Reduction of the ERP Measurement Time by a Weighted Averaging Using Deep Learning

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Clinical electroencephalography (EEG) relies on averaging multiple responses to estimate event-related potentials (ERPs).
• This averaging process is time-consuming, limiting its efficiency in clinical settings.
• Background EEG noise can obscure important ERP components.

Purpose of the Study:

• To develop and evaluate a deep learning-based weighted-averaging method for shortening ERP measurement time.
• To assess if the proposed method can maintain ERP waveform shape and peak latency while enhancing P300 amplitude.
• To quantify the reduction in measurement time and background EEG suppression achieved by the new method.

Main Methods:

• A novel weighted-averaging technique using deep learning (CNN or EEGNet) was proposed.
• The method was applied to P300 component analysis, comparing its output to conventional averaging.
• Key metrics evaluated included waveform shape (correlation coefficient), peak P300 amplitude, peak latency, and background EEG suppression (RMS of pre-stimulus activity).

Main Results:

• The deep learning method successfully maintained ERP waveform shape and P300 peak latency compared to conventional averaging.
• The weighted-averaging approach resulted in a higher P300 peak amplitude.
• Using EEGNet, a 13.7% reduction in measurement time was achieved, equivalent to approximately 40 seconds per 5 minutes of data acquisition.

Conclusions:

• Deep learning-based weighted averaging offers a promising approach to accelerate ERP measurements in clinical EEG.
• This method allows for faster data acquisition without compromising the integrity of ERP waveforms or latency.
• The enhanced P300 amplitude and reduced measurement time demonstrate the clinical utility of this innovative technique.