ESTRO 2023 - Abstract Book

S1691

Digital Posters

ESTRO 2023

adequate for the clinically observed variation. Therefore, starting from a clinical constraint of a CTV D95% ≥ 98% throughout treatment, we assess the number of false positives (replanning when CTV D95% ≥ 98%, respectively), as well as false negatives (no action taken when CTV D95%<98%, respectively). Results Between January and September 2022, 93 patients received a robust VMAT plan. 13 cases (39 CBCT’s) were analyzed due to a deviation >8 mm during treatment. Means ± standard deviations of PTV V95% were: on the plan CT 98.3%±0.5%, on the defCTs 96.7%±1.3%; on the CBCT’s 91.8%±4.2%. Respective figures for CTV V95% were: 99.9%±0.1%; 99.5%±0.6%; and 98%±2.4%. See Fig. 1. In 9/13 cases, an adapted plan was made. On the basis of CTV D95% ≥ 98%, we observed 3 false negatives and 8 false positives.

Figure 1: Boxplots of PTV (left) and CTV (right) V95%.

Conclusion The robust optimization ensured CTV coverage throughout the treatment for 90/93 (96.8%) of patients. The high false positive rate suggests there may be room to adjust robustness parameters for higher conformality without losing robustness. However, the false negatives indicate that the robust optimization is unable to predict all plan deviations in practice, and therefore evaluation on the CBCT remains an indispensable clinical tool.

[1] K. Crama et al, Radiot & Oncol, pp. S1039-S1040, PO-1864, 2020. [2] A. Dunlop et al, Clin. Transl. Rad. Onc., vol. 16, pp. 60-66, 2019.

PO-1936 Contour propagation and uncertainty estimation using deep learning in head and neck treatments

L. Rivetti 1 , A. Studen 2,3 , M. Sharma 4 , J. Chan 5 , R. Jeraj 6,3,7

1 University of Ljubljana, Faculty of Mathematics and Physics, Ljubljana, Slovenia; 2 Jo ž ef Stefan Institute, Experimental Particle Physics Department, Ljubljana, Slovenia; 3 University of Ljubljana , Faculty of Mathematics and Physics , Ljubljana, Slovenia; 4 University of California, Department of Radiation Oncology, San Francisco, USA; 5 University of California , Department of Radiation Oncology , San Francisco, USA; 6 Universirty of Wisconsin, Department of Medical Phyisics, Madison, USA; 7 Jo ž ef Stefan Institute, Reactor Physics, Ljubljana, Slovenia Purpose or Objective One of the key aspects of the daily on-line adaptive radiotherapy (DART) workflow is to ensure fast and accurate contours on the daily imaging. Convolutional neural networks (CNNs) were used to model a dense displacement field (DDF) which propagate the planning structures to the new daily anatomy. Although these methods are fast, none of them produce interpretable uncertainties of the DDF, leading to a careful review of the contours by the physicians. The purpose of this work is to develop a method which can quickly generate the daily structures along with its interpretable uncertainties to focus physician attention to critical regions and thus reduce intervention time. Materials and Methods In this work, CNNs combined with drop-out layers were used to generate a distribution of the DDF which maps the planning CT to the daily CBCTs in head and neck treatments. Daily probability maps of the position of all the OARs and targets were generated averaging the planning structures propagated with 100 samples of the DDF distribution. The performance of the method was assessed by calculating the dice similarity coefficient (DICE) and the target registration error (TRE) between structures propagated with the mean DDF and their daily ground-truth. In addition, the results were compared to those obtained using a different DIR method (Elastix). The predicted DDF uncertainty was assessed as the isolation of two source of uncertainties; image uncertainty (related to regions of the image with very low contrast tissue), and model uncertainty (related to an improper matching of the two images). Results It was found that the DICE and TRE distributions calculated with both methods (Elastix and the presented method) were not statistically significantly different for all the 11 structures evaluated in each patient of the test set. The esophagus structure had the lowest mean dice coefficient (0.73) and the maximum mean TRE (3.7 mm). Our method performed the registration in the order of seconds while Elastix required minutes. The mean standard deviation of the DDF obtained in