ESTRO 2021 Abstract Book

S369

ESTRO 2021

C. Seller Oria 1 , A. Thummerer 2 , J. Free 2 , J.A. Langendijk 2 , S. Both 2 , A.C. Knopf 2 , A. Meijers 2 1 University Medical Center Groningen, University of Groningen, Radiation Oncology, Groningen, The Netherlands; 2 University Medical Center Groningen, University of Groningen , Radiation Oncology, Groningen, The Netherlands Purpose or Objective Cone beam CT (CBCT) based synthetic CTs (sCT) produced with a deep convolutional neural network (DCNN) show high image quality, suggesting their potential usability in adaptive proton therapy workflows. However, the nature of workflows involving neural networks prevents the user from directly controlling their output. Therefore, quality control tools that monitor the sCTs and detect failures or outliers in the generated images are needed. This work evaluates the potential of using a range probing (RP) quality control tool to verify CBCT based sCTs generated by a DCNN. For the first time, in vivo RP measurements are used to experimentally assess the CT number accuracy in sCT images. Materials and Methods A RP quality control dataset of 7 head and neck cancer patients, consisting of repeat CTs (rCT), CBCTs and RP acquisitions, was retrospectively assessed. RP quality control measurements and rCTs were performed for in vivo dosimetry checks for head and neck cancer patients treated with intensity modulated proton therapy. Each RP field was composed of 9x9 pencil beams directed from a gantry angle of 90 degrees, with a spacing of 5 mm, thus covering an area of 40x40 mm 2 . The center of the RP field was aligned with the treatment isocenter, allowing pencil beams to intersect a wide variety of tissues (figure 1(a)). RP acquisitions in two different treatment fractions (referred as sessions 1 and 2) were collected for each patient (numbered from 1 to 7), resulting in a dataset of 14 RP fields with their corresponding rCT and CBCT. CBCT based sCTs were generated using a DCNN. Relative range errors between measured RP and simulated RP in rCTs, and between measured RP and simulated RP in sCTs, were computed and presented in the form of two-dimensional maps (figure 1(b)). The mean and 1.5 times the standard deviation were extracted from each relative range error map to assess the CT number accuracy in rCTs and sCTs. Range probes going through anatomically unstable regions (shoulders, trapezius muscles, base of the tongue and swallowing muscles) were excluded from the relative range error maps (figure 1(b)).

Results Mean relative range errors showed agreement between measured and simulated RP, ranging from -1.2% to 1.5% in rCTs, and from -0.7% to 2.7% in sCTs (figure 2). Standard deviations laid within a range of ±3% in rCTs and from -3% to 4.5% in sCTs.