Dementia Scale Classification with Sequential Model from Sleep Activity Data

Dementia, a disorder caused by brain diseases, has been found to influence the sleep patterns of patients. The finding indicates that monitoring sleep activity is helpful to detect the change in cognitive status. With this in mind, the aim of this study is to explore the possibility to develop a machine learning model for classifying the scores of dementia screening tests based on sleep activity data which could be recorded with less burden for participants. In this study, We collected sleep activity data from 124 elderly patients with varying cognitive states, including heart rate, respiratory rate and depth of sleep, using a single sensor. The score of Mini Mental State Estimation (MMSE) cognitive test is used to determine the level of cognitive states. First, we conducted a statistical analysis of the measured sleep activity data to find specific features observed in people with low-MMSE scores. Second, we utilized an efficient sequence model for capturing time-series changes in sleep activity for binary classification of the dementia scale to detect such low-MMSE people. Our findings revealed significant distinctions in sleep patterns between high and low cognitive status groups, and in the classification task, a maximum macro F1 score of 0.67 was achieved using LSTM models. Our results suggest the validity of using sleep activity data for the prediction of dementia classification.