A wavelet-packets based algorithm for EEG signal compression

Abstract

Transmission of biomedical signals through communication channels is being used increasingly in clinical practice. This technique requires dealing with large volumes of information, and the electroencephalographic (EEG) signal is an example of this situation. In the EEG, various channels are recorded during several hours, resulting in a great demand of storage capacity or channel bandwidth. This situation demands the use of efficient data compression systems. The objective of this work was to develop an efficient algorithm for EEG lossy compression. In this algorithm, the EEG signal is segmented and then decomposed through Wavelet Packets (WP). The WP decomposition coefficients are thresholded and those having absolute values below the threshold are deleted. The remaining coefficients are appropriately quantized and coded using a run-length coding scheme. The compressed EEG signal can be recovered by an inverse process. Extensive experimental tests were made by applying the algorithm to EEG records and measuring the compression rate (CR) and the distortion in signal segments. The WP transform showed a high robustness, allowing a reasonably low distortion after a compression-decompression process, for CR typically in the range 5-8. The algorithm has a relatively low computational cost, making it appropriate for practical applications.

Wearable Electroencephalography

“…Offline implementations of this technique have shown good results with approximately a 50% reduction of the raw data being achievable using lossless compression techniques [51]. Approximately 15% compression ratio is achievable when mildly lossy compression, where the original and end EEG signals do not match exactly, is deemed acceptable [52]. These levels are impressive, and the schemes should certainly be used where possible.…”

Toward Online Data Reduction for Portable Electroencephalography Systems in Epilepsy

“…It thus ranges from reasonably similar to the performance here, to noticeably better. Similarly, reported EEG data compression techniques (which aim to represent the EEG information in fewer digital bits, rather than the discontinuous method investigated here) report data reductions of approximately 40% using lossless encoding (based on Huffman coding) [34], and up to 90% data reduction when lossy transform-based compression (based upon thresholding wavelet coefficients) is used [35].…”

A Key Power Trade-off in Wireless EEG Headset Design

“…As an example, Huffman coding combined with derivative computation is shown to achieve a compression ratio of 0.4 for low computational complexity [14]. A lossy compression technique presented in [15] achieves compression ratios of 0.1 to 0.2. Another promising method is to utilise the mutual information that exists between channels [16].…”