An Improved Sparse Representation over Learned Dictionary Method for Seizure Detection

• DOI: 10.1142/s0129065715500355
• Publication Date: 2015-09-15T05:44:05Z
• Authors: Junhui Li; Weidong Zhou; Shasha Yuan; Yanli Zhang; Chengcheng Li; Qi Wu
• Keywords: Epileptic seizure

Automatic seizure detection has played an important role in the monitoring, diagnosis and treatment of epilepsy. In this paper, a patient specific method is proposed for long-term intracranial electroencephalogram (EEG) recordings. This based on sparse representation with online dictionary learning elastic net constraint. The learned could sparsely represent testing samples more accurately, constraint which combines 11-norm 12-norm not only makes coefficients but also avoids over-fitting problem. First, EEG signals are preprocessed using wavelet filtering differential filtering, kernel function applied to make closer linearly separable. Then dictionaries nonseizure respectively from original ictal interictal training optimization algorithm compose dictionary. After that, test coded over residuals associated sub-dictionary calculated, respectively. Eventually, classified as two distinct categories, or nonseizure, by comparing reconstructed residuals. average segment-based sensitivity 95.45%, specificity 99.08%, event-based 94.44% false rate 0.23/h latency -5.14 s have been achieved our method.