Multichannel EEG compression based on ICA and SPIHT

• PCA-based ICA is proposed to exploit the inter-channel correlations.
• 1-D signal is arranged in the form of matrix to exploit the intra-channel correlations.
• SPIHT algorithm is used to compress the signals in matrix form.

In this paper, we propose a novel approach for the compression of multichannel electroencephalograph (EEG) signals. The method assumes that EEG signals are the linear mixture of several independent components (ICs). To retain the ICs, the proposed scheme first applies an independent component analysis (ICA) with a preprocessing step of principal component analysis (PCA) to EEG signals. Then the compression scheme is composed of two parts: the ICs compression part and the residue compression part. Each IC is arranged in the form of matrix and then compressed with the algorithm of set partitioning in hierarchical trees (SPIHT). The residue signals are compressed in the same way as ICs, but with a higher compression ratio (CR). The appropriate combination of compression ratios of the ICs and the residue is explored to achieve desired performance. The compression scheme is tested with eight datasets sampled at two different frequencies. The experimental results demonstrate the high compression performance of the proposed approach and its potential usage in the EEG related telemedicine applications.