Investigating accuracy of 3D printed liver models with computed tomography

Abstract

Background: The aim of this study was to evaluate the accuracy of three-dimensional (3D) printed liver models developed by a cost-effective approach for establishing validity of using these models in a clinical setting.

Methods: Fifteen patients undergoing laparoscopic liver resection in a single surgical department were included. Patient-specific, 1-1 scale 3D printed liver models including the liver, tumor, and vasculature were created from contrast-enhanced computed tomography (CT) images using a cost-effective approach. The 3D models were subsequently CT scanned, 3D image post-processing was performed, and these 3D computer models (MCT) were compared to the original 3D models created from the original patient images (PCT). 3D computer models of each type were co-registered using a point set registration method. 3D volume measurements of the liver and lesions were calculated and compared for each set. In addition, Hausdorff distances were calculated and surface quality was compared by generated heatmaps.

Results: The median liver volume in MCT was 1,281.84 [interquartile range (IQR) =296.86] cm3, and 1,448.03 (IQR =413.23) cm3 in PCT. Analysis of differences between surfaces showed that the median value of mean Hausdorff distances for liver parenchyma was 1.92 mm. Bland-Altman plots revealed no significant bias in liver volume and diameters of hepatic veins and tumor location. Median errors of all measured vessel diameters were smaller than CT slice height. There was a slight trend towards undersizing anatomical structures, although those errors are most likely due to source imaging.

Conclusions: We have confirmed the accuracy of 3D printed liver models created by using the low-cost method. 3D models are useful tools for pre-operative planning and intra-operative guidance. Future research in this field should continue to move towards clinical trials for assessment of the impact of these models on pre-surgical planning decisions and perioperative outcomes.

Keywords: Three-dimensional (3D) printing; computed tomography (CT); liver resection; model; preoperative planning.

Conflict of interest statement

Conflicts of Interest: The authors have no conflicts of interest to declare.

Figures

Figure 1

Comparison of CT imaging quality between PCT and MCT. Liver position in both images is approximated by using point registration. CT, computed tomography; PCT, patient CT; MCT, model CT.

Figure 2

Computed tomography of 3D printed liver models. (A) Axial, sagittal and coronal views of liver model. Silicone parenchyma has higher attenuation than plastic elements (tumor, vessels); (B) inferior, left and anterior volume rendering views of 3D printed model.

Figure 3

Overview of segmentation process: based on patient CT images, region of interest masks are generated which allow to perform volumetric analysis as well as export them as 3D models. CT, computed tomography; VR, volume rendering (anterior view); SR, surface rendering.

Figure 4

Schematic representation of 2D landmark measurements performed on axial CT images. PVD, MHD, LHD and EDGE measurements are shown on example of PCT images, while TD and RHD on example of MCT images. PVD, portal vein diameter; MHD, middle hepatic vein diameter; LHD, left hepatic vein diameter; TD, distance from tumor to IVC; EDGE, distance from tumor to liver edge; RHD, right hepatic vein diameter; CT, computed tomography; PCT, patient CT; MCT, model CT.

Figure 5

Bland-Altman plot representing mean systemic difference of all measured parameters between patient CT and model CT. 95% confidence intervals and 1.96 standard deviation limits of agreement were used.

Figure 6

Hausdorff distance heatmaps presenting problematic areas. Warmer colors mean greater differences between surfaces. (A) Visual inspection confirms that most models were accurate with mean Hausdorff distance value lower than slice height; (B) still, a few models had issues with incomplete silicone filling, which is most prominent in case number 8, resulting in an increase of mean errors. (C) maximum Hausdorff distance values are additionally potentially overestimated in several models due to discrepancies in registration of inferior vena cava area.