Iceball growth 3D simulation model based on finite element method for hepatic cryoablation planning

Abstract

Background: Cryoablation simulation based on Finite Element Method (FEM) can facilitate preoperative planning for liver tumors. However, it has limited application in clinical practice due to its time-consuming process and improvable accuracy. We aimed to propose a FEM-based simulation model for rapid and accurate prediction of the iceball size during the hepatic cryofreezing cycle.

Methods: A 3D simulation model was presented to predict the iceball size (frozen isotherm boundaries) in biological liver tissues undergoing cryofreezing based on the Pennes bioheat equation. The simulated results for three cryoprobe types were evaluated in the ex vivo porcine livers and clinical data. In ex vivo experiments, CT-based measurements of iceball size were fitted as growth curves and compared to the simulated results. Eight patient cases of CT-guided percutaneous hepatic cryoablation procedures were retrospectively collected for clinical validation. The Dice Score Coefficient (DSC) and Hausdorff distance (HD) were used to measure the similarity between simulation and ground truth segmentation.

Results: The measurements in the ex vivo experiments showed a close similarity between the simulated and experimental iceball growth curves for three cryoprobe models, with all mean absolute error<2.9 mm and coefficient of determination>0.85. In the clinical validation, the simulation model achieved high accuracy with a DSC of 0.87 ± 0.03 and an HD of 2.0 ± 0.4 mm. The average computational time was 23.2 s for all simulations.

Conclusion: Our simulation model achieves accurate iceball size predictions within a short time during hepatic cryoablation and potentially allows for the implementation of the preoperative cryoablation planning system.

Keywords: Cryoablation; Finite element method; Iceball simulation; Liver tumor; Preoperative planning.

Fig. 1

The workflow chart of this study. FEM = finite element method

Fig. 2

The geometric model in the FEM-based cryoablation simulation. (a) The iceball (blue ellipsoid) derived from the cryoprobe active region was modeled within the simulated domain (gray cylinder). A 3 mm diameter hepatic artery (red tube) was modeled 20 mm away and parallel to the active region. The centers of the cryoprobe active region and simulated domain were geometrically coincident in the FEM simulation. (b) The modeling structures of the cryoprobe: Active region (A), Diameter (D), and Length (L). (c) The geometric model in the 2D axial view

Fig. 3

The ex vivo experimental setup. (a) The Hygea AI Epic cryoablation system. (b) The cryoprobes (RCL17, RBL20, RBL26, from top to bottom) in our experiments. (c) The porcine livers were preprocessed to simulate the 37 °C in vivo temperature conditions before cryoablation procedures. (d) The CT imaging-based measurements of the ellipsoid iceball (dotted line encircled), including the length of long axis L (blue straight line) and short axis l (red straight line)

Fig. 4

The size measurements on the cut surface of iceball at 5, 10, 15, and 20 min of the RCL17 cryofreezing in ex-vivo experiments

Fig. 5

The growth curves on the simulated and experimental iceball size for RCL17, RBL20, and RBL26 cryoprobe during the 20-minute freezing duration. RCL17 curves: long axis with MAE = 2.32 mm and r² = 0.92, short axis with MAE = 0.77 mm and r² = 0.98. RBL20 curves: long axis with MAE = 2.40 mm and r² = 0.88, short axis with MAE = 0.93 mm and r² = 0.98. RBL26 curves: long axis with MAE = 2.07 mm and r² = 0.89, short axis with MAE = 2.89 mm and r² = 0.86

Fig. 6

The 2D (a) and 3D (b) views of the simulated iceball at 5 min, 10 min, 15 min, and 20 min during the freezing cycle. In the 2D views, the outer, middle and inner circles represents 0℃, -20 ℃ and − 40 ℃ isotherm surfaces, respectively

Fig. 7

The 3D views of the ground truth segmentation (red), the simulated iceball (blue), and the cryoprobe (gray)

Fig. 8

The examples of 2D similarity comparisons between the simulated iceball and ground truth in Case 1, Case 4, and Case 7 (using RCL17, RBL20, and RBL26, respectively). The simulated iceball (blue) was superimposed on the intraoperative 5-minute, 10-minute, and 15-minute axial CT images and compared with ground truth segmentation (red)