Early Seizure Detection Based on Cardiac Autonomic Regulation Dynamics

Figures

• 

  Figure 1. Proposed framework for early seizure detection. The methodology comprises five stages: the acquisition of an ECG recording from the patient, the HRV signal extraction from the recording, the linear and non-linear analysis of the HRV time series, the pre-processing of the parameters derived from the HRV analysis, and the use of a classifier algorithm to identify any preictal states.

• 

  Figure 2. Relationship between Wo and Wp. RR interval data were extracted from multiple overlapping observation windows of Wo seconds in the ECG recording. Subsequently, a set of HRV parameters was calculated and inserted into prediction matrix X (which is shown in the shaded area of the overall matrix), with duration Wp (i.e., the duration of the period analyzed by the classifier). k, the last step of the overlapping observation window slide in a ECG sample; S, step, in seconds, that is the overlapping observation window slides through between the current extraction period and the start of the next one; Wo, observation window, in seconds, the length of the ECG segment used to extract the HRV parameters; Wp, prediction window, in seconds, the duration of the period analyzed by the classifier; xp(n), the pth parameter extracted in the nth sample.

• 

  Figure 3. Dynamics of HRV parameters. Each panel presents dynamic changes in one particular HRV parameter during four seizures and one interictal period experienced by the same patient. Clinical/EEG seizure onset is represented by the vertical red lines. Each of the signals associated with seizure-related findings has been synchronized to seizure onset. CSI, cardiosympathetic index; CVI, cardiovagal index; SampEn, sample entropy; SDNN, standard deviation of all normal RR intervals.

• 

  Figure 4. Dynamics of HRV parameters. Dynamic changes of four HRV parameters, from 12 patients, with the signals synchronized to seizure onset. Clinical/EEG seizure onset is represented by the vertical red lines. CVI, cardiovagal index; CSI, cardiosympathetic index; SampEn, sample entropy; SDNN, standard deviation of all normal RR intervals.

• 

  Figure 5. Early detection of seizures. Each chart presents the results of the early detection/prediction model, with positive output (indicating seizure prediction) highlighted using shading. Clinical/EEG seizure onset is represented by the vertical red lines. The top two panels show successful early prediction/detection of seizures in two different patients. The bottom panels are from a third patient: the left panel shows a successful seizure detection the right panel an example of a seizure which was not detected.

• 

  Figure 6. False positive rate of seizure prediction. Each chart presents the results of the early detection/prediction model, with positive output (indicating seizure prediction) highlighted using shading. Each of the graphs show the HRV signal from a different ECG recording of an interictal period.