2190 - Quantification of Lung Target Delineation Accuracy on High-Quality Cone-Beam CT

Purpose/Objective(s): Recent developments in high-quality CBCT imaging have allowed direct-to-unit (DTU) treatment, which forgoes simulation CTs. One potential use case is DTU for single-fraction lung SBRT. However, the accuracy of delineating a moving lung lesion on a free-breathing high-fidelity CBCT scan has not been evaluated. Here we investigate the quantitative accuracy of target delineation in high-fidelity free-breathing CBCT scans compared to an average 4D as ground truth, using patient scans and digital phantoms for lung tumors.

Materials/Methods: CBCT and 4DCT scans of three lung cancer patients treated with DTU were used. The gross tumor volume (GTV) was contoured as visualized on a 60-second free-breathing CBCT. As 60-sec CBCT is an average estimate of all breathing phases, the envelope of the moving GTV was assumed to estimate internal target volume (ITV). In addition to the patient scans, twelve digital thoracic lung models with incorporated respiratory and cardiac motions with various tumor sizes (diameter 5-20 mm) and motion ranges (5-15 mm) were generated using the “4D XCAT digital thoracic phantom” software. The digital models were then used to create 2D projection images from the corresponding acquisition geometry of 3D CBCT data. Tomographic image reconstruction was performed on the projection data using Feldkamp-David-Kress algorithm available by Reconstruction Toolkit (RTK), an open-source library. The delineated targets on reconstructed CBCT and average 4D scans were quantitatively compared for their volumes, distance from the center of mass (COM) and Dice similarity coefficients (DSC).

Results: Results for three clinical patients and 12 digital phantoms are shown in Table 1. For small tumors with a diameter < 10 mm, regardless of tumor motion, the target delineation on 60-sec CBCT showed low similarity to the ground truth (mean DSC=0.45). For tumor sizes >15 mm, CBCT exhibited improved similarity to the ground truth (mean DSC=0.8).

Conclusion: This work presents a quantitative analysis of lung target delineation on high fidelity CBCT. Our analysis shows that a 60-second CBCT scan under-delineates the tumor motion compared to average 4D ground truth. Therefore, margin uncertainty should be accounted for when CBCT scans are used for target delineation. Future work is needed to investigate the dosimetric consequences of the observed target delineation uncertainty.