gms | German Medical Science

Objectives: If 3D-printed models of acetabulum fractures exhibit the same shape as in the patient image data and how shape deviations can be measured.

Methods: Various kinds of acetabular fracture CT scans (n=33) were obtained from our PACS system, all of which were reconstructed with sub-millimeter slice thickness using a soft-tissue kernel. All cases were processed by creating an initial segmentation that was refined by a surface model reconstruction filter. The resulting 3D surface mesh was printed with an Ultimaker 2 printer using polylactid acid (PLA) material. Then, the printed models were CT-scanned using a lung kernel and reconstruction with sub-millimeter slice thickness. Using the same surface model filter, the print CTs were converted into surface meshes, which allow for robust and precise shape comparison. To this end, we registered corresponding patient and print mesh models with a two-step approach: In the first step, the meshes were coarsely matched based on hand-labelled anatomical landmarks on the surface. In the second step, the registration was refined using a robust iterative closest point (ICP) method provided by the open source Visualization Toolkit program library. As the registration of non-equal shapes is ambiguous, we repeated the process several times, slightly varying the landmark positions. This way, we were able to average out wrong registrations, in which the algorithm chose to match deformed surfaces that increased the computed shape deviation.

Results and conclusion: To compare the registered patient mesh and print mesh shapes, we calculated the distance from each surface point on the patient mesh to its nearest neighbor on the print mesh. Specifically, we recorded the maximum surface distance for each registration. As each of the variation runs results in slightly different maximum distances for each patient-print pair, we compute the median of the maximum shape deviation over all runs, thus filtering outliers. The mean of all maximum distances between patient model and printed model for all cases was 5.33mm (standard deviation ± 1.9mm). Figure 1 shows the average surface distance from patient to print mesh for each of the 33 cases. The mean surface distances over all cases were lower than 0.9 mm. We presented a method for precisely quantifying the error of 3D printed acetabulum fracture models compared to the original patient data as shown in Figure 1. This method aims to fill a current gap in the routine use of 3D printed models by allowing the surgeon to precisely identify whether the printed model of a patient's fracture is valid and whether it accurately represents the patient's anatomy. Furthermore, the location of any detected errors can be visualized on the model, providing additional information to the surgeon. This has practical implications in improving the safety of 3D printed models that are increasingly being used for the routine treatment of complex orthopaedic trauma cases.