Influence of the correlation modeling period on the prediction accuracy of infrared marker-based dynamic tumor tracking using a gimbaled X-ray head

Purpose: To assess the utility of 10 s and 20 s modeling periods, rather than the 40 s currently used, in the clinical construction of practical correlation models (CMs) in dynamic tumor tracking irradiation using the Vero4DRT.

Methods: The CMs with five independent parameters (CM parameters) were analyzed retrospectively for 10 consecutive lung cancer patients. CM remodeling was performed two or three times per treatment session. Three different CMs trained over modeling periods of 10, 20, and 40 s were built from a single, original CM log file. The predicted target positions were calculated from the CM parameters and the vertical displacement of infrared markers on the abdomen (PIR) during the modeling. We assessed how the CM parameters obtained over modeling periods of T s (T = 10, 20, and 40 s) were robust to changes in respiratory patterns after several minutes. The mimic-predicted target positions after several minutes were computed based on the previous CM parameters and PIR during the next modeling. The 95th percentiles of the differences between mimic-predicted and detected target positions over 40 s (E95robust,T: T = 10, 20, and 40 s) were then calculated.

Results: Strong correlations greater than 0.92 were observed between the E95robust,20 and E95robust,40 values. Meanwhile, irregular respiratory patterns with inconsistent amplitudes of motion created differences between the E95robust,10 and E95robust,40 values of ≥10 mm.

Conclusions: The accuracies of CMs derived using 20 s were almost identical to those obtained over 40 s, and superior to those obtained over 10 s.