Natural language processing-based classification of early Alzheimer's disease from connected speech

Area of Science:

• Neurology
• Computational Linguistics
• Biomarkers

Background:

• Automated analysis of connected speech via Natural Language Processing (NLP) presents a potential biomarker for Alzheimer's disease (AD).
• Identifying specific speech types most sensitive and specific for AD detection remains a challenge.

Purpose of the Study:

• To evaluate the efficacy of different connected speech types in distinguishing early Alzheimer's disease (AD) patients.
• To assess the performance of a language model in classifying cognitively unimpaired (CU) individuals based on amyloid status.

Main Methods:

• A language model was applied to automatically transcribed connected speech from 114 Flemish-speaking individuals.
• Five distinct types of connected speech were analyzed to differentiate between early AD patients and amyloid-negative CU individuals, and between amyloid-negative and amyloid-positive CU individuals.

Main Results:

• The language model achieved up to 81.9% sensitivity and 81.8% specificity in distinguishing AD patients from amyloid-negative CU subjects.
• Classification accuracy for differentiating amyloid-positive from amyloid-negative CU individuals was up to 82.7% sensitivity and 74.0% specificity.
• Autobiographical interviews demonstrated superior performance compared to scene descriptions.

Conclusions:

• Autobiographical interviews are valuable for the automated analysis of connected speech in AD detection.
• Clinical AD classification prior to speech analysis can enhance the accuracy of amyloid status classification in healthy individuals.