Non-invasive Prediction of CYP11B2-Defined Subtypes in Primary Aldosteronism Using ¹⁸F-Pentixafor PET/CT and Machine...

Area of Science:

• Nuclear medicine imaging
• Machine learning in healthcare
• Endocrinology and metabolic diseases

Background:

• Primary aldosteronism (PA) is a common cause of secondary hypertension.
• Accurate subtyping of PA is crucial for guiding treatment decisions.
• Current diagnostic methods can be invasive and may not always be definitive.

Purpose of the Study:

• To develop and validate an interpretable machine learning model for non-invasive prediction of PA pathological subtypes.
• To integrate clinical data, radiomics, and deep learning (DL) features from 18F-AlF-NOTA-Pentixafor PET/CT.
• To enhance diagnostic accuracy and facilitate personalized management of PA.

Main Methods:

• Retrospective analysis of 89 PA patients undergoing 18F-Pentixafor PET/CT.
• Development of predictive models using support vector machine (SVM) algorithm.
• Integration of clinical, radiomics, and DL features with two-stage feature selection.
• Model performance evaluated using AUC, calibration, and decision curve analysis; interpretability via SHAP.

Main Results:

• The combined model achieved superior diagnostic accuracy (AUC 0.907, sensitivity 1.000, F1-score 0.923) in the test set.
• The integrated model significantly outperformed individual clinical, radiomics, and DL models (p<0.01).
• SHAP analysis highlighted key radiomics and DL features, aligning with biological markers like CXCR4 and CYP11B2 expression.

Conclusions:

• An interpretable machine learning model can accurately predict surgically confirmed PA subtypes non-invasively.
• This AI-driven approach may decrease reliance on invasive adrenal vein sampling.
• The model supports personalized surgical decision-making for PA patients.