Recurrent Neural Network based Prediction of Transition to Mild Cognitive Impairment Using Unobtrusive Sensor Data

Abstract

Alzheimer’s disease (AD) is the most common neurodegenerative disease among older adults. Mild Cognitive Impairment (MCI) is a transitional stage where older adults begin to exhibit symptoms that could be precursors to AD progression. Detecting the MCI stage early can help prevent or delay the onset of an advanced stage in AD and thus enhance the quality of life among older adults. This neurodegeneration process resonates with the inherent temporal nature of this disease progression. Patterns/trends in daily activities/routine measured over time are better indicators of the disease progression and help detect the transitional stage, MCI. This paper aims to leverage the activity patterns derived through unobtrusive sensors at historical time points and investigate the effect of these activity trends in predicting the progression at a future time point. This study proposes a prediction model leveraging Long short-term memory recurrent neural networks (RNN). From the daily activity/routine standpoint, walk and sleep-related measures are used as input features to the model along with the diagnostic label derived from neuropsychological assessment data, and the transition to MCI is predicted at a future time point. The initial experiment and results show that the study approach proposed in this paper can predict the progression yielding an 82 percent overall prediction accuracy and 90 percent accuracy in predicting degenerating cases. These results encourage future experiments with other extended activity features and further fine-tuned RNN model.