Toward Automated Cognitive Assessment in Parkinson's Disease Using Pretrained Language Models

The study published on arXiv on November 13, 2025, focuses on automated cognitive assessment in Parkinson's disease using advanced natural language processing (NLP) models. It addresses the challenge of extracting nuanced cognitive experiences from unstructured patient narratives, which can provide insights into the cognitive and emotional changes faced by individuals with Parkinson's. The research evaluated three model families, including Bio_ClinicalBERT, Meta-Llama-3-8B-Instruct, and GPT-4o mini, revealing significant variations in performance across different cognitive categories. Notably, the Meta-Llama-3-8B-Instruct model achieved the highest overall F1-scores, particularly excelling in context-dependent categories, indicating its potential utility in clinical settings. This advancement in AI-driven cognitive assessment could lead to improved patient care and a deeper understanding of Parkinson's disease.