Quantitative susceptibility mapping and MRS-based multimodal machine learning for early Parkinson's disease...

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Parkinson's disease (PD) presents diagnostic challenges due to its progressive nature and heterogeneity.
• Early detection of PD is crucial for timely intervention and management.
• Current diagnostic methods may lack sensitivity in the early stages.

Purpose of the Study:

• To develop and validate a multimodal machine learning model for early Parkinson's disease detection.
• To integrate neurochemical metabolites and Quantitative Susceptibility Mapping (QSM)-based radiomic features for enhanced diagnostic accuracy.
• To identify key biomarkers for early PD diagnosis and improve model interpretability.

Main Methods:

• Utilized multicenter Parkinson's disease (PD) cohorts.
• Developed and evaluated machine learning models including Random Forest, Support Vector Machine, eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine.
• Integrated neurochemical metabolite data with QSM-based radiomic features.
• Employed SHapley Additive exPlanations (SHAP) for feature importance analysis and model interpretability.

Main Results:

• The eXtreme Gradient Boosting (XGBoost) model demonstrated superior predictive performance.
• Achieved high Area Under the Curve (AUC) values of 0.984 (training) and 0.973 (test) for early PD detection.
• Identified key diagnostic biomarkers through SHAP analysis, enhancing model interpretability.

Conclusions:

• The developed multimodal machine learning model shows significant potential for early Parkinson's disease detection.
• Integration of neurochemical and radiomic data improves diagnostic accuracy.
• The findings offer insights into the neurobiological mechanisms underlying early PD and support clinical application.