Attention-deficit/hyperactivity disorder (ADHD) is among the most common neurodevelopmental disorders globally, influencing approximately 5-7% of children and 2-5% of adults. Characterized by persistent patterns of inattention, hyperactivity, and impulsivity, symptoms typically manifest during preschool years and intensify as the child enters a school environment. The effects of ADHD extend into adulthood, impacting academic, personal, and social functioning.
Current diagnostic strategies heavily rely on subjective behavioral assessments, which can lead to biases and inaccuracies, contributing to missed diagnoses, especially in contexts with limited specialized expertise. Electroencephalography (EEG) presents an appealing solution for developing objective ADHD biomarkers by measuring neural activity with high temporal resolution. Recent advances in machine learning utilizing EEG data have shown potential for automating ADHD classification, yet these methods usually focus solely on brain signals and ignore the broader socioeconomic context that significantly impacts neurodevelopment and risk for ADHD.
Research emphasizes that factors like socioeconomic status (SES), including parental education, income level, and neighborhood quality, influence brain development throughout an individual's life, establishing a prominent link between SES and neurodevelopment. These adversities can correlate with deficits in attention and executive function, which overlap with ADHD symptomatology.
Despite established relationships between SES and neural function, machine learning approaches have rarely merged socioeconomic information with neurophysiological data. Previous studies typically analyze either temporal or spectral features from EEG data without combining multiple data representations, often missing complementary insights.
In this research, we introduce a multimodal deep learning architecture that incorporates three key components: (1) temporal features from EEG through one-dimensional convolutional-recurrent networks, (2) spectro-temporal features via two-dimensional convolutional networks employing attention mechanisms, and (3) economic factors analyzed through feedforward processing. Our objective is to enhance ADHD risk screening accuracy by integrating socioeconomic information into EEG data processing.
Using the Cognitive Electrophysiology in Socioeconomic Context dataset, which encompasses EEG records from 127 young adults across four cognitive tasks along with detailed SES information, we aim to explore two critical avenues:
Tasks include auditory oddball (passive attention processing), visual oddball (active target detection), flanker task (response inhibition), and visual search efficiency — allowing us to comprehend if socioeconomic context consistently provides discriminatory power regardless of the cognitive demands.
Analyses were conducted using five-fold stratified cross-validation, ablation studies, and benchmarking against an established EEG classification model.
Findings indicate that the multimodal model outperforms EEG-only models across the four cognitive tasks, achieving increased accuracy of 2.1–5.9% and sensitivity improvements reaching 12.2%. Notably, the positive-class F1-scores reached high values between 96.4–99.8%, further illustrating the benefits of integrating socioeconomic context in ADHD screening.
Leave-One-Subject-Out Cross-Validation demonstrated an accuracy range of 0.86–0.91 for the EEG-only model compared to 0.92–0.96 for the multimodal approach, accompanied by benchmarks showcasing heightened sensitivity and specificity.
The model's EEG backbone was also validated using an independent pediatric dataset, achieving 80.4% subject-independent accuracy—showing continued performance beyond initial training scenarios. Nonetheless, the study notes that self-report screening instruments, rather than clinical diagnoses, primarily generated labels, clarifying that the proposed methodology is intended as a proof-of-concept for ADHD risk screening and not as a diagnostic tool.
This research underlines the feasibility of conducting context-aware ADHD risk screening, integrating environmental influences alongside neurophysiological signals. By accommodating socioeconomic context, tools for assessing ADHD risk can refine diagnostic accuracy and modify treatment strategies per individual needs. Further exploration is essential to track the implications of this model across varied populations and clinical settings.
The integration of socioeconomic context with multimodal EEG data presents a promising direction for improving ADHD risk screening methodologies. This work not only enhances the understanding of how various factors may intersect with neurodevelopment but also sets a foundation for future studies aiming to develop more precise ADHD diagnostic tools.