EEG Lossy Compression and its impact on EEG-based pattern recognition

Abstract

Electroencephalogram (EEG) has been widely used in EEG-based pattern recognition
systems such as epileptic seizure, sleep stage, emotion, alcoholics and person recognitions.
However, one of the major challenges of EEG is the huge amounts of data that
need to be processed, transmitted and stored. Developing effective EEG compression
algorithms is therefore necessary. As EEG lossy compression algorithms achieve
a much higher Compression Ratio (CR) than lossless ones, most studies related to
EEG compression focus on lossy algorithms. Studies also indicate that EEG signals
under brain disorders, for example, epilepsy are different from the normal EEG, especially
on frequency. Numerous lossy compression algorithms have been proposed
with none being reported as the best algorithm. This project focuses on developing
an EEG lossy compression algorithm to maximize CR while minimising the loss
of information. This research also focuses on developing an EEG lossy compression
algorithm for epileptic EEG signals.
EEG lossy compression algorithms allow advanced CR compared to lossless ones;
however data is lost in the reconstructed signals including diagnosing and biometric
information, which may negatively affect EEG-based applications. Little work has
been done in evaluating the effect of lossy compression on EEG-based seizure recognition
systems. Hence, this research evaluates the impacts of EEG lossy compression
on EEG-based pattern recognition systems including person, seizure, alcoholics, age,
and gender recognition systems.
Evaluation experiments conducted on a wide range of public EEG datasets show
that the proposed EEG lossy compression algorithms give better compression performances
than some recent lossy compression algorithms. In addition, lossy compression
algorithms do have the impact on EEG-based pattern recognition systems
as the recognition performances decrease when compression increases. However, it is feasible to apply lossy compression to EEG-based pattern recognition systems and
using lossy compression is still more advantageous than using lossless approaches.

Date of Award	2019
Original language	English
Supervisor	Wanli Ma (Supervisor) & Dat TRAN (Supervisor)