The Dosimetric Impacts of CT-Based Autocontouring Algorithm for Breast Cancer Radiotherapy Planning

BACKGROUND In many institutions, the delineation of tumors and organs at risk (OARs) is still performed manually, which is both time-consuming and resource-intensive. Furthermore, inter-observer variability leads to inconsistencies in contouring accuracy. In recent years, several automatic contouring methods have been developed to overcome these limitations. However, these approaches have not yet provided sufficiently accurate results for clinical practice. One possible reason is that most auto-contouring algorithms have been developed and validated using CT images, which may not be optimal for achieving accurate automated delineation. PURPOSE To achieve effective tumor control in radiotherapy and to minimize radiation exposure to organs at risk (OARs), accurate delineation of target volumes and OARs is an essential component of radiotherapy (RT) planning. Additionally, precise contouring is crucial for the reliable assessment of RT-related toxicity. In recent years, artificial intelligence (AI)-based models have been developed, providing high accuracy in delineating various anatomical regions within a shorter time. The present study aimed to dosimetrically evaluate the usability of a new-generation auto-contouring algorithm (DirectORGANS), which automatically detects and contours organs directly on computed tomography (CT) images acquired at the simulator prior to breast radiotherapy planning. METHODS The CT images of 30 patients were used in this study. All patients who underwent breast-conserving surgery (BCS) subsequently received radiotherapy. The breast, defined as the target volume, was automatically contoured for all patients using the DirectORGANS algorithm at the CT simulator. The CT datasets were then imported into the Eclipse Treatment Planning System (TPS) for contour review and dose evaluation. On the same CT image sets, the breast volumes were manually delineated by an experienced physician according to the RTOG atlas, and these manual contours were used as the reference structures. For each patient, volumetric modulated arc therapy (VMAT) plans were generated based on the reference contours (RefPlan). A dose prescription of 40 Gy in 15 fractions was administered to the clinical target volume (CTV). The dose parameters for both the manually delineated and automatically generated contours of the target volume were obtained from the dose–volume histogram (DVH) of the same treatment plan. To evaluate target coverage, the conformity index (CI) and homogeneity index (HI) were calculated. Statistical comparisons between manual and automatic contouring results were performed using the Wilcoxon signed-rank test with SPSS software, and a p-value < 0.05 was considered statistically significant. RESULTS Statistically significant differences were observed between the manual contours (MC) and auto contours (AC) in terms of the homogeneity index (HI), conformity index (CI), and the clinical target volume (CTV) coverage by the 95% isodose line, which resulted from variations in breast contouring (p < 0.001). Furthermore, there was a statistically significant difference between the time required for manual and automatic contouring (p < 0.001). CONCLUSION To evaluate the dosimetric impact of using potentially inaccurate auto contours directly for treatment planning, the breast doses were evaluated from planned RefPlan. The current results indicate that clinician manual contouring using the RTOG Atlas is not reasonably concordant. The differences between clinician contours and auto contours may be due to the DirectORGANS algorithm contour a larger breast volume than RTOG-atlas guidelines adhered to by clinicians. While the DirectORGANS algorithm can serve as an initial tool, clinicians will need to adjust the contours to ensure conformity with the RTOG atlas. DirectORGANS algorithm is suitable for use in RT planning to minimize differences between physicians and shorten the duration of this contouring step.