Detection of an Autism EEG Signature From Only Two EEG Channels Through Features Extraction and Advanced Machine...

Area of Science:

• Neuroscience
• Computational Biology
• Medical Technology

Background:

• Previous studies demonstrated high accuracy in distinguishing autism spectrum disorder (ASD) from typical development using standard electroencephalogram (EEG) data.
• The feasibility of using limited EEG derivations, as available in neonatal units, for ASD detection remained to be investigated.

Purpose of the Study:

• To determine if two EEG derivations (C3 and C4) are sufficient for accurate ASD detection using advanced machine learning.
• To assess the performance of machine learning algorithms on limited EEG data compared to previous findings.

Main Methods:

• Utilized 1-minute artifact-free EEG segments from C3 and C4 channels from prior studies.
• Extracted 1588 quantitative features using the Python tsfresh package.
• Employed a hybrid machine learning system (TWIST) combining evolutionary algorithms and neural networks for feature selection and classification.

Main Results:

• Identified 12 key EEG features from C3-C4 data in study 1 and 36 in study 2.
• Achieved 100% accuracy in classifying ASD vs. typical development in study 1.
• Attained 94.95% accuracy in classifying ASD vs. other neuropsychiatric disorders in study 2.

Conclusions:

• A limited EEG segment, when analyzed with advanced computational algorithms, contains significant information for autism detection.
• This approach holds promise for early ASD identification in newborns using standard neonatal EEG equipment.
• The findings support the potential for non-invasive, early screening of ASD signatures at birth.