3D bone reconstruction of the pelvis and hip through segmentation of tin-filtered ultra-low-dose CT compared with standard CT – a feasibility study

Purpose: 3D bone surface model reconstructions of CT exams are used for surgical planning. However, the accuracy of these models in tin-filtered ultra-low-dose CT examinations is unknown. Thus, we evaluated the accuracy of 3D-reconstructed bone models obtained from tin-filtered ultra-low-dose CT (tfCT) compared to standard CT (sCT) and calculated the radiation dose reduction. Materials and Methods: IRB-approved prospective study with informed consent. Patients undergoing a pelvic sCT examination were scanned additionally with tfCT. 3D surface model reconstructions (left and right hemipelvis, and left and right proximal femur) of both scans were created, for each patient independently by two readers. The average error between the surface models of both methods was obtained using root mean square error (RMSE). The dose parameters that were used were volume CT dose index (CTDIvol) and dose length product (DLP). Radiation dose differences of both scans were evaluated using the Wilcoxon signed-rank test. Results: 23 subjects (9 males and 14 females; mean age 31±8 years) were included. The mean RMSE was 0.47-0.48 mm for the right hemipelvis (left hemipelvis: 0.48-0.5 mm) and 0.52-0.56 for the proximal right femur (left femur: 0.5-0.57 mm). The average CTDIvol was 8.19±4.45 mGy for sCT examinations and 1.09±0.02 mGy for tfCT exams. The mean DLP was 220.45±124.36 mGy*cm for sCT examinations, and 29.12±1.54 mGy*cm for tfCT exams. The average CTDIvol and DLP were substantially lower (-87%) for tfCT compared to sCT (p