Pattern Recognition in Human Sleep Electroencephalogram Using Sparse Optimization

In this paper, we addressed the decomposition problem of electroencephalogram and the detection of sleep spindles and K-complexes. We propose an approach to decompose the electroencephalogram into three components, including transient, low-frequency, and oscillation. The decomposition problem can be attributed to the convex optimization. The objection function of the optimization included the 1 norm of the transient and the Fourier transform coefficient of the oscillation. Based on the methods of the Douglas-Rachford variable splitting and the Alternating Direction Method of Multipliers, the electroencephalogram is decomposed efficiently. The detection of sleep spindles and K-complexes is achieved by applying the Teager-Kaiser Energy Operator on the oscillation and the low-frequency signal, respectively. The performance of the proposed approach is tested using a human DREAMS dataset. The average spindle detection accuracy of our method is 0.97 and the F1 score is 0.69. The average K-complex detection accuracy of our method is 0.97 and the F1 score is 0.46.