Screening of Mild Cognitive Impairment Through Conversations With Humanoid Robots: Exploratory Pilot Study.

Area of Science:

• Neurology
• Gerontology
• Artificial Intelligence

Background:

• Dementia is a growing global health concern, necessitating early detection methods.
• Current methods for detecting cognitive decline often rely on specific examinations, limiting their everyday applicability.
• Conversational humanoid robots offer a potential solution for continuous monitoring and care of older adults.

Purpose of the Study:

• To investigate the early detection of mild cognitive impairment (MCI) using natural conversations with humanoid robots.
• To assess the utility of prosodic and acoustic speech features for distinguishing between individuals with MCI and cognitively normal (CN) individuals without formal testing.

Main Methods:

• An exploratory study involving 94 participants (47 MCI, 47 CN).
• Collected conversational data from both a Mini-Mental State Examination (MMSE) and natural interactions with a humanoid robot.
• Extracted 17 prosodic and acoustic features, performed statistical significance tests, and conducted automatic classification using a support vector machine (SVM).

Main Results:

• Significant differences in speech features were found between MCI and CN groups in both MMSE conversations (5/17 features) and everyday robot conversations (16/17 features).
• Everyday conversations with the robot revealed significant differences in response time, speech duration, jitter, shimmer, and F0cov.
• Automatic classification achieved 66.0% accuracy for MMSE speech and 68.1% accuracy for robot conversations.

Conclusions:

• Prosodic and acoustic features from everyday conversations with humanoid robots can facilitate early and simple screening for MCI.
• This robot-based approach demonstrates comparable accuracy to the traditional MMSE for classifying individuals with MCI.
• Humanoid robots show promise for non-invasive, accessible cognitive health monitoring in older populations.