Radiomics-enhanced <sup>18</sup>F-AV45 PET/MRI for integrative assessment and centiloid estimation of amyloid-β burden in Alzheimer's disease.
Authors
Affiliations (3)
Affiliations (3)
- Department of Nuclear Medicine, Shanghai East Hospital, School of Medicine, Tongji University, Shanghai, China.
- Department of Rehabilitation, Guangdong Work Injury Rehabilitation Hospital, Guangzhou, China. [email protected].
- Department of Nuclear Medicine, Shanghai East Hospital, School of Medicine, Tongji University, Shanghai, China. [email protected].
Abstract
Reliable assessment of cerebral amyloid-β (Aβ) deposition is essential for the diagnosis and management of Alzheimer's disease (AD). This study aimed to evaluate the feasibility of integrating radiomics-enhanced <sup>18</sup>F-florbetapir positron emission tomography/magnetic resonance imaging (<sup>18</sup>F-AV45 PET/MRI) features for Aβ status evaluation and to further explore their potential for continuous Centiloid prediction in AD. Ninety-four subjects who underwent (<sup>18</sup>F-AV45 PET/MRI (60 Aβ-positive, 34 Aβ-negative) were retrospectively included. Standardized uptake value ratio (SUVr) features were extracted from seven cortical regions (frontal, temporal, parietal, occipital, insular, cingulate, and white matter), and corresponding T1-weighted images' radiomics features were computed. Three feature sets (PET, radiomics, and combined) were analyzed using logistic regression (LR), k-nearest neighbor (kNN), and linear discriminant analysis (LDA) with 10-fold cross-validation. The best performing model was further interpreted using SHapley Additive exPlanations (SHAP) analysis. Additionally, Centiloid regression was performed using random forest, ElasticNet, and ExtraTrees regressors. The combined feature achieved the best performance with the LR model, with area under the receiver operating characteristic curve = 0.9373, accuracy = 0.8723, F1-score = 0.898). SHAP analysis identified biologically meaningful features derived from both radiomics and PET modalities, showing clear inter-group separation. In Centiloid regression, the ExtraTrees model achieved strong agreement with measured values. This framework provides an interpretable and quantitative solution for amyloid evaluation, enabling both categorical Aβ status discrimination and continuous Centiloid estimation from routine PET/MRI data. This approach represents a proof-of-concept for supporting <sup>18</sup>F-AV45 PET-based assessment in AD. This study demonstrates that radiomics-enhanced PET/MR features can reliably predict both Aβ status and Centiloid values without specialized processing platforms, offering a clinically deployable, standardized, and interpretable approach to improve AD diagnosis and monitoring. Radiomics-enhanced <sup>18</sup>F-AV45 PET/MRI enabled quantitative Aβ evaluation in AD. Radiomics-enhanced <sup>18</sup>F-AV45 PET/MRI provided a noninvasive and interpretable assessment to improve clinical confidence. Radiomics-enhanced <sup>18</sup>F-AV45 PET/MRI allowed Centiloid estimation without specialized platforms for wider clinical use.