Semi-automatic segmentation of pelvic bone tumors: Usability testing

Abstract

In this study, we focus on studying inter-user variability of a semi-automatic image processing pipeline used for 3D model production and subsequent PSI design in pelvic tumor resection. Six retrospective cases of pelvic bone tumors were segmented. Three different users (trained engineer (TE), orthopedic student (OS), expert orthopedic surgeon (OE)) performed multimodal semi-automatic segmentation on registered CT and MRI sequences. Inter-user variability was evaluated on the tumor models using the symmetrical Hausdorff distances and the DICE similarity coefficient. The mean symmetrical Hausdorff distance was 1.1 mm between TE and OS, 3.8 mm between TE and OE, and 3.6 mm between OS and OE. The mean values for the DICE similarity coefficient were 0.91 between TE and OS, 0.79 between TE and OE, and 0.82 between OS and OE. In patient 5, the DICE coefficient between TE and OE, and OS and OE dropped to 0.18 and 0.24 respectively. This study suggests that the inter-user variability between two segmentations of the same tumor cannot be overlooked, especially in complex pelvic tumors: OE expertise seems mandatory for tumor segmentation validation. The collaboration between engineers and clinicians also seems crucial for developing this type of pipeline for patient-specific instruments design purposes.

Keywords

• Surgical cutting guides
• Digital chain
• Patient-specific instruments
• Reconstruction surgery
• 3D printing
• Pelvic bone tumor