Dementia Care Research and Psychosocial Factors

Area of Science:

• Neurolinguistics
• Computational Linguistics
• Artificial Intelligence in Medicine

Background:

• Advancements in AI and NLP offer valuable tools for clinical diagnostics.
• Previous research utilized AI-NLP to identify linguistic markers differentiating Primary Progressive Aphasia (PPA) subtypes from picture descriptions.
• This study aimed to replicate and extend these findings using a more ecologically valid language sample.

Purpose of the Study:

• To investigate the efficacy of AI-driven NLP in discriminating PPA subtypes using speech samples from the "What did you do for work?" prompt.
• To identify specific linguistic features (sentence length, word frequency, content word usage) that differentiate PPA subtypes.
• To validate the use of NLP tools in classifying atypical cognitive-behavioral syndromes.

Main Methods:

• Collected speech samples from 58 PPA participants and 16 healthy controls.
• Extracted digital language markers using Quantitext, an NLP-based toolbox.
• Quantified sentence length, word frequency, content word usage, and syntax frequency for each sample.
• Utilized one-way ANOVA to detect differences in language markers among groups.

Main Results:

• Individuals with non-fluent variant PPA (nfvPPA) produced significantly shorter sentences and used lower-frequency words compared to controls and other PPA subtypes.
• Semantic variant PPA (svPPA) participants used lower-frequency words than controls.
• Logopenic variant PPA (lvPPA) participants produced fewer content words than nfvPPA and control individuals.
• No significant differences in syntax frequency were observed across groups.

Conclusions:

• NLP tools demonstrate potential in discriminating PPA subtypes from unstructured language samples, aligning with previous research.
• AI-driven classification of atypical cognitive-behavioral syndromes shows promise.
• Further analyses are underway to evaluate the impact of symptom severity and validate findings against structured language samples.