A Digital Twin Framework for Modeling Multi-Organ Radiomics Stability in Total-Body PET Imaging

<h3>Rationale and Objectives</h3> Radiomics features derived from total-body PET imaging are sensitive to data acquisition variability, thereby limiting their reliability for quantitative modeling and digital twin applications. This study aimed to model multi-organ radiomics stability under dose perturbation and harmonization variability using a structured digital twin framework. <h3>Materials and Methods</h3> A multi-center total-body PET dataset comprising 550 patients was retrospectively analyzed. Seven dose levels were simulated from list-mode data (standard dose, and dose reduction factors (DRFs) of 2, 4, 10, 20, 50, and 100). A total of 107 International Biomarker Standardization Initiative (IBSI)-compliant radiomics features were extracted from standardized uptake value (SUV)-normalized images across 21 anatomical regions. Feature reproducibility was quantified using intraclass correlation coefficient — ICC (2,1). Stability trajectories across dose levels were transformed into multi-dimensional descriptors, including dose sensitivity (slope), area under the stability curve (AUSC), stability limit, and harmonization gain. These components were integrated into feature-level Radiomics Stability Scores (RSS), organ-level Twin Stability Index (TSI), and Digital Twin Readiness Scores (DTRS). <h3>Results</h3> Radiomics stability was organ-dependent and trajectory-driven. Skeletal structures, such as the femur, demonstrated high global stability (mean AUSC ≈ 45.1) with low dose sensitivity (mean absolute slope ≈ 6.1×10⁻⁴), maintaining >93% stable features at DRF=100 (ICC ≥ 0.75). In contrast, the liver showed substantially lower stability (mean AUSC ≈ 20.3) and earlier stability collapse, with stable feature fractions decreasing to ∼64% at DRF ≥10. Harmonization provided selective improvements but did not uniformly mitigate dose-induced instability. <h3>Conclusion</h3> Digital twin readiness in total-body PET cannot be inferred from single reproducibility metrics. Perturbation-aware stability modeling provides a structured foundation for identifying organ-specific suitability in quantitative theranostic and digital twin applications.