Epileptic seizure detection using heart rate variability

Epileptic seizures are recurring brief episodes of abnormal excessive or synchronous neuronal activity in the brain, and are often accompanied by changes in various autonomic functions like heart rate (HR). A better approach for detecting epileptic seizures is by using electrocardiogram (ECG) signals because ECG acquisition is relatively easier as compared to EEG. In this paper a new technique is proposed for detection of seizures in epileptic patients using the electrocardiogram (ECG) signal. Feature sets for analysis of HRV (heart rate variability) comprises of parameters from multiple domains. For temporal analysis activity, mobility and complexity features are identified and for spectral analysis mean of absolute deviation of Fast Fourier Transform coefficients and spectral entropy are identified for seizure detection. These features are classified by using two different approaches i.e. by setting threshold and by using linear support vector machine where average latency by threshold approach was found to be better than linear SVM. The performance parameters for the proposed technique using threshold approach for classification are accuracy (94.2%), sensitivity (84.1%) and specificity (94.5%) which shows that the proposed algorithm detects epileptic seizures efficiently. Comparison of performance of this model was done with those proposed earlier using ECG signal and this model was found to be better.