Differentiating between common PSP phenotypes using structural MRI: a machine learning study.

Area of Science:

• Neuroimaging
• Neurology
• Machine Learning

Background:

• Differentiating Progressive Supranuclear Palsy-Richardson's syndrome (PSP-RS) from PSP-Parkinsonism (PSP-P) is clinically challenging.
• Distinguishing between these phenotypes is crucial for prognosis and clinical trial eligibility.

Purpose of the Study:

• To distinguish between PSP-RS and PSP-P using MRI structural data.
• To evaluate the diagnostic performance of MR Parkinsonism Index (MRPI) and machine learning models.

Main Methods:

• 62 PSP-RS, 40 PSP-P patients, and 33 controls underwent 3T MRI.
• Cortical thickness and volumes were extracted; MRPI and MRPI 2.0 were calculated.
• Machine learning algorithms (XGBoost, Random Forest) were applied to structural MRI data.

Main Results:

• MRPI and MRPI 2.0 showed AUCs of 0.88 and 0.81, respectively.
• Machine learning models combining MRPI with volumetric/thickness data achieved higher accuracy (AUC 0.93).
• The best model used XGBoost with MRPI, cortical thickness, and subcortical volumes, performing well even in early-stage PSP.

Conclusions:

• Combined MRPI and volumetric/thickness data improve differential diagnosis between PSP-RS and PSP-P.
• Structural MRI analysis aids in early diagnosis of common PSP phenotypes.
• Accurate differential diagnosis is vital for patient management and clinical trial enrollment.