Behavioral Analysis for Alzheimer's Disease

As the world's older adult population increases, the number of people living with Alzheimer's disease (AD) and related dementia is growing rapidly, making them an increasingly serious health and social concern. Early diagnosis of dementia is important to ensure appropriate management and treatment of the disease, but it remains challenging because the biomarkers and extensive neuropsychological tests for the diagnosis can be invasive, time-consuming, and expensive.

From this perspective, we are developing behavioral analytics to help early detection of AD and related dementia. Specifically, we are focusing on sensor data derived from daily behaviors such as gait, speech, and drawing/handwriting, which can be easily collected both inside and outside of a clinic (e.g., at home). Our automated analysis pipeline extracts behavioral markers for AD and related dementia, for example: increased left-right asymmetry and step-to-step fluctuations in gait; language and speech disturbances in daily conversation; and increased pen pressure variability and reduced smoothness in drawings.

These behavioral features are then used by machine learning models for various types of health-related applications. These applications include (i) detection of AD at an early stage (i.e., mild cognitive impairment, MCI); (ii) differentiation of dementia subtypes; (iii) detection of risk factors for AD (i.e., mental health issues like loneliness and depression); and (iv) prediction of adverse effects of cognitive impairments on activities of daily life (e.g., risk of car accidents when driving).