Reliability of Automated Amyloid PET Quantification: Real-World Validation of Commercial Tools Against Centiloid...

Area of Science:

• Nuclear Medicine
• Neuroimaging
• Medical Software Validation

Background:

• Growing demand for amyloid PET quantification faces practical challenges.
• Automated software platforms are increasingly adopted to overcome these limitations.
• Evaluating the reliability of commercial tools against the original Centiloid Project method is crucial.

Purpose of the Study:

• To assess the reliability of three commercial automated software platforms (BTXBrain, MIMneuro, SCALE PET) for Centiloid quantification.
• To compare their performance against the original Centiloid Project method.
• To evaluate their concordance with visual interpretation of amyloid PET scans.

Main Methods:

• Retrospective analysis of 332 amyloid PET scans ([18F]Florbetaben and [18F]Flutemetamol) with paired T1-weighted MRI.
• Centiloid values calculated using BTXBrain, MIMneuro, and SCALE PET, compared to the original Centiloid method.
• Agreement assessed via Pearson's correlation, ICC, Passing-Bablok regression, Bland-Altman plots, and ROC curves for visual interpretation concordance.

Main Results:

• BTXBrain (R=0.993, ICC=0.986) and SCALE PET (R=0.992, ICC=0.991) showed excellent correlation; MIMneuro had slightly lower agreement (R=0.974, ICC=0.966).
• Software-specific biases were observed: BTXBrain underestimated (slope=0.872), MIMneuro overestimated (slope=1.053), and SCALE PET had minimal bias (slope=1.014).
• All platforms demonstrated excellent agreement with visual interpretation (AUC > 0.996).

Conclusions:

• Three automated platforms offer acceptable reliability for Centiloid quantification, despite software-specific biases.
• Observed biases did not impede feasibility in aiding image interpretation, as confirmed by concordance with visual readings.
• Users must recognize platform-specific characteristics for accurate diagnostic threshold application and longitudinal change interpretation.