The performances of two autosegmentation algorithms were evaluated on 28 anonymized pelvic CT scans as a pilot study for the clinical implementation of a semi-automatic workflow. Four organs at risk (OARs), namely the rectum, bladder, and femoral heads, were contoured manually by an expert radiation oncologist (RO)—considered as the ground truth (GT)—and by model-based segmentation (MBS) and deep learning (DL) algorithms. Autocontouring performances were evaluated using a qualitative scoring system, contouring time analysis, and five geometrical indices: the 95th percentile Hausdorff Distance (95HD), Dice Similarity Coefficient (DSC), Surface Dice Similarity Coefficient (SDSC), Added Path Length (APL), and Relative Added Path Length (RAPL). Considering total median value for the four OARs, both MBS and DL showed clinically acceptable results with differences between the two algorithms being not statistically significant for almost all indices. The DL autocontouring algorithm achieved high geometric accuracy, high scores from the ROs, and consistent performances with all validation indices for every OAR. The MBS algorithm achieved high geometric accuracy for the femoral heads and bladder. The DL algorithm required 30 s to contour all the OARs, and the MBS algorithm required 90 s, showing a time gain compared with the manual contours, which took 20 min for each case. The DL autocontouring algorithm obtained promising but preliminary results with every evaluation metric and for every analyzed OAR. The application of the MBS algorithm as the only contouring tool still presents challenges.

Accurate dose calculation is crucial in radiation therapy, particularly when employing advanced techniques such as VMAT. Monte Carlo (MC) algorithms are widely recognized as the gold standard for dose calculation. This study aimed to validate an MC dose calculation model in RayStation^® TPS and to compare its performance against the Collapsed Cone (CC). The MC model was evaluated using measured data from four matched Elekta^® Versa HD LINACs. Validation followed AAPM TG-157, TG-244 and IAEA TECDOC guidelines, including homogeneous media, heterogeneous media and advanced techniques (VMAT, IMRT, dMLC). MC showed high agreement with measurements, with most results within a 2% dose difference or passing a gamma criteria ≥95%. Performance was comparable or superior to CC in many cases, especially in small-field stereotactic plans. Impact of grid size and statistical uncertainty was assessed to find the optimal values. This study indicated a 2 mm grid size with 0.5% uncertainty as the optimal balance between accuracy and computation time. In conclusion, this study demonstrated that the MC model can provide accurate dose calculations for treatment plans, with computation times compatible with clinical practice. The use of MC appears to be particularly advantageous for advanced techniques and in the presence of tissue heterogeneities.