From Diagnostic Imaging to Systemic Prevention: An Integrated CEM–AI Framework for Oncologic, Osteoporotic, and Vascular Risk Stratification

Introduction: Contrast‑Enhanced Mammography (CEM) has evolved from a purely diagnostic tool into a hybrid modality capable of integrating structural and functional information. Our previous retrospective studies demonstrated three key findings: (i) CEM provides highly accurate tumor size estimation (MAE 0.46 mm) and detects additional lesions with surgical impact; (ii) breast density (BD) and background parenchymal enhancement (BPE) represent two independent physiological biomarkers—BD reflecting structural composition and correlating with bone mineral density (BMD), while BPE reflects vascular–hormonal activity and correlates with systolic blood pressure (SBP); (iii) artificial intelligence (AI) models, including multilayer perceptrons, deep neural networks, and bifurcated architectures, capture nonlinear relationships and can simultaneously predict BD, BPE, and systemic surrogates. These findings suggest that CEM may serve as a platform for multidimensional risk stratification.

Methods: We integrated results from five retrospective studies (n = 205–314 patients) evaluating CEM accuracy, BD–BPE interactions, and AI‑based prediction models. Imaging features (BD, BPE), clinical variables (age), and systemic indicators (BMD, SBP) were analyzed using linear regression, scikit‑learn models, and multi‑output deep learning architectures. Performance metrics included MAE, MSE, R², AUC, and inter‑reader agreement.

Results: CEM showed superior dimensional accuracy compared with mammography and ultrasound (p < 0.00001). BD and BPE demonstrated weak linear correlation (R² ≈ 0.14) but strong independent associations with systemic markers (BD–BMD r = 0.71; BPE–SBP r = 0.30). AI models improved reproducibility (K from 0.54 to 0.82), reduced reading time by 35%, and enabled simultaneous prediction of BD, BPE, BMD, and SBP. Multi‑output networks revealed latent systemic information embedded in CEM images.

Conclusions: CEM, when combined with BD, BPE, and AI‑based modeling, may extend beyond oncologic diagnosis to support integrated risk stratification for cancer, osteoporosis, and vascular disease. These findings justify a prospective multicenter validation study to confirm the systemic predictive potential of CEM‑derived biomarkers.