A Comprehensive Benchmark of EEG-Based BCI Deep Learning Models for MCI and Dementia Classification

A new study has evaluated ten deep learning models for classifying mild cognitive impairment (MCI) and dementia using electroencephalography (EEG) brain signals. The researchers aimed to provide a standardized evaluation framework, addressing inconsistencies found across previous studies.

To achieve this, the team used a unified preprocessing pipeline and five-fold subject-wise cross-validation on four resting-state EEG datasets. Each experiment was replicated ten times to ensure the reliability of the results.

The findings revealed that SCCNet achieved the highest mean subject-level accuracy, sensitivity, and F1 score. In contrast, ShallowConvNet had the highest mean segment-level accuracy, specificity, and precision. The study also found that subject-level aggregation improved the mean accuracy for all evaluated models.

However, the researchers noted significant variations in performance across datasets and diagnostic tasks. Interestingly, higher computational cost did not necessarily translate to better classification performance. Several compact architectures managed to compete with substantially larger models in terms of classification accuracy.

The study provides a reproducible reference for comparing EEG-based dementia classification models under consistent, subject-independent evaluation conditions. This benchmarking approach can help researchers choose the most suitable deep learning model for their specific EEG dataset and diagnostic task.