EEG Seizure Detection with Low Ranked Shared Dictionary Learning and Slantlet Transform

With over 50 million people suffering from seizure related illnesses, automatic detection and prediction of seizures is an important issue in biomedical signal processing. In this paper, a new seizure detection method centered on the Low-Rank Shared Dictionary Learning algorithm is proposed. First, EEG signals are preprocessed using Slantlet transform to make the input signals more discriminative. Then dictionaries of seizure and nonseizure signals are respectively learned using seizure and nonseizure training samples with the dictionary learning algorithm. The experimental study shows that the proposed approach produces superior results with 95.91% sensitivity, 99.62 % specificity, and 99.53% accuracy, in comparison with those obtained with the state-of-the-art methods.