Brain structural features with functional priori to classify Parkinson's disease and multiple system atrophy using...

Area of Science:

• Neuroimaging
• Machine Learning
• Neurology

Background:

• Clinical 2D MRI has limitations for early Parkinson's disease (PD) and multiple system atrophy (MSA) diagnosis.
• The diagnostic and therapeutic potential of MRI in these conditions is underexplored.

Purpose of the Study:

• To develop a machine learning framework for distinguishing PD from MSA using reconstructed clinical MRI data.
• To identify neuroimaging biomarkers specific to PD and MSA.

Main Methods:

• Employed a structure-constrained super-resolution network (SCSRN) to reconstruct 2D MRI data from 56 PD and 58 MSA patients.
• Utilized a functional template for feature extraction and hierarchical SHAP for feature selection.
• Trained Extra Trees and logistic regression models on the reconstructed data.

Main Results:

• Achieved 95.65% accuracy and 99% AUC in distinguishing PD from MSA on the test set.
• Identified significant neuroimaging biomarkers, including larger fourth ventricular and smaller brainstem volumes.
• Clarified the positive and negative impacts of various features in predicting PD and MSA.

Conclusions:

• The novel framework effectively utilizes reconstructed clinical 2D MRI for PD and MSA differentiation.
• Provides insights into disease-specific neuroimaging biomarkers and brain morphology alterations.
• Highlights the potential of advanced AI techniques for neurodegenerative disease diagnosis.