Neural fingerprinting on MEG time series using MiniRocket

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Signal Processing

Background:

• Neural fingerprinting identifies individuals via brain activity recordings.
• Current M/EEG methods often use second-order statistics, ignoring temporal dynamics.
• Magnetoencephalography (MEG) and electroencephalography (EEG) are key neuroimaging techniques.

Purpose of the Study:

• To evaluate advanced multivariate time series classification for neural fingerprinting.
• To assess the efficacy of the ROCKET classifier on short MEG resting-state data segments.
• To improve individual identification accuracy and efficiency in neuroimaging.

Main Methods:

• Applied the RandOm Convolutional KErnel Transformation (ROCKET) classifier.
• Utilized short time segments (1-second windows) of resting-state MEG data.
• Tested on a cohort of 124 subjects for individual classification.

Main Results:

• Achieved classification accuracies exceeding 99% for subject identification.
• Demonstrated superior performance compared to previous neural fingerprinting methods.
• Required significantly shorter time segments for accurate classification.

Conclusions:

• Multivariate time series classification, specifically ROCKET, enables highly accurate neural fingerprinting.
• This approach effectively captures individual temporal dynamics in MEG data.
• The method offers a more efficient and accurate way for individual identification using neuroimaging.