An Intrinsically Explainable Method to Decode P300 Waveforms from EEG Signal Plots Based on Convolutional Neural...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Deep learning models for electroencephalography (EEG) signal decoding often function as "black boxes", limiting clinical interpretability.
• P300 waveforms are crucial neural signals for brain-computer interface (BCI) applications, particularly for communication.

Purpose of the Study:

• To develop an intrinsically explainable method for decoding P300 waveforms from EEG signals.
• To enable convolutional neural networks (CNNs) to process EEG data visualized as images for enhanced interpretability.
• To implement and validate a P300-based speller for individuals with amyotrophic lateral sclerosis (ALS).

Main Methods:

• EEG signals were transformed into image representations for simultaneous visual interpretation and CNN analysis.
• A P300-based speller system was developed using the image-based EEG decoding method.
• The method was validated on a public dataset comprising EEG data from ALS patients.

Main Results:

• The proposed method successfully identified P300 signatures in EEG data from 8 ALS patients.
• The P300-based speller achieved comparable letter identification rates to existing state-of-the-art methods.
• The approach provides clinically relevant explainability (XAI) in EEG signal decoding.

Conclusions:

• The developed method offers a straightforward and explainable approach to decoding P300 signals using CNNs.
• This technique enhances the potential of BCI spellers as alternative communication tools for ALS patients.
• The integration of visual interpretability with AI-driven decoding represents a significant advancement in BCI research.