Prediction of Cognitive Impairment Using a Deep Learning Autoencoder Algorithm from a Singapore Study

Dementia is a decline in cognitive function, typically diagnosed when the acquired impairment becomes severe enough to impact social or occupational functioning. Between no cognitive impairment (NCI) and dementia, there are many intermediate states. Predictive cognitive impairment can be useful for initiating treatment to prevent further brain damage. Several deep learning-based approaches have been proposed for the classification of Magnetic Resonance Imaging (MRI) to diagnose Alzheimer’s disease (AD) or dementia. However, diagnoses of cognitive impairment are not based solely on MRI data but also on other associated factors. In this study, the aim is to predict cognitive impairment using both neuroimaging markers and other associated factors, employing a deep learning autoencoder algorithm based on data from a Singapore study. A novel method has been proposed for applying autoencoders to a multiclass classification task and provide a feature importance analysis. The performance of the autoencoder model was compared with two widely used machine learning classification algorithms, namely, the multinomial logistic regression (MLR) and Extreme Gradient Boosting (XGBoost). The results show that the autoencoder algorithm performs well and outperforms the competing methods.