Best practice for single-trial detection of event-related potentials: Application to brain-computer interfaces

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Event-related potentials (ERPs) are key in electroencephalogram (EEG) analysis for brain-computer interfaces (BCIs) and cognitive neuroscience.
• Single-trial ERP detection offers deeper insights into brain dynamics than grand averages.
• Existing methods often require complex processing and numerous hyperparameters, limiting application with limited data.

Purpose of the Study:

• To review state-of-the-art methods for single-trial ERP detection, focusing on techniques suitable for BCI applications.
• To identify efficient methods with minimal hyperparameters, applicable to datasets with limited trials.
• To benchmark classification methods for single-trial ERP detection.

Main Methods:

• Review of current single-trial ERP detection techniques, including temporal filtering, spatial filtering, and classification.
• Focus on methods with few hyperparameters and ease of implementation for limited datasets.
• Benchmarking of various classification methods using EEG data from a rapid serial visual presentation task.

Main Results:

• Single-trial ERP detection achieved an area under the ROC curve > 0.9.
• Effective detection was possible with fewer than ten sensors and 20 trials.
• The number of trials is a critical factor for robust classifier performance, more so than the number of sensors.

Conclusions:

• Efficient single-trial ERP detection is feasible for BCI applications using methods with minimal hyperparameters.
• Robust classification requires careful selection of the number of trials.
• This research provides a benchmark for selecting appropriate methods for single-trial ERP detection in BCI.