A Robustness Comparison of Two Algorithms Used for EEG Spike Detection

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Scalp electroencephalography (EEG) is crucial for identifying epileptogenic networks and understanding the central nervous system.
• Automatic detection of neural transients like spikes and sharp waves requires a reliable gold standard.
• Intracerebral EEG data with Gaussian noise serve as a robust gold standard, mimicking scalp EEG output.

Purpose of the Study:

• To compare the robustness of Discrete Wavelet Transform (DWT) and Continuous Wavelet Transform (CWT) methods for detecting epileptiform transients.
• To evaluate the sensitivity and selectivity of DWT and CWT under decreasing Signal-to-Noise Ratios (SNR).
• To provide insights into the performance of these methods for analyzing noisy EEG data.

Main Methods:

• Utilized intracerebral EEG data mixed with Gaussian noise to simulate scalp EEG conditions.
• Applied two automatic detection methods: Discrete Wavelet Transform (DWT) and Continuous Wavelet Transform (CWT).
• Assessed method performance by varying the Signal-to-Noise Ratio (SNR) from 10 dB down to -10 dB.

Main Results:

• The Discrete Wavelet Transform (DWT) approach demonstrated superior stability in sensitivity, consistently detecting spikes as SNR decreased.
• The Continuous Wavelet Transform (CWT) approach exhibited greater stability in selectivity, effectively rejecting false spikes.
• Both methods showed varying performance based on SNR, highlighting the challenges of artifact and noise in EEG analysis.

Conclusions:

• DWT is more robust for sensitive detection of neural transients in low SNR conditions.
• CWT is more effective for selective detection, minimizing false positives in noisy EEG.
• The choice between DWT and CWT depends on whether sensitivity or selectivity is prioritized for epileptiform transient detection.