Evaluation Of 4d Cone Beam Computed Tomography For Target Localization In Stereotactic Body Radiation Therapy For Early Lung Cancer

To assess the accuracy and reproducibility of automated registration of 4D cone-beam CT (CBCT) in target localization and verification for lung stereotactic body radiation therapy (SBRT) compared to 3D CBCT (manual and automatic soft tissue localization). Early stage non-small cell lung carcinoma patients were treated with image guided SBRT using a VMAT technique. Localization images were obtained using CBCT before shift, after shift and after treatment on each treatment day. Tumor registration on 3D CBCT was performed by four independent observers as well as automatically with dual registration using vertebral clipbox and soft tissue mask. Dual registration with clipbox and mask was also performed on 4D reconstructed CBCT. Krippendorff’s alpha was used to assess agreement and Bland-Altman analysis was used to assess limits of agreement. Six patients and 114 cone beam images were analyzed. Four right-sided and 2 left-sided tumors were assessed with an equal number of central and peripheral tumors. Mean (+/- SD) shift amplitude for manual registration was 2.54 ± 2.65 mm, 2.98 ± 2.83 mm, and 1.93 ± 1.40 mm in the right-left (RL), cranio-caudal (CC) and anterior posterior (AP) direction respectively. Agreement between observers was very high with Krippendorff’s alpha greater than 0.87 in all directions. Automatic dual registration on 3D CBCT produced similar shifts with the mean difference between mean manual and automatic techniques of 0.4 mm in all directions. Dual registration on 4D CBCT produced differences of 0.2 mm RL, 0.4 mm CC and 0.02 mm AP. Agreement between both manual and 3D automatic and manual and 4D automatic was high with Krippendorff’s alpha greater than 0.9 in all directions. Maximum limits of agreement were -0.21 to 0.13 and -0.10 to 0.19 for the 3D and 4D automatic techniques, respectively. Linear regression analysis revealed association between maximum respiratory motion and disagreement between manual and automatic 3D registration (R2 = 0.41, p < 0.01), but did not reveal a similar correlation with 4D CBCT (R2 = 0.02, p = 0.10). Automatic dual registration using 4D CBCT provides accurate tumor localization and unlike automatic dual registration using 3D CBCT is unaffected by large respiratory motion. Four-dimensional CBCT reduces the potential for systematic error making it a desirable approach for image guidance.