Radiomics stability in total-body PET varies across organs when acquisition conditions change. A digital twin framework using 550 PET examinations, seven simulated dose levels and 107 features across 21 anatomical regions maps how stability shifts under dose reduction and harmonisation. The analysis, appeared in press in Academic Radiology, shows that skeletal structures retain robust feature behaviour more consistently than several soft-tissue organs. It combines dose sensitivity, cumulative stability, stability limits and harmonisation response to assess digital twin readiness.
Modelling Stability Beyond Single Metrics
The framework evaluates radiomics behaviour under controlled perturbations introduced through dose reduction and harmonisation. Dose levels range from full acquisition to progressively reduced count conditions, allowing reproducibility to be assessed across structured degradation scenarios. Intraclass correlation coefficients quantify agreement at each level, while additional descriptors capture how stability evolves rather than whether it persists.
Dose Sensitivity Coefficients describe the rate at which reproducibility declines with reduced counts, whereas the Area Under the Stability Curve summarises global stability across all perturbation levels. A stability limit defines the threshold at which reproducibility falls below acceptable agreement, identifying operational boundaries for feature reliability. These components are combined into composite measures that reflect both magnitude and resilience of stability.
Radiomics Stability Scores integrate normalised descriptors into feature-level indices, while organ-level aggregation produces a Twin Stability Index. A Digital Twin Readiness Score enables comparison across anatomical regions. This multi-dimensional structure transforms stability assessment into a state-space representation, capturing how features respond to realistic acquisition variability rather than relying on static reproducibility indicators.
Organ-Specific Patterns in Radiomics Robustness
Stability varies markedly across anatomical regions, with skeletal structures demonstrating consistently strong performance. The femur shows high cumulative stability and minimal sensitivity to dose reduction, maintaining a large proportion of reproducible features even at extreme reductions. Iliac bones and sacral bone display similar behaviour, reflecting robust stability trajectories.
Must Read: Whole-Body MRI Screening Enters Clinical Practice
In contrast, soft-tissue organs exhibit greater susceptibility to perturbation. The liver shows lower cumulative stability and earlier decline in reproducible features, with stability fractions decreasing substantially at moderate dose reductions. Kidneys and other soft-tissue regions follow similar patterns, indicating limited resilience under reduced count conditions.
Hierarchical differences emerge when stability is evaluated across thresholds. At conventional agreement levels, many organs maintain high proportions of stable features even at low doses. Stricter thresholds reveal earlier collapse in soft tissues, while skeletal regions preserve reproducibility. This separation highlights organ-specific stability boundaries and supports the identification of critical dose levels beyond which quantitative modelling becomes unreliable.
The distribution of stability and readiness scores shows that high intrinsic stability does not automatically translate into digital twin suitability. Some organs demonstrate moderate stability yet limited readiness due to constrained resilience or sensitivity to perturbation components.
Limited and Variable Impact of Harmonisation
Harmonisation strategies address non-biological variability arising from differences in acquisition parameters, yet their effects are uneven across organs. Radiopharmaceutical-based adjustments are most frequently associated with improvements, while manufacturer- and pixel-spacing–based approaches also contribute to selected regions.
Localised gains in reproducibility are observed in organs such as the colon, ribs, and bladder, where harmonisation increases agreement under specific conditions. However, average improvements across dose levels are generally small and often negative, indicating that harmonisation does not consistently enhance stability across the full perturbation range.
The findings show that harmonisation mitigates systematic variability but does not compensate for noise-driven degradation introduced by low-dose acquisition. Stable organs may still exhibit reduced performance after adjustment, and improvements in isolated conditions do not necessarily extend across all dose levels.
Visual representations of multi-dimensional stability profiles highlight distinct patterns across organs. Skeletal regions display broad, balanced profiles with strong performance across metrics, whereas organs such as the liver and heart show constrained profiles driven by higher sensitivity and limited harmonisation benefit. These differences underline the need for multi-parameter evaluation when assessing digital twin readiness.
Radiomics stability in total-body PET imaging varies across organs and depends on how features respond to acquisition perturbations. Trajectory-based modelling reveals that robustness cannot be captured by a single reproducibility measure but requires integrated evaluation of stability magnitude, decay behaviour and resilience boundaries. Skeletal structures maintain strong performance under reduced dose conditions, whereas several soft-tissue organs show earlier degradation despite moderate baseline agreement. Harmonisation provides selective improvement but does not fully address instability linked to count reduction. These findings support the use of perturbation-aware frameworks to identify suitable regions and features for digital twin applications in quantitative imaging.
Source: Academic Radiology
Image Credit: iStock
References:
Azimi M, Cheraghi M & Zaidi H (2026) A Digital Twin Framework for Modeling Multi-Organ Radiomics Stability in Total-Body PET Imaging. Academic Radiology: In Press.
