Harmonisation of PET imaging features with different amyloid ligands using machine learning-based classifier.

Area of Science:

• Neuroimaging
• Machine Learning
• Biomarker Development

Background:

• Amyloid-beta (Aβ) positron emission tomography (PET) imaging is crucial for Alzheimer's disease diagnosis.
• Different Aβ PET ligands (e.g., 18F-florbetaben and 18F-flutemetamol) exist, posing challenges for harmonizing data across studies.
• Standardized Aβ assessment is needed for reliable clinical trials and diagnosis.

Purpose of the Study:

• To develop a novel, ligand-independent machine learning classifier for harmonizing different Aβ PET ligands.
• To assess the accuracy and concordance of the classifier in detecting Aβ positivity.
• To evaluate the correlation of machine learning-based cortical tracer uptake (ML-CTU) values between different Aβ PET ligands.

Main Methods:

• A machine learning classifier was developed using paired 18F-florbetaben (FBB) and 18F-flutemetamol (FMM) PET images.
• The classifier was adapted from a previous FBB PET classifier to process FMM PET images, creating a ligand-independent Aβ PET classifier.
• Concordance rates for Aβ positivity and correlations of ML-CTU values between FBB and FMM were analyzed.

Main Results:

• The ligand-independent classifier demonstrated high accuracy (AUC = 0.958) in detecting Aβ positivity across different ligands.
• Good to excellent concordance (87.5%) was observed between FBB and FMM using the classifier, outperforming visual assessment.
• Highly correlated ML-CTU values (R = 0.903) were found between FBB and FMM.

Conclusions:

• The developed machine learning classifier effectively harmonizes different Aβ PET ligands in a clinical setting.
• This harmonization facilitates biomarker-guided diagnosis and supports clinical trials for anti-Aβ therapies.
• The novel classifier shows promise for improving the consistency and reliability of Aβ PET imaging in Alzheimer's research.