Predicting Parkinson's disease trajectory using clinical and functional MRI features: A reproduction and replication...

Area of Science:

• Neuroimaging
• Biomarkers
• Neurodegenerative Diseases

Background:

• Parkinson's disease (PD) lacks early diagnostic and progression biomarkers.
• Neuroimaging biomarkers show promise but are sensitive to data processing variations.
• Evaluating biomarker robustness is crucial for clinical application.

Purpose of the Study:

• To reproduce and replicate neuroimaging models for predicting PD state and progression.
• To assess the impact of methodological variations on biomarker performance.
• To provide recommendations for enhancing reproducibility in neuroimaging research.

Main Methods:

• Utilized the Parkinson's Progression Markers Initiative (PPMI) dataset.
• Reproduced and replicated machine learning models using fALFF and ReHo features from resting-state fMRI.
• Investigated variations in cohort selection, feature extraction, and input features.

Main Results:

• Achieved statistically significant predictive performance (R2 > 0) using a pipeline closely matching the original study.
• Partial reproduction using provided data yielded results comparable to the original findings.
• Identified challenges in reproducibility due to the complexity of neuroimaging studies.

Conclusions:

• The study successfully reproduced and replicated key findings, confirming the potential of neuroimaging biomarkers for PD.
• Methodological variations can significantly impact results, underscoring the need for standardization.
• Recommendations are provided to improve the reproducibility of future neuroimaging research in PD.