Mild cognitive impairment detection based on EEG and HRV data
Résumé
Brain volume decrease is usually connected to neurodegeneration and aging. In this environment,
an important percentage of elderly persons suffer from mild cognitive impairment (MCI), a kind of
dementia that can lead to Alzheimer’s disease (AD). Since the symptoms of cognitive impairment are
scarcely discernible, developing a safe and effective method for early MCI detection has emerged as
an important challenge. According to this regard, numerous cognitive training tests can be targeted
to help aging people retain a good quality of life, especially in the case of fragility disorders. A
Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) task was initially created to
detect the early stages of AD. This task specifically targets various tests related to specific cognitive
domains. However, it has since developed into a popular diagnostic tool for many kinds of dementia,
such as MCI. Several low-cost equipment, such as electroencephalography (EEG) and heart rate
variability (HRV), may be useful for predicting MCI. On the other side, various machine learning
(ML) models can be employed to extract/analyse relevant features from biomedical and physiological
signals, especially in the context of anomaly detection and classification. To this regard, we developed
a new method based on ML models to categorize MCI and healthy control (HC) patients during
the CERAD task using EEG and HRV multimodal data. Our dataset includes 15 subjects who were
randomly assigned to training and testing groups of 7 HC and 8 MCI, respectively. Our raw EEG
and HRV data are analyzed to extract time, frequency, and non-linear features. A scaling step is
employed to reduce the significant disparity between features. For the classification task, five ML
models are evaluated, including support vector machine (SVM), k-nearest neighbors (KNN), decision
tree (DT), random forest (RF), and gradient boosting (GB). To enhance accuracy, a hybrid ML model
with a voting system is developed, combining the top ML models with the highest accuracy rates.
A comparison step is performed between the use of ML and hybrid ML models. The experimental
findings demonstrated the efficacy of our proposed technique, which included a hybrid ML model. An
average accuracy of 93.86%, a sensitivity of 93.87%, and a specificity of 93.53% are achieved. The
obtained results allow one to conclude that the first CERAD test plays a prominent role as a novel
biomarker with an ultra-short duration for early MCI identification through the combination of EEG
and HRV signals.
Origine | Fichiers produits par l'(les) auteur(s) |
---|