ESTRO 2021 Abstract Book

S262

ESTRO 2021

Rotterdam, The Netherlands

Purpose or Objective CyberKnife’s Synchrony tumor tracking is based on continuous position measurements of external fiducials and a linear or dual quadratic correlation model to translate measured external markers motion into internal tumor motion, the latter being compensated by tracking. Prior to start of treatment, the correlation model is built with 15 X-rays taken in different phases of the patient’s respiratory cycle. During treatment, every 45- 150 sec. a new X-ray image is taken to update the correlation model, following the first in, first out principle for acquired images. We hypothesized that in the presence of slow drifts in respiratory motion, the currently implemented correlation models would be inadequate, and that use of correlation models with an explicit time dependence would improve tracking accuracy. A new tracking algorithm applying such models was implemented and validated. Materials and Methods In current Synchrony tumor tracking, linear correlation models follow the expression x int = α·x ext +b (L models). In this study, we have introduced the use of correlation models that include a linear time dependence, x int =α · x ext +b(t) (L T models). In a decision tree, tracking can switch from one correlation model to the other, depending on presence of statistically relevant time trends. To validate the new tracking models for a large number (120.000) of patients, a respiratory track generator was built for typical robotic treatments (≥20min). In this generator, the well-known respiratory tracks defined by Lujan et al. (Med. Phys., 1999), were complemented with variations and disturbances in respiratory motion (RM) baseline, amplitude, period and shape, while also adding noise. For each generated motion track, the applied RM features were randomly drawn from distributions found in literature. Additionally, linear baseline drifts of 0, 0.25 and 0.5mm/min were added to motion tracks in CC direction. For all motion tracks, we compared the clinical Linear Motion Tracking, LMT, with the novel Mixed Motion Tracking, MMT, approach that can switch from L to L T models in case of detected baseline drifts. Results Comparisons between MMT and LMT are presented in Fig. 1 and Fig 2. MMT resulted in at least 1.6mm reduction in mean 3D tracking error for 10% of the investigated 120.000 patients. MMT was better than LMT in 93.3% and 96.7% of analyzed patients with 0.25 and 0.5mm/min linear baseline drifts, respectively. MMT was also able to mitigate low frequency baseline disturbances with at least 0.9mm reduction in mean 3D error for 10% of these patients group. MMT proved to be robust: only for 2% of investigated patients there was a deterioration of >0.2mm when using MMT instead of LMT.