Identification of language difficulties using explainable AI and multimodal EEG-fMRI brain network biomarkers

What happened
A team of researchers developed an artificial intelligence model that combines EEG and fMRI brain data to identify language difficulties in people aged 5 to 21. The study analyzed brain activity from 173 participants, including those with language challenges and typically developing controls. The model was trained and tested using thousands of machine learning configurations and evaluated with cross-validation methods. The researchers used explainable AI techniques to determine which brain features were most important for identifying language impairments.
Key details
- The model achieved up to 88.1% balanced accuracy in age subgroups (5–9, 10–13, and 14–17 years).
- EEG data was collected using a 129-channel Geodesic HydroCel System at 500 Hz.
- fMRI data was acquired using both 1.5T and 3T Siemens scanners with multiband EPI sequences.
- A total of 4,221 machine learning models were evaluated.
- SHAP, an explainable AI method, was used to interpret the model's neural feature importance.
Why it matters
Clinicians and researchers working with children and adolescents who have language difficulties now have access to an AI tool that can objectively identify these challenges using brain data. This approach provides a measurable way to understand the neural basis of language impairments, which may help improve diagnosis and support strategies. The use of explainable AI also allows professionals to see which specific brain features are linked to language difficulties, offering clearer guidance for future research and intervention.
