Benchmarking motor imagery algorithms for pediatric users of brain-computer interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Brain-computer interfaces (BCIs) offer vital communication and control for children with neurological disabilities.
• Current BCI algorithm development often overlooks pediatric user needs and developmental differences.

Purpose of the Study:

• To evaluate the performance of 12 state-of-the-art motor imagery classification algorithms on pediatric datasets.
• To identify algorithms suitable for, and potential age-related effects in, pediatric BCI applications.

Main Methods:

• Tested 12 motor imagery classification algorithms on three datasets from 94 typically developing children (ages 5-17).
• Compared performance metrics, including Area Under the Curve (AUC), across different algorithms and age groups.

Main Results:

• Non-deep learning algorithms generally outperformed deep learning algorithms (mean AUC 0.64-0.65 vs. 0.57).
• Filter Bank Common Spatial Pattern (FBCSP) and ShallowConvNet showed significant age effects.
• Children as young as 6 demonstrated measurable motor imagery activations (AUC up to 0.8).

Conclusions:

• Pediatric users can produce reliable motor imagery signals for BCI use.
• Existing BCI algorithms require evaluation and potential adaptation for pediatric populations, especially considering developmental changes in EEG patterns.
• Further research is needed to optimize BCI algorithms for young users.