Toward Sensor-Based Early Diagnosis of Cognitive Impairment using Poisson Process Models

Abstract

Sensor-based assessment in combination with machine learning algorithms provide the potential to augment current practices of the (early) diagnosis of cognitive impairment. The goal of this paper is to detect cognitive impairment in elderly adults using sensor-based measures installed in the home. Longitudinal time-series data of sensor signals are analyzed with Poisson process (PP) models and supervised machine learning algorithms to identify individuals with mild cognitive impairment (MCI) and dementia. We examine two types of PP models: a homogeneous PP which assumes a constant rate of change for each sensor, and a non-homogeneous PP which incorporates contextual information by separately estimating the arrival rate for each task. Our results indicate that the proposed approach can effectively distinguish between patients with dementia and healthy individuals, as well as patients with MCI and healthy individuals based on the sensor-based PP features. Sensor-based assessment that relies on the non-homogeneous PP is further found to be more effective for the task of interest compared to homogeneous PP, as well as expert-based assessment. Findings from this research have the potential to help detect the early onset of cognitive impairment in elderly adults, and demonstrate the ability of computational models and machine learning to predict cognitive health, thus, contributing toward advancing aging-in-place. Clinical Relevance-This examines a computational method to quantify cognitive decline for elderly adults using home-based sensors. eventually contributing to ambulatory clinical biomarkers for dementia.