Temporal Harmonization: Improved Detection of Mild Cognitive Impairment from Temporal Language Markers using...

Area of Science:

• Neurology
• Computational Linguistics
• Artificial Intelligence

Background:

• Mild Cognitive Impairment (MCI) represents an early dementia stage with detectable cognitive and behavioral changes.
• Early MCI detection is critical for timely interventions, clinical trial enrichment, and therapeutic development.
• Linguistic markers offer a non-invasive, cost-effective approach for MCI identification.

Purpose of the Study:

• To analyze linguistic markers from conversations to differentiate MCI from cognitively normal (NL) individuals.
• To investigate the potential of temporal linguistic dynamics across multiple conversations for enhanced MCI detection.
• To develop a method mitigating individual speaking style variations for improved model generalization.

Main Methods:

• Analysis of linguistic markers from multi-turn conversations between participants and healthcare professionals.
• Development of a temporal harmonization technique to standardize linguistic features across subjects.
• Application of machine learning models to predict MCI using harmonized temporal linguistic data.

Main Results:

• Temporal dynamics in conversations reveal fine-grained linguistic changes relevant to MCI.
• The proposed temporal harmonization method effectively reduces inter-subject variability in linguistic features.
• Machine learning models utilizing harmonized temporal features significantly improved MCI prediction accuracy.

Conclusions:

• Subject-invariant harmonized temporal linguistic features enhance the accuracy of MCI detection.
• This approach offers a promising avenue for non-invasive, early detection of cognitive decline.
• The methodology has implications for improving clinical trial selection and understanding MCI progression.