Spectral change or Jensen gap? Log-ratio baseline correction for time-frequency M/EEG is negatively biased

Area of Science:

• Neuroscience
• Signal Processing

Background:

• Time-frequency analysis of M/EEG data commonly uses baseline correction to assess spectral changes.
• Log-ratio baseline correction, widely used in software, divides spectral power by baseline power and log-transforms the ratio.

Purpose of the Study:

• To identify and address the negative bias inherent in log-ratio baseline correction.
• To propose and validate an unbiased alternative for M/EEG spectral analysis.

Main Methods:

• Introduced a
• mean-log-ratio
• correction method by subtracting mean log-transformed baseline power from target window log-transformed power.
• Evaluated methods using real and simulated M/EEG data.

Main Results:

• Log-ratio correction demonstrated a negative bias, misrepresenting true spectral power increases as decreases.
• This bias was observed across different software implementations.
• The proposed mean-log-ratio method and simple mean-subtraction correction did not exhibit bias.

Conclusions:

• Log-ratio baseline correction is not recommended due to its inherent negative bias.
• The mean-log-ratio correction offers a viable, unbiased alternative for M/EEG spectral analysis.
• Accurate baseline correction is crucial for reliable interpretation of M/EEG results.