Algorithmic clustering based on string compression to extract P300 structure in EEG signals.

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• P300 is a crucial Event Related Potential for Brain Computer Interfaces (BCIs).
• Variability in P300 signal structure across subjects and time poses a detection challenge.
• P300 spellers utilize the oddball paradigm for human-computer communication.

Purpose of the Study:

• To address P300 signal variability using algorithmic clustering.
• To identify the underlying structure of P300 signals.
• To explore the efficacy of string compression for P300 analysis.

Main Methods:

• Utilized Normalized Compression Distance (NCD) for structure extraction.
• Developed a novel signal-to-ASCII process for event data transformation.
• Applied hierarchical clustering and multidimensional projection for analysis.

Main Results:

• Demonstrated good clustering performance, highlighting structure extraction capabilities.
• Validated results using two distinct datasets recorded in varied scenarios.
• Identified potential for the approach to serve as an electrode-selection criterion.

Conclusions:

• NCD-driven clustering effectively discovers structural characteristics of EEG signals.
• The methodology is a suitable complementary tool for P300 analysis.
• The approach aids in understanding and utilizing P300 signals more robustly.