Identifying neurocognitive disorder using vector representation of free conversation

Area of Science:

• Computational linguistics
• Neuroscience
• Artificial intelligence in healthcare

Background:

• Recent research explores natural language processing (NLP) for dementia identification.
• Existing methods often rely on picture description tasks, which may not be ideal for clinical settings.
• Free conversation offers a simpler, task-free approach less prone to learning effects.

Purpose of the Study:

• To develop a machine learning model for discriminating dementia patients from healthy individuals.
• To utilize features extracted from unstructured free conversation data via NLP.
• To assess the feasibility of using free conversation for dementia screening in clinical practice.

Main Methods:

• Recruited patients from a specialized dementia outpatient clinic and healthy volunteers.
• Transcribed participant conversations and performed morphological analysis using NLP.
• Converted text data into real-valued vectors for machine learning model training.
• Utilized 432 datasets for model development and evaluation.

Main Results:

• The machine learning model achieved a classification accuracy of 0.900.
• Sensitivity was 0.881, and specificity was 0.916 in discriminating dementia.
• High accuracy was demonstrated in classifying dementia versus non-dementia subjects.

Conclusions:

• Machine learning models can effectively discriminate between dementia and non-dementia subjects using free conversation data.
• NLP-based feature extraction from unstructured conversation is a viable method for dementia identification.
• This approach shows promise for practical clinical application in dementia screening.