Copenhagen, Denmark

ESTRO 2022

Session Item

May 07
14:15 - 15:15
Poster Station 1
05: Intra-fraction & real-time adaptation
Jan-Jakob Sonke, The Netherlands
Poster Discussion
AI-based online adaptive CBCT-guided radiotherapy for bladder cancer using SIB and fiducial markers
Sana Azzarouali, The Netherlands


AI-based online adaptive CBCT-guided radiotherapy for bladder cancer using SIB and fiducial markers

Sana Azzarouali1, Karin Goudschaal1, Duncan den Boer1, Jorrit Visser1, Maarten Hulshof1, Arjan Bel1

1Amsterdam UMC, Radiation Oncology, Amsterdam, The Netherlands

Show Affiliations
Purpose or Objective

Accurate external beam radiotherapy of the bladder with a simultaneous integrated focal boost (SIB) is challenging due to variable bladder filling. With daily online adaptation of the GTVs and OARs using AI-driven Cone Beam CT (CBCT)-guided radiotherapy, we aim to reduce toxicity while maintaining target coverage. The purpose of this is study is to analyze the workflow in the presence of fiducial markers and to further enhance the accuracy of the boost dose on the tumor.

Material and Methods

Five patients with muscle invasive bladder cancer were treated on a ring-based linac integrated with a CBCT and software platform for both treatment planning and delivery (Ethos, Varian, USA). In 20 fractions the bladder and first lymph nodes received a dose of 40 Gy combined with a SIB of 15 Gy to the tumor. Fiducial markers were used in four patients. Two pretreatment CT images were made at t = 0 and t = 15 min. The first planning CT was used for manual delineation and to make a reference plan (VMAT, 6MV FFF). A PTV margin of 7 mm was used, but patient specifically extended in those directions where the PTV did not cover the complete bladder on second pretreatment CT due to intrafraction bladder filling . During each fraction a synthetic CT scan (sCT) was produced by deformable registration of the planning CT to the CBCT. A structure set based on the anatomy of the day was generated by using deformable registration and a convolutional neural network. Manual corrections to the target structure were performed if necessary, after which a scheduled plan was generated by calculating the dose of the reference plan on the sCT. Subsequently, an adapted plan was generated by running a new optimization. A second CBCT was acquired for position verification prior to delivery. To evaluate the fraction and whether the bladder was covered by the PTV, a post-treatment CBCT was acquired. The duration of each step, manual corrections, planning decisions, intrafractional bladder filling and target coverage were monitored.


The median treatment time was 32 min (Fig. 1A). For each week the treatment time was significantly lower than the week before except for week 3 demonstrating the presence of a learning curve (Fig. 1B). Compared to the fully automatic workflow, an additional 5 min was needed if manual corrections were done. Manual corrections were made in 77% of all fractions. The adaptive plan was used in 99% of all fractions. For the adaptive and scheduled plan 100% and 60% of the cases resulted in a PTV V95>98%, respectively (Fig. 2A). The bladder was covered by the PTV on CBCT3 in 87% of the fractions (Fig. 2B). In the other 13% a median of 0.9 cm of the bladder was outside the PTV due to intrafractional bladder filling.


This study shows that daily online adaptive CBCT-based RT with SIB is feasible for bladder cancer. However, the workflow would benefit from a shortening of the calculation time and lowering the need for manual corrections to reduce the effect of intrafractional bladder filling.