Artificial intelligence-driven electroencephalogram analysis for early attention deficit hyperactivity disorder
Manjusha Deshmukh1, Mahi Khemchandani2
1Computer Engineering Department, Saraswati College of Engineering, Navi Mumbai, Maharashtra, India.
Insights
This study enhances attention deficit hyperactivity disorder (ADHD) detection using electroencephalogram (EEG) analysis. Short-time Fourier transform (STFT) with LightGBM models achieved 96% accuracy, identifying a 5-channel system for cost-effective diagnosis.
Area of Science:
- Neuroscience and Biomedical Engineering
- Computational Psychiatry
- Medical Signal Processing
Background:
- Attention Deficit Hyperactivity Disorder (ADHD) shares symptoms with other mental health conditions, complicating diagnosis.
- Early diagnosis of ADHD is crucial for timely intervention and improved patient outcomes.
- Current diagnostic methods can be subjective and time-consuming.
Purpose of the Study:
- To identify specific brain regions associated with ADHD in children.
- To develop a reliable framework for early ADHD diagnosis using advanced electroencephalogram (EEG) analysis.
- To enable prompt intervention through accurate and efficient diagnostic tools.
Main Methods:
- EEG signal decomposition using Discrete Cosine Transform (DCT), Short-Time Fourier Transform (STFT), and Empirical Mode Decomposition (EMD).
- Machine learning (ML) algorithms, specifically Light Gradient-Boosting Machine (LightGBM), were employed for ADHD detection.
- Comparative analysis of decomposition techniques and electrode configurations for optimal diagnostic accuracy.
Main Results:
- STFT demonstrated superior performance compared to DCT and EMD in ADHD detection.
- A 96% accuracy rate was achieved using STFT with LightGBM on 19 electrode sites.
- A reduced 5-electrode configuration (Fp1F3C3C4P4) using STFT and LightGBM yielded 91% accuracy.
Conclusions:
- STFT combined with LightGBM represents a state-of-the-art approach for ADHD detection via EEG.
- A minimal 5-channel electrode setup can maintain high diagnostic accuracy, paving the way for scalable solutions.
- This research provides a foundation for developing cost-effective and efficient ADHD diagnostic systems.
Objective:
Mental health (MH) and attention deficit hyperactivity disorder (ADHD) are inextricably linked, having the same symptoms and complications. The goal of this research is to pinpoint the precise brain areas that cause ADHD in children and to make it possible to diagnose the disorder early. The study intends to provide a trustworthy diagnosis framework that enables prompt intervention using cutting-edge electroencephalogram (EEG) data analysis and machine learning (ML) approaches.
Method:
This study uses EEG decomposition for improved ADHD detection. Decomposition techniques, such as the discrete cosine transform (DCT), short-time Fourier transform (STFT), and empirical mode decomposition (EMD), are used to break down EEG signals into sub-bases. As STFT demonstrated the highest accuracy, in further studies ML algorithms use STFT sub-bands on various combinations of brain regions as feed-ins to detect ADHD.
Result:
The results demonstrate that STFT methods outperform DCT and EMD. The trial outcomes revealed that, when utilizing a combination of 19 electrode sites, the STFT approach achieved the best accuracies, specifically 96% with light gradient-boosting machine (LightGBM) models. However, when utilizing STFT with LightGBM, the combination of Fp1F3C3C4P4 (5 electrode placements) yields 91% accuracy and 93% on Fp1F3C3C4P4 as well as Fp1F3C3C4F8.
Novelty:
While our previous research has separately investigated the efficacy of EMD and STFT/DCT, this presents the first comprehensive, head-to-head comparison of all three techniques within a unified framework. We conclusively demonstrate that STFT-based features, when paired with a LightGBM classifier, achieve a new state-of-the-art accuracy of 96%. Building on this superior model, we conduct a novel and granular electrode-reduction analysis to identify a minimal 5-channel configuration that maintains over 91% accuracy, directly addressing the need for scalable and cost-effective diagnostic systems and establishing a clear pathway for their development.
