EP 114. Uncovering epileptic seizures – A feasibility study for the semiological analysis of hidden patient motion during epileptic seizures

Abstract

Aims Semiological analysis of epileptic seizures is performed to categorize resection candidates and provides valuable information about lateralization and localization of the epileptogenic zone. Seizure related movements of inpatients are evaluated based on video-EEG recordings, but can be hidden during seizures through e.g. blanket occlusion, staff in front of the camera, image blurring or if the patient is outside of the field of view. Based on this problem, we propose the use of accelerometer sensors attached to the patient extremities, which allows the continuous extraction of motion features, even under occlusion. In this study we focus our analysis on the separation of ictal and postictal movement. Methods From 15 patients with 57 motor seizures, we selected the most suitable patient that consistently performed both ictal and postictal movements in all of his seizures (n = 6). Half of those occurred while the patient was completely covered by the blanket. The accelerometers (Shimmer) were attached to the wrists and ankles of the patient. The remaining three seizures allowed a clear view on the patient, allowing to manually track movements with a commercial video-based system (MaxTRAQ). Ictal and postictal phase of all seizures were determined by video-EEG analysis. Comparative measurements were made for the manual tracking and the accelerometer data. We developed a new metric for accelerometers, which we call agility, accounting for rotation as well as for rapid movement of the extremities. The agility is the sum over absolute changes in the acceleration vector divided by the elapsed time. Results To evaluate the metrics, we chose the threshold that provides 90% sensitivity for the ictal phase and report the respective specificity. Video-analysis In related works, the trajectory-length of a bodypart in 2D successfully distinguished movements of interest (MOIs) in frontal lobe and temporal lobe epilepsy. In this study, the trajectory-length in [pixels] was 1780 ± 702 during ictal and 935 ± 783 during postictal phase, yielding a specificity of 66.7% (threshold 1150). Furthermore, the covered area has previously been used to separate MOIs in hypermotor and automotor seizures. This area in [pixels2] reached 45,160 ± 44,563 during ictal and 16,535 ± 10,702 during postictal phase with a specificity of 61.1% (threshold 22,250). Accelerometer-analysis For hypermotor seizure detection, the standard deviation of the acceleration was used in current literature. In the three completely occluded seizures, this value in [m/s2] was 1.8 ± 0.6 during ictal and 2.5 ± 0.9 during postictal phase at a 55.6% specificity (threshold 2.8). The new agility metric in [Hz] yielded 3.1 ± 0.9 during ictal and 5.8 ± 3.0 during postictal phase, providing a separation specificity of 66.7% (threshold 4.2) at 90% sensitivity, thus matching the best video-based metric. Conclusions While video tracking and accelerometers measure different aspects of the seizure related movements, their accuracies are on par, showing that accelerometer data are suitable for the quantification of seizures occurring under a blanket. Interestingly, video metrics based on the range of movements are higher in the ictal phase, while the metrics we calculated from accelerometer data are higher in the postictal phase. We deem further research for these complementary quantification approaches extremely valuable, as it opens up new possibilities for continuous quantitative semiological analysis.