Algorithm for Detection of Convulsive Seizures in Severe Encephalitis Based on Multiscale Compressed Shapelets

Encephalitis is a serious disease for neurological dysfunction caused by inflammation of the brain parenchyma. Recurrent convulsive seizures or nonconvulsive status are the main causes of many neurological sequelae. Accurately identifying convulsive seizure signals in continuous electroencephalogram (EEG) signals collected from patients with severe encephalitis can help doctors effectively make diagnoses and give treatment plans. Traditionally, fourfold-scale compressed amplitude-integrated electroencephalography (aEEG) is used for detection. In this article, a novel convulsive seizure detection method of encephalitis based on multiscale aEEG signal is proposed. First, continuous EEG (cEEG) signal is converted into multiscale aEEG. Second, multiscale shapelets from multiscale aEEG signals are extracted to form a multiscale ictal waveform codebook. Third, the dynamic time warping (DTW) algorithm is used to calculate the similarity between the codebook and the actual waveform. Finally, random forest (RF) classifier is applied to train and test the method. In the dataset collected, the accuracy, sensitivity based on the event, specificity, $F1$ score, mean absolute error, missed detection, false detection, and false positive ratio based on the event obtained by the proposed method reach 96.32%, 95.70%, 97.22%, 96.85%, 0.037, 4.71%, 3.70%, and 0.06 times/h, respectively.