Alzheimer's Dementia Detection Using Perplexity from Paired Large Language Models

2025-06-11

AI Summary

Early detection of language deterioration in Alzheimer’s disease (AD) remains challenging for automated systems. Method: We propose an enhanced paired perplexity method tailored for instruction-tuned large language models (LLMs), the first to adapt this paradigm to Mistral-7B-Instruct. Our approach integrates task-oriented prompt engineering and human–model response contrastive analysis to construct interpretable decision boundaries and uncover implicit AD-related linguistic patterns encoded in the LLM. Contribution/Results: The method enables both behavioral interpretation of LLMs and high-fidelity synthetic AD language data generation for training augmentation. On the ADReSSo-2022 benchmark, it achieves a 3.33% accuracy improvement over the prior state-of-the-art paired perplexity method and a 6.35% gain over the ADReSS 2020 champion system. It uniquely balances predictive performance, robustness, and interpretability—advancing both clinical NLP and trustworthy AI for neurodegenerative disorder screening.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Alzheimer's dementia (AD) is a neurodegenerative disorder with cognitive decline that commonly impacts language ability. This work extends the paired perplexity approach to detecting AD by using a recent large language model (LLM), the instruction-following version of Mistral-7B. We improve accuracy by an average of 3.33% over the best current paired perplexity method and by 6.35% over the top-ranked method from the ADReSS 2020 challenge benchmark. Our further analysis demonstrates that the proposed approach can effectively detect AD with a clear and interpretable decision boundary in contrast to other methods that suffer from opaque decision-making processes. Finally, by prompting the fine-tuned LLMs and comparing the model-generated responses to human responses, we illustrate that the LLMs have learned the special language patterns of AD speakers, which opens up possibilities for novel methods of model interpretation and data augmentation.

Problem

• Detecting Alzheimer's dementia using large language models
• Improving accuracy in AD detection with interpretable decision boundaries
• Identifying AD language patterns for model interpretation and data augmentation

Innovation

• Uses Mistral-7B LLM for dementia detection
• Improves accuracy by 3.33% over existing methods
• Detects AD with interpretable decision boundaries