An explainable and interpretable model for attention deficit hyperactivity disorder in children using EEG signals
Smith K Khare1, U Rajendra Acharya2
1Electrical and Computer Engineering Department, Aarhus University, 8200, Aarhus, Denmark.
Computers in Biology and Medicine
|February 24, 2023
Summary
This study introduces a novel explainable machine learning model for detecting Attention Deficit Hyperactivity Disorder (ADHD) in children using EEG signals, achieving high accuracy and providing interpretable insights for clinical use.
Area of Science:
- Neuroscience
- Machine Learning
- Biomedical Signal Processing
Background:
- Attention Deficit Hyperactivity Disorder (ADHD) is a neurodevelopmental disorder impacting sleep, mood, anxiety, and learning.
- Electroencephalogram (EEG) signals offer potential for ADHD detection but are complex and difficult to interpret visually or with standard machine learning models.
- Subtle differences in EEG signals between individuals with ADHD and healthy controls are challenging to discern, leading to unreliable machine learning model decisions.
Purpose of the Study:
- To develop and validate an explainable machine learning model for accurate ADHD detection in children using EEG data.
- To extract and analyze hidden information from EEG signals through a combination of signal processing techniques.
- To enhance the interpretability and explainability of ADHD detection models for clinical application.
Main Methods:
- Utilized Variational Mode Decomposition (VMD) and Hilbert Transform (HT) for feature extraction from EEG signals.
- Employed an Explainable Boosted Machine (EBM) model to classify 41 statistical parameters derived from EEG.
- Applied explainability techniques like LIME, SHAP, PDP, and Morris sensitivity for feature, channel, and brain region importance analysis.
Main Results:
- Achieved high diagnostic performance with 99.81% accuracy, 99.78% sensitivity, and 99.84% specificity.
- Demonstrated excellent model performance with an F-1 score of 99.83% and an Area Under the Curve (AUC) of 100%.
- Identified the frontal brain region as having the highest interpretability and explainability in ADHD detection.
Conclusions:
- The developed explainable model is highly reliable, robust, interpretable, and suitable for clinicians to detect ADHD in children.
- The findings provide crucial insights into the model's decision-making process, enhancing trust and clinical utility.
- Early and rapid ADHD diagnosis using explainable AI can potentially reduce treatment costs and lengthy diagnostic procedures.
