[Classification of mild cognitive impairment and normal cognition using an automated voice-based testing...

Area of Science:

• Artificial Intelligence in Medicine
• Speech Signal Processing
• Neuroscience

Context:

• Early detection of cognitive decline is crucial for timely intervention.
• Mild cognitive impairment (MCI) affects memory and thinking, often preceding dementia.
• Existing tools for MCI detection can be resource-intensive.

Purpose:

• To develop a machine learning model for classifying individuals with mild cognitive impairment (MCI) versus cognitively normal (CN) status.
• To utilize speech features derived from spoken answers to neuropsychological tasks for classification.
• To create an accessible and user-friendly tool for cognitive decline screening.

Summary:

• A Gaussian Naive Bayes model was trained using age, gender, cognitive test scores, and speech features from spoken responses.
• The model achieved an Area Under the Curve (AUC) of 0.866, with 75% accuracy, distinguishing MCI from CN individuals.
• Speech data from tasks like time orientation, sentence recall, and digit span memory were analyzed.

Impact:

• The developed model demonstrates potential for easy-to-use, accessible MCI detection.
• Integration into smartphone applications and telephone services could facilitate widespread cognitive health monitoring.
• This approach may enable earlier identification of individuals with MCI, improving patient outcomes.