Abstract
PURPOSE: To develop a mathematical framework to estimate the in silico A0 threshold based on the technical specifications of a specific ablation confirmation software package for thermal ablation of liver tumors that can then be used to identify the impact of different sources of error. METHODS: To estimate in silico A0 thresholds, we developed a simulation framework incorporating technical parameters and biological effects. Technical parameters were segmentation error, registration error, and slice thickness, and biological effects were tissue shrinkage and microscopic satellite lesions; these parameters and effects were all modeled using statistical distributions. For each permutation of parameters, a logistic regression was fitted to determine the observed MAM required to achieve ≥ 99% probability of true complete tumor coverage (i.e., the A0 threshold). The mathematical framework was integrated into a web application to estimate the A0 threshold and the reliability of the commonly used 5-mm A0 threshold based on several software performance characteristics. RESULTS: A total of 15,000,000 simulations (10,000 simulations × 1500 parameter permutations) were run and summarized. Tumor and ablation zone segmentation most greatly influenced the A0 threshold, with thresholds of 3.4 and 8.4 mm for 1- and 5-mm errors, whereas slice thickness had a relatively small effect, with A0 thresholds of 2.9 and 3.4 mm for thicknesses of 1 and 5 mm, respectively. CONCLUSION: This framework provides a method to determine software-specific in silico A0 thresholds and evaluate the reliability of existing 5-mm criteria based on software performance metrics. The results further show that ablation confirmation software should have registration and segmentation errors of ≤ 3 mm to reliably use a 5-mm A0 threshold.