AI-driven ablative margin assessment after HCC radiofrequency ablation: a case-control feasibility study.
Authors
Affiliations (3)
Affiliations (3)
- Interactive Graphics and Simulation Group (IGS), Department of Computer Science, University of Innsbruck, Innsbruck, Austria.
- Interventional Oncology/Stereotaxy and Robotics (SIP), Department of Radiology, Medical University Innsbruck, Innsbruck, Austria.
- Interventional Oncology/Stereotaxy and Robotics (SIP), Department of Radiology, Medical University Innsbruck, Innsbruck, Austria. [email protected].
Abstract
Thermal ablation of hepatocellular carcinoma (HCC) requires an adequate safety margin around the tumor to ensure treatment success. In this retrospective study, we evaluated an artificial intelligence (AI) pipeline for automatic safety margin assessment on computed tomography (CT) scans, with the goal of supporting radiologists in taking immediate clinical actions. We included CT scans from patients with primary liver cancer treated at the local university hospital with radiofrequency ablation (RFA) between 2010 and 2022. The study group consisted of 43 randomly picked lesions with local tumor progression, with an average diameter of 35 mm [24-45] (median [interquartile range]). For comparison, we randomly selected 96 lesions showing no local tumor progression via propensity matching (diameter 28 mm [17-37]). To calculate the minimum ablative margin (MAM), we coregistered the intraprocedural patient scans with a recently proposed AI-based image registration model. Overall, 139 HCC cases were evaluated, with an average follow-up interval of 25 months [8-42] (median, interquartile range). The calculated MAM was found to be a significant predictor of local tumor progression (odds ratio 0.42, 95% confidence interval 0.35-0.50, p = 0.001). Significant differences in local tumor protection rate were found between the lesions grouped by MAM (p < 0.001). Our AI-driven pipeline enabled fast and reliable assessment of the efficacy of RFA of HCC in relation to ablative margins. The ablative margins have been automatically quantified using a novel method that combines AI-based segmentation with AI-driven image registration. By delivering robust margin quantification quickly, the proposed pipeline can help clinicians in assessing ablative margins intraoperatively. An AI-driven tool can rapidly and automatically measure the treatment margin after liver cancer ablation. Inadequate safety margins calculated with the proposed pipeline are associated with a higher risk of local tumor recurrence. A MAM of 2.50 mm suffices to lower the risk of local tumor progression to below 10%.