Advanced machine learning techniques reveal multidimensional EEG abnormalities in children with ADHD: a framework for
Ying Mao1,2, Xuchen Qi2,3,4, Lingyan He5
1Department of Special Examination, Shaoxing Peoples' Hospital, Shaoxing, Zhejiang, China.
Frontiers in Psychiatry
|March 3, 2025
Summary
Machine learning accurately identified electroencephalography (EEG) patterns in children with Attention Deficit Hyperactivity Disorder (ADHD). This research enhances understanding of ADHD
Area of Science:
- Neuroscience
- Computational Biology
- Medical Informatics
Background:
- Attention Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopmental disorder impacting attention and impulse control in children and adults.
- Electroencephalography (EEG) offers potential for novel diagnostic tools and personalized treatments for ADHD.
- Understanding EEG characteristics in pediatric ADHD is crucial for advancing diagnostic and therapeutic strategies.
Purpose of the Study:
- To explore distinctive EEG features in children with ADHD using advanced machine learning and feature selection.
- To identify brain function abnormalities in pediatric ADHD compared to healthy controls through a data-driven approach.
- To utilize the SHapley Additive exPlanations (SHAP) algorithm for interpreting machine learning model decisions regarding ADHD classification.
Main Methods:
- Extracted multidimensional EEG characteristics, including power spectral density (PSD), fuzzy entropy (FuzEn), and mutual information (MI) functional connectivity.
- Employed four machine learning algorithms: random forest (RF), XGBoost, CatBoost, and LightGBM for classification tasks.
- Applied the SHAP algorithm to assess feature importance and interpret the classification models.
Main Results:
- Achieved a 99.58% classification accuracy for pediatric ADHD detection using the CatBoost model with 206 optimal features (53 PSD, 5 FuzEn, 148 MI).
- Identified increased theta, alpha, and beta rhythm power, an elevated theta/beta ratio (TBR), and enhanced whole-brain functional connectivity in children with ADHD.
- The SHAP analysis highlighted the contribution of specific EEG features to accurate ADHD classification.
Conclusions:
- Machine learning effectively extracts EEG features crucial for differentiating pediatric ADHD from healthy controls.
- Multidimensional EEG characteristics provide insights into the electrophysiological mechanisms underlying ADHD.
- These findings support the potential for automated ADHD diagnosis using advanced EEG analysis.
