Software-specific A0 Thresholds for Minimum Ablative Margins after Liver Tumor Thermal Ablation: A Large-Scale Simulation Study

Take-Away Points

• ■ Major Focus: To quantify how technical measurement errors and biologic uncertainties shift the observed minimum ablative margin (MAM) required to meet an in silico A0 definition for liver tumor thermal ablation confirmation software.

• ■ Key Results: In 15 000 000 simulations (10 000 simulations × 1500 parameter permutations), segmentation error most greatly impacted the A0 threshold (3.4 mm at 1-mm error vs 8.4 mm at 5-mm error), followed by registration error (3.4 mm at 1 mm vs 7.0 mm at 5 mm), whereas section thickness had minimal effect (3.4 mm at 1 mm vs 2.9 mm at 5 mm).

• ■ Impact: A 5-mm A0 criterion is technically reliable only when both segmentation and registration errors are ≤ 3 mm; even at 3 mm, the threshold becomes borderline if both error sources coexist, underscoring the need for software-specific calibration and transparent reporting of performance metrics and exclusion criteria.

Ablation confirmation software is increasingly used to assess intraprocedural treatment adequacy during liver tumor thermal ablation. In this context, the A0 threshold refers to the observed MAM associated with a predefined very high probability of complete tumor coverage. However, MAM is not a direct histologic measurement but an imaging-derived surrogate obtained by coregistering pre- and postablation images; therefore, it depends on segmentation and registration accuracy. Because the true microscopic margin cannot be verified after in situ necrosis, commonly applied A0 thresholds such as MAM ≥ 5 mm may be method dependent rather than universally applicable.

In this simulation study, Paolucci et al developed a modeling framework to derive software-specific in silico A0 thresholds based on technical parameters (segmentation error, registration error, section thickness) and biological factors (tissue shrinkage, microscopic satellite lesions), each modeled probabilistically. Across 1500 parameter permutations, 10 000 simulations per permutation generated paired true and observed MAM values for tumors 10–30 mm in size and candidate margins of −5 to 10 mm. A logistic model identified the observed MAM corresponding to a ≥99% probability of true complete coverage (MAM_true > 0), which defined the A0 threshold. The framework was also implemented as a web application to estimate software-specific A0 values and the reliability of a 5-mm criterion.

Segmentation error had the greatest impact on A0, exceeding the effect of registration error. Increasing registration error raised A0 (3.4 mm at 1 mm, 4.9 mm at 3 mm, 7.0 mm at 5 mm). Excluding cases with registration error > 3 mm, a common clinical practice, produced lower A0 estimates than analyses retaining all registrations, with larger discrepancies at higher error levels. Section thickness had minimal influence, with only a marginal decrease in A0 (≤ 0.5 mm across 1–5 mm). Biological effects shifted A0 in opposite directions: tissue shrinkage reduced A0 (3.4 mm at 0% vs 1.8 mm at 30%), whereas microscopic satellite lesions substantially increased A0 as their probability rose (3.4 mm at 0% vs 7.9 mm at 75% and 7.7 mm at 100%).

These findings suggest that A0 thresholds should be interpreted relative to software-specific measurement performance rather than applied as uniform margin cutoffs. Prospective technical and clinical validation studies are warranted to assess how such thresholds generalize across platforms and to support standardized reporting of measurement accuracy and case exclusions of acquisition protocols, registration and segmentation performance, predefined quality thresholds, and nonevaluable or excluded cases.

Highlighted Article

• Paolucci I, Albuquerque J, Siddiqi NS, et al. The effects of measurement errors on minimum ablative margins after thermal ablation of liver tumors: a simulation study. CVIR Oncol 2026;2(1). Published online January 14, 2026. doi: https://doi.org/10.1007/s44343-025-00029-9.