overall finding of this thesis was that most non-invasive, ambulatory devices investigated for automated seizure detection have high sensitivities and acceptable false alarm rates for GTCS and PNES. This was evident in the first study where 13 PNES from five patients and 11 motor epileptic seizures were recorded during video-EEG monitoring. The sensitivity for detecting PNES and GTCS using 3D accelerometry was found to be 100% and 72.7%, respectively, with a FAR of 2.4 per 24 hours. The systematic review found that few studies have investigated the utility of ambulatory devices in the automated detection of focal seizures or the combination of both focal and GTCS. Moreover, the use of differing algorithms for each study, was found to increase heterogeneity. This implied that results between each study could not be directly compared, even when similar parameters were being used. This may hinder the ability to investigate these devices from moving to phase 4 studies where they may be tested in an out of hospital setting, the ultimate goal of utilising automated seizure detection. Conclusion The automated wireless device accelerometer tested in this study was shown to be sensitive in detecting and classifying both GTCS and PNES. There are a limited number of studies investigating the automated detection of focal seizures that are experienced by the majority of patients with epilepsy. The inability to compare studies due to the use of varied machine learning algorithms, may limit the use of such devices in phase 4 studies. It is essential that future studies involve larger population groups and determine automated detection for focal seizures. Furthermore, using pre-defined algorithms that have already displayed high sensitivities and low false alarm rates, particularly for the automated detection of GTCS and PNES, should be a priority for future research.