Language Modeling Screens Parkinson's Disease with Self-reported Questionnaires.

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Parkinson's disease (PD) poses a significant public health challenge, particularly for aging populations.
• Current diagnostic methods for PD often rely on motor symptoms and invasive procedures, hindering early detection.
• Developing accessible, non-invasive early diagnostic tools for PD is crucial.

Purpose of the Study:

• To establish a transferable artificial intelligence (AI) model, named Quest2Dx, for the non-invasive diagnosis of Parkinson's disease using health questionnaires.
• To address challenges in AI model development, such as missing data and the need for questionnaire-specific modeling.
• To enhance the interpretability of AI models in disease diagnosis.

Main Methods:

• Developed Quest2Dx, a novel language modeling approach for analyzing health questionnaire data.
• Implemented a transferable AI model designed to work across different questionnaires and handle missing responses.
• Validated Quest2Dx on the PPMI and Fox Insight datasets.

Main Results:

• Quest2Dx achieved high diagnostic accuracy, with Area Under the Receiver Operating Characteristic Curve (AUROC) scores of 0.977 (PPMI) and 0.974 (Fox Insight).
• Demonstrated strong cross-questionnaire validation performance, achieving AUROCs of 0.920 (PPMI to Fox Insight) and 0.952 (Fox Insight to PPMI).
• Identified key predictive questions, offering insights into PD indicators.

Conclusions:

• Quest2Dx represents a significant advancement in low-cost, non-invasive Parkinson's disease screening.
• The model's transferability and interpretability offer a promising approach for PD detection in primary care settings.
• This AI-driven tool has the potential to improve early diagnosis and management of Parkinson's disease.