Validating a data-driven EEG-based dementia screening test

With the number of dementia patients expected to double in Europe and triple globally by 2050, particularly affecting low- and middle-income nations, there is an urgent need for reliable, cost-effective screening methods to enable early interventions. This study investigates an EEG-based dementia screening tool, comparing it to conventional cognitive screening tests for early dementia detection. The in-house developed software automatically detects and removes EEG artifacts, and employs generalized linear models to extract spectral EEG features linked to cognitive decline. Clustering and cross-validation are used to address multiple comparison problem and prevent overfitting. Finally, a neural network combines candidate spectral biomarkers into a single value, predictive of cognitive ability. We validated the system implementation on 448 senior participants who also underwent with five traditional cognitive tests (MoCA, ADAS, ACE III, Eurotest and Phototest) and eight minutes of EEG recording. Specificity and sensitivity of our test were 95% and 75% respectively with an overall accuracy of 94% in sample with 10% dementia prevalence, outperforming the other five conventional tests. This result highlights the potential of EEG as a low-cost, non-invasive, and efficient dementia screening tool, particularly suitable for primary healthcare settings or as a monitoring tool for disease progression and treatment efficacy.