Classification of mild cognitive impairment using machine learning with dynamic functional connectivity from resting-state functional MRI

The early diagnosis of mild cognitive impairment (MCI) is crucial for effective treatment. Resting-state functional magnetic resonance imaging (rs-fMRI) combined with machine learning has shown promise for the diagnosis of MCI. However, because rs-fMRI data tend to include substantial noise and the limited amount of available rs-fMRI data especially for MCI, it is important to develop a robust model to counter the effects of noise and data imbalance. Therefore, we propose a preprocessing method and classify preprocessed rs-fMRI data into cognitively normal and MCI groups using a machine learning model. Specifically, during preprocessing, we perform principal component analysis, window-based functional connectivity analysis, and feature selection based on hypothesis testing for differences. The highest classification performance from the fivefold cross-validation was an accuracy of 0.847, recall of 0.670, precision of 0.635, and F1 score of 0.633.