Refining ADHD diagnosis with EEG: The impact of preprocessing and temporal segmentation on classification accuracy
Sandra García-Ponsoda1, Alejandro Maté2, Juan Trujillo1
1Lucentia Research Group - Department of Software and Computing Systems, University of Alicante, Rd. San Vicente s/n, San Vicente del Raspeig, 03690, Spain; ValgrAI - Valencian Graduate School and Research Network of Artificial Intelligence, Camí de Vera s/n, Valencia, 46022, Spain.
Computers in Biology and Medicine
|November 1, 2024
Summary
Preprocessing and segmenting electroencephalogram (EEG) data improves attention-deficit/hyperactivity disorder (ADHD) diagnosis. Later EEG segments and specific channels yielded the highest accuracy, suggesting cognitive fatigue influences ADHD classification.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Clinical Psychology
Background:
- Electroencephalogram (EEG) signals are crucial for diagnosing attention-deficit/hyperactivity disorder (ADHD).
- Noise and artifacts in EEG data can compromise diagnostic accuracy and reliability.
- Advanced preprocessing and segmentation techniques are vital for enhancing ADHD classification from EEG.
Purpose of the Study:
- To evaluate the impact of preprocessing and segmentation on ADHD diagnosis using EEG.
- To identify optimal EEG segments, channels, and features for accurate ADHD classification.
- To explore the role of cognitive fatigue in ADHD diagnosis via EEG analysis.
Main Methods:
- Applied artifact subspace reconstruction (ASR) and independent component analysis (ICA) for EEG preprocessing.
- Segmented EEG recordings and extracted statistically significant features.
- Utilized Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and XGBoost models for classification across different EEG segments and channels.
Main Results:
- Machine learning models trained on later EEG segments demonstrated significantly higher accuracy in ADHD classification.
- The highest accuracy of 86.1% was achieved using data from P3, P4, and C3 channels.
- Key features contributing to classification included Kurtosis, Katz fractal dimension, and power spectrums in Delta, Theta, and Alpha bands.
Conclusions:
- EEG preprocessing and segmentation are critical for reliable ADHD diagnosis.
- Cognitive fatigue may play a significant role in distinguishing ADHD, particularly in later EEG segments.
- Further research into cognitive fatigue and advanced segmentation strategies can improve ADHD diagnostic accuracy.
