Multimodal Neurocognitive Analysis: Assessing Children's Learning Disabilities Using EEG and Psychometric Measurements

This open-source research project comes from the Monterrey Institute of Technology and Higher Education (Mexico). Its core is to integrate electroencephalogram (EEG) signals with standardized psychometric test data to build an attention-gated fusion model, aiming to provide a more precise classification and understanding framework for children with reading disabilities (RD) and mathematical disabilities (MD). The project code and dataset are fully open-source, serving as an excellent example for interdisciplinary research in neuroscience, machine learning, and clinical psychology.

Traditional diagnosis of children's learning disabilities relies on psychometric tests (e.g., reading speed, math performance, attention, and IQ assessments), which can indicate "what aspects" a child has difficulties with but cannot explain the neural-level "why". Phenomena show: Children with similar scores may have vastly different neural activity patterns; RD and MD children have overlapping cognitive indicators (e.g., IQ, attention) but different brain processing patterns. Core scientific questions: Do EEG and psychometric data carry complementary information? Can fusion provide a richer portrait of disabilities?

The study subjects are 76 children aged 7-13, divided into a reading disability group (RDG, ~38 people) and a mathematical disability group (MDG, ~38 people). Each child contributes baseline data:

• EEG: 3-minute resting state (eyes open) + task state (RDG reads text + answers comprehension questions; MDG performs arithmetic operations)
• Psychometrics: Reading speed/error rate/comprehension, math accuracy/reaction time, attention, IQ Note: Task-state EEG of the two groups is for inter-group comparison, not intra-individual comparison. The dataset comes from the Monterrey Institute of Technology and Higher Education and is publicly hosted on Mendeley Data.

The project builds the workflow through three Jupyter Notebooks:

1. Exploratory Data Analysis: Understand data characteristics, inter-group cognitive differences, and variable correlations.
2. EEG Preprocessing and Feature Extraction: Use MNE-Python to process EEG and output spectral-spatial feature matrices.
3. Classification and Fusion:
   • Baseline 1: EEG only (Random Forest/SVM, stratified k-fold cross-validation)
   • Baseline 2: Psychometrics only (same classifiers)
   • Innovative Model: Attention-gated fusion → Embed EEG and psychometric features separately, learn trust weights, fuse, then classify.

The project proposes three verifiable hypotheses: H1 (Neural Differentiation): There are differences in EEG band power between RD and MD groups (especially theta and beta waves), with more obvious differences in task states. H2 (Task Modulation): Task-state EEG differs from resting-state EEG within groups, reflecting increased cognitive demands. H3 (Multimodal Advantage): The accuracy of the attention-gated fusion model is higher than that of single-modal baselines.

The project emphasizes interpretability: It visualizes the neural basis behind model decisions (which brain regions and frequency bands drive classification) through projection topographic maps. This is crucial for clinical practice—doctors not only need to know the classification result but also understand the "why".

Research Significance:

1. Diagnostic Assistance: Provide objective neural markers for early identification of learning disabilities.
2. Personalized Intervention: Develop targeted plans based on neurocognitive characteristics.
3. Methodology: The attention-gated framework can be extended to other multimodal neuroimaging studies.
4. Open Science: The dataset and code are fully open-source, promoting reproducibility and expansion in the field.