EEG-based multi-feature fusion assessment for autism.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Psychiatry

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental condition impacting social communication and cognitive abilities.
• Current diagnostic methods for ASD can be subjective and lack objective biomarkers.
• Identifying reliable biomarkers is crucial for early and accurate diagnosis.

Purpose of the Study:

• To develop an objective method for autism spectrum disorder (ASD) diagnosis using electroencephalogram (EEG) signals.
• To identify key EEG-based biomarkers for ASD through multi-feature fusion and machine learning.
• To evaluate the classification accuracy of the proposed method for ASD detection.

Main Methods:

• Utilized a multi-feature fusion technique to extract diverse features from EEG signals, including power spectrum analysis, bicoherence, entropy, and coherence.
• Employed the minimum redundancy maximum relevance (mRMR) algorithm for optimal feature selection.
• Input selected features into three distinct classifiers, including SVM-linear, to assess classification performance.

Main Results:

• A combination of nine selected EEG features, processed by an SVM-linear classifier, yielded a high classification accuracy of 91.38% for ASD.
• The multi-feature fusion approach effectively captured complex patterns within EEG signals relevant to ASD.
• The study demonstrated the potential of specific EEG features as reliable biomarkers for ASD.

Conclusions:

• The proposed multi-feature fusion method based on EEG signals offers a promising objective approach for the clinical diagnosis of autism spectrum disorder (ASD).
• The identified EEG biomarkers and classification model provide a foundation for developing more accurate and accessible diagnostic tools for ASD.
• Further validation and research are warranted to integrate this method into clinical practice for autism diagnosis.