3D Ablation Margin Detection of Colorectal Pulmonary Metastases Treated with Cryoablation Using Deformable CT Image Registration and AI-segmentation Predicts Local Response Outcomes

To evaluate the efficacy of various deformable image registration (DIR) processes to quantify three-dimensional (3D) minimal ablative margins (MAM) and correlate this MAM with local disease progression after CT-guided cryoablation of colorectal pulmonary metastases (CPM).  

Materials and Methods:

A retrospective, single-institution study was conducted on patients with CPM who underwent percutaneous cryoablation (PC) from March 2016 to April 2023. The first group of patients (n=38 patients, 76 tumors) had MAM quantified through manual segmentation of the tumor and ablation zone, performed by two board-certified interventional radiologists. In the second group where the entire lung was imaged, (n=25 patients, 31 tumors) MAM was calculated using a biomechanical model-based algorithm, following AI-driven auto-segmentation of the lungs, tumors, and ablation zones.

Results:

The study included 38 patients, with a total of 76 CPM treated across 63 sessions. Demographic details are presented in table 1. The median follow-up period was 2.75 years (range: 0.8–8.7 years). Primary efficacy at 1 year was 87%. Of the 76 ablated CPM lesions that underwent manual contouring, 56 (74%) had a MAM greater than 0 mm but less than or equal to 5 mm; 15 (20%) had a MAM greater than 5 mm; and 5 (6%) had a MAM of 0 mm. For the biomechanical DIR with internal boundaries, 14 (39%) lesions had a MAM of 0 mm, 20 (56%) had a MAM between 0 and 5 mm, and 2 (6%) had a MAM greater than 5 mm. In univariable Cox-proportional hazards regression, manually contoured 3D MAM (HR 0.04, 95% CI [0.01–0.17], p< 0.001) and freeze duration (HR 0.82, 95% CI [0.69–0.98], p=.025) and MAM calculated using biomechanical DIR with internal boundaries (HR 0.19, 95% CI [0.04–0.96], p=.045) were significantly associated with reduced risk of local tumor progression.

Conclusion:

3D quantification of MAM significantly predicts local tumor progression for predicting LTP for CPM undergoing PC.