Optimized Complex-Valued Spatio-Temporal Graph Convolutional Networks for attention deficit hyperactivity disorder
R Lakshmi1, Vanathi Balasubramanian2
1Department of Artificial Intelligence and Data Science at SRM Valliammai Engineering College, Chennai, Tamil Nadu, India.
Computers in Biology and Medicine
|March 25, 2026
Summary
A new method uses optimized complex-valued spatio-temporal graph convolutional networks (CSTGCN) to detect Attention Deficit Hyperactivity Disorder (ADHD) in pediatric EEG signals with high accuracy. This approach improves upon existing models for ADHD and learning disability detection.
Area of Science:
- Neuroscience
- Machine Learning
- Biomedical Signal Processing
Background:
- Attention Deficit Hyperactivity Disorder (ADHD) is a neurodevelopmental condition impacting various cognitive and emotional functions.
- Electroencephalogram (EEG) signals are crucial for ADHD diagnosis but present challenges like complexity and non-stationarity.
- Existing diagnostic methods for ADHD and Learning Disabilities (LD) using EEG signals have limitations in feature interpretability and accuracy.
Purpose of the Study:
- To propose an optimized complex-valued spatio-temporal graph convolutional network (CSTGCN) for accurate detection of ADHD in pediatric EEG signals.
- To enhance the classification performance for both ADHD and Learning Disabilities (LD) by addressing signal complexity and feature extraction challenges.
- To introduce a novel framework integrating advanced signal processing and deep learning for improved neurodevelopmental disorder detection.
Main Methods:
- Preprocessing of EEG signals using Multi-Window Savitzky-Golay Filter (MWSGF) for noise reduction and Synchro Transient Extracting Transform (STET) for feature extraction.
- Utilizing a Complex-Valued Spatio-Temporal Graph Convolutional Network (CSTGCN) for classifying EEG signals into ADHD/No-ADHD and LD/No-LD categories.
- Employing the Red-Billed Blue Magpie Optimization Algorithm (RBBMO) to optimize the weights of the CSTGCN model for enhanced performance.
Main Results:
- The proposed CSTGCN-ADHD-EEG framework achieved a high accuracy of 99.46% and a precision of 98.32% in classifying pediatric EEG signals.
- The model demonstrated superior performance compared to existing methods in detecting both ADHD and Learning Disabilities.
- Effective noise and artifact removal were achieved through MWSGF and STET, leading to robust feature extraction.
Conclusions:
- The optimized CSTGCN framework offers a highly accurate and effective method for detecting ADHD and LD in pediatric EEG data.
- This advanced deep learning approach overcomes limitations of traditional EEG analysis for neurodevelopmental disorders.
- The findings suggest significant potential for the CSTGCN-ADHD-EEG model in clinical diagnostic applications for children.
