Purpose: This study aims to develop a deep learning-based method to predict the ITV in CBCT images for Lung SBRT patients and to evaluate the dosimetry of a lung SBRT treatment plan with the target of ITVpredict and the feasibility of its clinical application.
Methods: We used DL model (Mask R-CNN) to identify and delineate lung tumor ITV in CBCT images, enrolled 78 phantom cases and 156 patient cases who receive SBRT treatment. The results of the DL-based method were compared with the ground truth (4DCT) using four metrics, including DSC, Relative Volume Index (RVI), 3D Motion Range (R3D), and HD. Moreover, we involved 60 SBRT patients to evaluate the dosimetry of a lung SBRT treatment plan with the ITVpredict. The dosimetric parameters included R100%, R50%, Gradient Index (GI) and D2cm, which were evaluated via Plan4DCT and Planpredict.
Results: The DSC value for 4DCTAVG vs. CBCT was 0.83 ± 0.18, indicating a high similarity of tumor contouring in CBCT and 4DCT. The R3D, RVI, and HD values was 0.94±0.04, 1.37±0.36, 0.79±0.02, and 6.79±0.68, respectively. These results showed a good correlation between ITVpredict and ITV4DCT.However, the average volume of ITVpredict was 10% less than that of ITV4DCT (P = .333). No significant difference in dose coverage was found in V100% for the ITV with 99.98±0.04% in the ITV4DCT versus 97.56±4.71% in the ITVpredict (P = .162). Dosimetry parameters of PTV, including R100%, R50%, GI and D2cm showed no statistically significant difference between plans(p>0.05).
Conclusion: The current improved method yielded a good similarity between ITVpredict and ITV4DCT. This study confirmed that the treatment plan based on ITVpredict produced by our model automatically could meet clinical requirements. Thus, for lung SBRT patients, the model has great potential for using CBCT images for ITV contouring, which can be used in treatment planning.
Cone-beam CT, Treatment Planning