Detection of Neurocognitive Dysfunction via Hyperventilation Stimuli-induced EEG Responses using Machine Learning Approaches

Hyperventilation (HV) alters cerebral blood flow and neural excitability, making it a crucial physiological stimulus for detecting neurocognitive dysfunctions like mild cognitive impairment (MCI) and Alzheimer’s disease (AD). This study leverages HV-induced EEG responses to classify MCI and AD using support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost) machine learning (ML) models. Power-based features, including relative powers at delta, theta, alpha, and beta frequency bands, power asymmetry, organization score, and cognition index, were extracted and analyzed. Pearson correlation and paired t-tests were applied to select significant features. Power spectrum analysis and radar plots visualized frequency-dependent neural alterations across pre-HV, during-HV, and post-HV phases, showing increased relative powers at delta, theta bands, power asymmetry, organization score and reduced relative powers at alpha, beta bands, cognition index, spectral entropy in AD, reflecting cognitive decline. The XGBoost model achieved the highest classification accuracy, outperforming RF and SVM, demonstrating superior feature learning and interpretability. The findings highlight HV-induced EEG as a promising tool for early neurocognitive dysfunction detection, providing deeper insights into MCI and AD progression.