Automatic Seizure Detection Using Sparse Representation Over Learned Dictionary

Epilepsy is a chronic neurological disorder, which is characterized by the sudden and excessive electrical discharges neuronal activities in the brain. The recurrent seizure has greatly influenced the life of epileptic patients in the field of body, psychology, emotion. Epilepsy results from excessive discharges in the brain neurons, so electroencephalogram can be used to record the changes of potential in the brain. Electroencephalogram (EEG) is an important tool for doctor to identify the seizure area in the brain. Epilepsy expert can analyze long-term EEG visually, but the EEG recording is usually very long and it is a heavy and time-consuming work in reality. In this condition, epilepsy tend to be too tired to analyze the EEG., and some mistakes may occur. With the development of information technology, the computers are used to analyze and monitor EEG recordings. Making the most of some detection algorithm, we can successfully detect seizure part in the recordings of EEG, which is valuable for helping doctors analyze EEG recordings and diagnose epilepsy. And at the same time we can also save lots of time. So automatic seizure detection has played an important role in the monitoring, diagnosis and treatment of epilepsyIn this paper, a novel method is proposed for patient specific seizure detection in the long-term intracranial EEG recordings based on sparse representation with online dictionary learning algorithm and elastic net constraint.In this study, we used online dictionary algorithm to construct more effective learning dictionary with the constraint of elastic net instead of using original training samples directly. Later, the test samples are sparsely coded over the learned dictionary. Lastly, we can get detection results by comparing the reconstructing residuals over ictal and interictal dictionary. In this improved sparse representation, the learned dictionary could sparsely represent the testing samples more accurately, and the elastic net constraint which combines the l1-norm and l2-norm not only makes the coefficients sparse but also avoids over-fitting problem.The seizure detection framework proposed in this paper is shown as follows. Firstly, the EEG signals are preprocessed using wavelet filtering and differential filtering, and the kernel function is applied to make the samples closer to being linearly separable. Then the dictionaries of seizure and non-seizure are respectively learned from original ictal and interictal training samples with online dictionary algorithm to compose the training dictionary. After that, the test samples are sparsely coded over the learned dictionary and the residuals associated with ictal and interictal sub-dictionary are calculated, respectively. Eventually, the test samples are classified as two distinct categories, seizure or non-seizure, which yields the class that minimizes reconstructed residual. According to the analysis and evaluation of this detection method, it can be found that our detection method has low computational cost, small amount of computation burden and good performance for real time system, which is meaning for the application of our method in clinical study.