Fast spike detection in EEG using eigenvalue analysis and clustering of spatial amplitude distribution

Area of Science:

• Neuroscience
• Computational Neuroscience
• Signal Processing

Background:

• Spike detection is crucial for analyzing neural signals.
• Existing methods can be computationally intensive.
• A need exists for faster, efficient spike detection algorithms.

Purpose of the Study:

• To develop and evaluate a fast spike detection method using eigenvalue analysis and clustering.
• To assess the performance of the proposed method in terms of precision and recall.
• To determine the processing time efficiency compared to recording duration.

Main Methods:

• Eigenvalue analysis of gradients from neighboring samples to detect negative peaks.
• Clustering of detected peaks based on amplitude distribution at scalp electrodes.
• Scoring of negative peaks using electrode information and cluster assignment.
• Spike detection based on score threshold and number of clusters.
• Performance evaluation using precision and recall metrics.

Main Results:

• A tradeoff between precision and recall was observed.
• Maximum average recall reached 0.90 in two subjects.
• Average precision was 0.21, with a false positive rate significantly higher than the true positive rate.
• The method completed spike detection in approximately one-eighth of the recording time.

Conclusions:

• The proposed method offers a computationally efficient approach to spike detection.
• Further refinement is needed to improve precision while maintaining high recall.
• The speed advantage makes this method promising for real-time neural signal analysis.