ESTRO 2024 - Abstract Book

S4140

Physics - Intra-fraction motion management and real-time adaptive radiotherapy

ESTRO 2024

[2] Kierkels, R.G.J. et al., (2019). Automated Robust Proton Planning Using Dose-Volume Histogram-Based Mimicking of the Photon Reference Dose and Reducing Organ at Risk Dose Optimization. International Journal of Radiation Oncology*Biology*Physics, 103(1), 251–258. https://doi.org/10.1016/j.ijrobp.2018.08.023

72

Digital Poster

Prediction accuracy of lung tumor stereotactic body radiotherapy using CyberKnife Synchrony tracking

Tin Lok Chiu, Tsz Ching Fok, Natalie Kar Wei Yip, Fu Ki Yeung, Chi Wai Cheung, Siu Ki Yu

Hong Kong Sanatorium & Hospital, Medical Physics Department, Hong Kong, Hong Kong

Purpose/Objective:

The CyberKnife Synchrony® system enables active tumor tracking. Errors in prediction models should be quantified to accurately determine tumor margins, and their relationships with anatomical characteristics should be established.

Material/Methods:

Data from 42 patient with lesions in the thoracic cavity who underwent CyberKnife SBRT were analyzed. Log files of their treatment sessions were extracted. 21 patients underwent fiducial tracking treatment, and 17 patients underwent XSight® lung tracking. 48 tumors were treated, and 45 plans were included in the analysis. Margins due to prediction errors and the difference between the current target position and the prediction made 115ms earlier, which were due to the head latency of the linear particle accelerator, were determined in each anatomical direction using an expansion method to include 95% and 99% target coverage. The statistical distributions of these errors were investigated. Moreover, the relationships between errors and tumor anatomical characteristics were quantified and analyzed. For the expansion method 1 , a box was defined, such that the volume was the smallest available dimension to include certain percentage of data points. The first step was to determine the distance that covered that percentage of data points along each axis. After acquiring these initial dimensions, the Euclidean distance between data points outside the box and the origin was calculated. Starting from the closest data point, the box was expanded to include this point and the dimensions were updated. This step was repeated until the coverage was increased to the defined percentage. In this way, the six directions of the box for each treatment session were determined. The location and anatomical characteristics of a tumor affect the prediction errors. Hence, the relationship between the margin expanded for the prediction error was investigated using a linear mixed model. Using a backward selection model, the statistically significant parameters were retained to construct the linear model. Moreover, these error terms were split into groups according to the targets. The interfractional changes in the correlation and prediction errors were analyzed by measuring the standard deviation within the treatment course. The correlation between these standard deviation values and anatomical characteristics was also assessed to analyze their effect. The following tumor characteristics were included: the lateral distance from the sternum; the perpendicular distance to the posterior chest wall; the perpendicular distance to the apex of the lung; the perpendicular distance to the