Improved interpretation of ¹⁸F-florzolotau PET in progressive supranuclear palsy using a normalization-free...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Neurology

Background:

• 18F-florzolotau PET is an emerging biomarker for progressive supranuclear palsy (PSP).
• Interpretation challenges include lack of consensus in visual reading and biases in semi-quantitative analysis.
• Clinical overlap between PSP and Parkinsonian multiple system atrophy (MSA-P) necessitates improved diagnostic tools.

Purpose of the Study:

• To develop a reliable discriminative classifier for 18F-florzolotau PET to differentiate PSP and MSA-P.
• To create a normalization-free deep-learning (NFDL) model for enhanced PET analysis.

Main Methods:

• Development of a normalization-free deep-learning (NFDL) model for 18F-florzolotau PET analysis.
• Comparison of NFDL model accuracy against conventional semi-quantitative classifiers.
• Analysis of regions driving NFDL classifier decisions and correlation of NFDL-guided radiomic features with clinical severity.

Main Results:

• The NFDL model achieved significantly higher accuracy in differentiating PSP and MSA-P compared to semi-quantitative classifiers.
• Identified disease-specific topographies driving the NFDL classifier's decisions.
• NFDL-guided radiomic features showed correlation with the clinical severity of PSP.

Conclusions:

• The NFDL model demonstrates potential for early and accurate differentiation of atypical parkinsonism using 18F-florzolotau PET.
• The model's independence from subjective interpretation, MR-dependent, and reference-based preprocessing makes it broadly applicable.