Software-based versus visual assessment of the minimal ablative margin in patients with liver tumours undergoing percutaneous thermal ablation (COVER-ALL): a randomised phase 2 trial.

Background: Tumour coverage with an optimal minimal ablative margin is crucial for improving local control of liver tumours following thermal ablation. The minimal ablative margin has traditionally been assessed through visual inspection of co-registered CT images. However, rates of local tumour control after thermal ablation are highly variable with visual assessment. We aimed to assess the use of a novel software-based method for minimal ablative margin assessment that incorporates biomechanical deformable image registration and artificial intelligence (AI)-based autosegmentation.

Methods: The COVER-ALL randomised, phase 2, superiority trial was conducted at The University of Texas MD Anderson Cancer Center (Houston, TX, USA). Patients aged 18 years or older with up to three histology-agnostic liver tumours measuring 1-5 cm and undergoing CT-guided thermal ablation were enrolled. Thermal ablation was performed with the aim of achieving a minimal ablative margin of 5 mm or greater. Patients were randomly assigned (1:1) to the experimental group (software-based assessment) or the control group (visual assessment) by use of dynamic minimisation to balance covariates. Randomisation was performed intraprocedurally after placement of the ablation applicator. Assessment of oncological outcomes and adverse events were masked to treatment allocation. All analyses were conducted on an intention-to-treat basis. The primary endpoint was the minimal ablative margin on post-ablation intraprocedural CT. A preplanned interim analysis for superiority was done at 50% patient enrolment. Adverse events were recorded with the Common Terminology Criteria for Adverse Events. This trial is registered with ClinicalTrials.gov (NCT04083378), and recruitment is complete.

Findings: Patients were enrolled and treated with thermal ablation between June 15, 2020, and Oct 5, 2023. 26 patients were randomly assigned to the control group (mean age 58·1 [SD 14·8] years; 18 [69%] male and eight [31%] female; 11 [42%] colorectal cancer liver metastasis; median tumour diameter 1·7 cm [IQR 1·3-2·3]) and 24 to the experimental group (mean age 60·5 [14·4] years; 16 [67%] male and eight [33%] female; ten [42%] colorectal cancer liver metastasis; median tumour diameter 1·8 cm [1·5-2·5]). The interim analysis showed a mean minimal ablative margin of 2·2 mm (SD 2·8) in the control group and 5·9 mm (2·7) in the experimental group (p<0·0001), prompting halting of enrolment in the control group. A further 50 patients were enrolled to a non-randomised experimental group (mean age 56·5 [SD 11·7] years; 27 [54%] male and 23 [46%] female; 30 [60%] colorectal cancer liver metastasis; median tumour diameter 1·5 cm [IQR 1·3-2·2]); among these patients, the mean minimal ablative margin was 7·2 mm (SD 2·8). Grade 1-3 adverse events were reported in five (5%) of 100 patients: three (12%) of 26 in the control group and two (3%) of 74 in the experimental groups. No grade 4-5 adverse events or treatment-related deaths were reported.

Interpretation: Software-based assessment during CT-guided thermal ablation of liver tumours is safe and significantly improves the minimal ablative margin compared to visual assessment. Adoption of software-based assessment as a standard component of thermal ablation should be considered to achieve the intended minimal ablative margin.

Funding: US National Institutes of Health and US National Cancer Institute.