A Geometric and Dosimetric Assessment of Autocontouring for the Delineation of Pelvic Nodal Radiation Therapy Targets
Abstract
Purpose
Auto-contouring (AC) of the pelvic nodal Clinical Target Volume (CTV) for prostate cancer radiation therapy can reduce contouring time and inter-physician variability. A vendor tool provides two AC models based on different contouring guidelines, which have not been evaluated at our institution. In this study, geometric and dosimetric analyses were used to compare the models’ auto-contoured pelvic lymph nodes against the manually contoured pelvic lymph nodes to establish guidelines for clinical introduction.
Methods
A retrospective study of 50 prostate cancer patients with pelvic lymph node CTVs delineated by one of three attending Radiation Oncologists (ground truth CTV) was performed. A deep-learning tool (AutoContour, RADformation) generated two auto-contours (AC-CTV) (Model 1, Model 2) for each patient. AC-CTVs were compared to the ground truth CTVs geometrically using the Dice similarity coefficient (DSC), mean surface distance (MSD), and volume difference. An auto-planner (RapidPlan, Varian) generated two new plans per patient, one for each AC-derived Planning Target Volume (AC-PTV). Dosimetric quality was assessed using the ground truth PTV dose coverage (V100%) when the original targets were overlaid on the new plans. Metrics were compared using the Wilcoxon signed-rank test.
Results
AC-CTVs based on Model 1 showed significantly better geometric agreement with ground truth CTVs than those based on Model 2 (median DSC: 0.765 vs. 0.669; median MSD: 3.79 mm vs. 6.30 mm; both: p<0.001). Plans based on Model 1 showed significantly better ground truth PTV dose coverage (median V100%: 86.05% vs. 66.00%, p<0.001) than those based on Model 2, however both plans based on AC-PTVs led to clinically unacceptable underdosing to the ground truth PTV.
Conclusion
Model 1 provides a geometrically superior starting point for pelvic nodal CTV delineation. However, reliance on unedited targets results in underdosing, confirming the need for mandatory physician review and editing prior to clinical use.