Session Item

This study investigates the accuracy of four models for estimating internal liver tumor motion based on continuous monitoring of external marker motion combined with four strategies for sparse intrafractional imaging of the internal tumor position.

The study includes 15 patients with liver tumors previously treated in free breathing with three-fraction SBRT. At all fractions the 3D internal tumor motion (INT) was monitored by implanted electromagnetic transponders (Calypso) while the vertical external motion (EXT) of a marker block on the patient’s abdomen was monitored by a camera. The ability of the following four external-internal motion correlation models (ECM) to estimate INT as function of EXT was investigated: (1) a simple linear model, (2) an augmented linear model, (3) an augmented quadratic model, and (4) an extended quadratic model (Fig 1.a). First, the ECM was constructed by fitting the internal and external motion during the first 60s of each fraction, and the fit accuracy was calculated as the root-mean-square error (RMSE) between ECM estimated and actual internal motion. Next, the following four strategies for updating each ECM during the remaining part of the fraction were simulated, and the RMSE of the estimated tumor motion was calculated: (A) no ECM update, (B) sampling the internal tumor position every 3s and continuously update all ECM parameters based on samples of the last 2 minutes, (C) sampling the internal tumor position every 3s and continuously updating the constant term of the ECM based on the five last samples, (D) sampling the internal tumor motion continuously for 20s before each field delivery and updating the entire ECM based on these samples.

The mean duration of the investigated intrafraction motion traces was 25 min [range: 14-44 min]. The fit accuracy was in general best for the augmented quadratic model (ECM 3, Fig 1.b). Cases with a simple linear correlation between internal and external motion were well described by all ECMs (Fig 2, column 1). For cases with hysteresis, augmented ECMs (2,3,4) were needed to fit the data well (Fig 2, column 2), while quadratic ECMs (3-4) were needed for non-linear (banana-shaped) motion (Fig 2, column 3).
The best estimation of tumor motion throughout the treatment fractions was obtained with the augmented linear ECM 2, which could account for hysteresis while being more robust against irregular motion than the quadratic ECMs (Fig 1.b). The best ECM update strategy was by regular tumor position sampling every 3s throughout the treatment (Strategies B-C).

Although the augmented quadratic ECM in general had higher fit accuracy, the augmented linear ECM combined with continuous updates every 3s provided the most accurate tumor position estimation throughout the fraction. This strategy could provide accurate internal motion monitoring based on an external gating camera and sparse 3Hz kV imaging, which is possible at conventional linacs.