Intra- and Inter-Subject Perspectives on the Detection of Focal Onset Motor Seizures in Epilepsy Patients

Abstract

Focal onset epileptic seizures are highly heterogeneous in their clinical manifestations, and a robust seizure detection across patient cohorts has to date not been achieved. Here, we assess and discuss the potential of supervised machine learning models for the detection of focal onset motor seizures by means of a wrist-worn wearable device, both in a personalized context as well as across patients. Wearable data were recorded in-hospital from patients with epilepsy at two epilepsy centers. Accelerometry, electrodermal activity, and blood volume pulse data were processed and features for each of the biosignal modalities were calculated. Following a leave-one-out approach, a gradient tree boosting machine learning model was optimized and tested in an intra-subject and inter-subject evaluation. In total, 20 seizures from 9 patients were included and we report sensitivities of 67% to 100% and false alarm rates of down to 0.85 per 24 h in the individualized assessment. Conversely, for an inter-subject seizure detection methodology tested on an out-of-sample data set, an optimized model could only achieve a sensitivity of 75% at a false alarm rate of 13.4 per 24 h. We demonstrate that robustly detecting focal onset motor seizures with tonic or clonic movements from wearable data may be possible for individuals, depending on specific seizure manifestations.

Keywords: digital health; eHealth; epilepsy; mHealth; mobile health; multimodal; seizure detection; wearables.

Figure A1

The Empatica E4 wrist-worn wearable device used in this study (left), and the Android phone application that connects to the wearable via Bluetooth and records the data stream (right).

Figure 1

Overview of how the feature and baseline windows were chosen, for the three different groups of features by modality. This calculation would result in one feature vector, for the next the windows would all be shifted by an interval of T = 2 s to the right. Abscissa not to scale.

Figure 2

Data set flowchart of the participant selection process. KCL: King’s College London; UKF: University Medical Center Freiburg; E4: Empatica E4 wrist-worn wearable device.

Figure 3

Selection of examples of true positive detections for each of the three participants in the intra-subject evaluation. Seizures shown are: (a) UKF1-4; (b) UKF2-3; (c) and KCL1-3 (see Table A1). Due to the grace period of 2 minutes around a seizure event, the detection for KCL1-3 counts as a true positive. Each plot of a seizure shows the raw ACC signal (top), the raw EDA signal and feature 2b (middle), and the estimated heart rate and signal quality index of the BVP signal (bottom). The regions highlighted in red mark the ground truth as labeled by experts, and those highlighted in green mark the seizure intervals, as predicted by the respective model, trained on the data of all the other seizures of the participant. The seizure onset and offset are additionally marked by the black vertical bars. All signals shown are normalized between −1 to 1 only for these plots. The original value ranges before normalization can be found in Table A3.

Figure 4

Feature importance scores per intra-subject evaluation for the seizure detection models of the three selected participants: (a) UKF1; (b) UKF2; (c) KCL1 (see Table A2); (d) Feature importance scores of the model resulting from training the GTBM model on the seizure data of all three inter-subject training participants. Blue, red, and yellow bars show the importance scores for the features grouped by biosignal modality ACC, EDA, and BVP, respectively. Horizontal lines mark the mean scores of the groups. The ordinate is unitless; the scores can be interpreted qualitatively. The feature labels correspond to the listing of features in the Materials and Methods.

Figure 5

Seizure UKF2-2, a false negative. Compare also to Figure 3. Data shown from top to bottom: raw ACC, raw EDA and feature 2b, heart rate and BVP signal quality index. The red overlay is the seizure ground truth. The seizure onset and offset are additionally marked by the black vertical bars. All signals shown are normalized between −1 to 1 only for these plots. The original value ranges before normalization can be found in Table A3.