A feasibility study of intrafractional tumor motion estimation based on 4D-CBCT using diaphragm as surrogate

To investigate the intrafractional stability of the motion relationship between the diaphragm and tumor, as well as the feasibility of using diaphragm motion to estimate lung tumor motion.

Eighty-five paired (pre and posttreatment) daily 4D-CBCT images were obtained from 20 lung cancer patients who underwent SBRT. Bony registration was performed between the pre- and post-CBCT images to exclude patient body movement. The end-exhalation phase image of the pre-CBCT image was selected as the reference image. Tumor positions were obtained for each phase image using contour-based translational alignments. Diaphragm positions were obtained by translational alignment of its apex position. A linear intrafraction model was constructed using regression analysis performed between the diaphragm and tumor positions manifested on the pretreatment 4D-CBCT images. By applying this model to posttreatment 4D-CBCT images, the tumor positions were estimated from posttreatment 4D-CBCT diaphragm positions and compared with measured values. A receiver operating characteristic (ROC) test was performed to determine a suitable indicator for predicting the estimate accuracy of the linear model.

Using the linear model, per-phase position, mean position, and excursion estimation errors were 1.12 ± 0.99 mm, 0.97 ± 0.88 mm, and 0.79 ± 0.67 mm, respectively. Intrafractional per-phase tumor position estimation error, mean position error, and excursion error were within 3 mm 95%, 96%, and 99% of the time, respectively. The residual sum of squares (RSS) determined from pretreatment images achieved the largest prediction power for the tumor position estimation error (discrepancy < 3 mm) with an Area Under ROC Curve (AUC) of 0.92 (P < 0.05).

Utilizing the relationship between diaphragm and tumor positions on the pretreatment 4D-CBCT image, intrafractional tumor positions were estimated from intrafractional diaphragm positions. The estimation accuracy can be predicted using the RSS obtained from the pretreatment 4D-CBCT image.