ESTRO 2023 - Abstract Book

S1889

Digital Posters

ESTRO 2023

Purpose or Objective Glioma patients undergoing stereotactic brain radiation therapy (SBRT) must do MRI localization to ensure the accurate location of the target during treatment. If using pseudo-CT generated from planning MRI, the patient only needs a quick MRI scan which can not only provide excellent soft-tissue contrast for the target and organs at risk (OAR), but also provide CT-like density information for dose calculation. Cumbersome and error-prone CT-MRI registration can also be eliminated to reduce planning delays. The study aims to evaluate the feasibility of commercial synthetic CT software in stereotactic radiotherapy for glioma by using multi-omics evaluation methods. Materials and Methods All glioma patients who had planning MRIs and were treated in SBRT from 2021 to 2022 in our institution were included. The Philips MRCAT Brain based on a convolutional neural network converts planning MRIs to pseudo-CTs. The pseudo-CTs use initial delineation, registration, and treatment plans (Eclipse version ： 15.6) of planning CTs. Dosimetric validation was performed using main dose-volume histogram (DVH) endpoints in respect to SABR 6.1 guidelines (Dmin, Dmax, Dmean, D95%, R50%, R100%) as well as local and global 3D gamma analysis with 1-3%/2mm, 1%/1mm criteria and a 10% threshold to the maximum dose. And tumor control probability (TCP) of planning target volume (PTV) and normal tissue complication probability (NTCP) of OARs were used for biological evaluation. Then the original radiomics features of PTV and OARs were extracted for radiomics comparison. Wilcox-test was used for comparison between initial planning CTs and synthetic CTs, and a significant difference level was set at 0.05. Results 34 patients were included, with 34 planning CTs and mDixon-T1pseudo-CTs, and 30 mDixon-T1-CE pseudo-CTs. The median PTV was 3.4 cc (range 0.5-27.3), with no differences in DVH metrics for PTVs and OARs. And there was no difference between the two pseudo-CT cohorts. The median 3D local gamma passing rates (1%/2mm, 1%/1mm) between planning CTs and mDixon-T1 pseudo-CTs were 99.4% (range 88.1%-100%), 92.6% (range 63.5%-99.6%). In biological evaluation, all differences of TCPs were <0.3%, normal brain's NTCPs were < 2%, and other OAR’s NTCPs were < 0.02%. Then we analyzed 60 original radiomics features (47 outliers were excluded). It was found that more than 85% features of PTV and OARs have differences. Conclusion The study is the first to compare pseudo-CTs generated by MRCAT Brain from planning MRIs with initial planning CTs for SBRT in glioma. These quantitative results strongly indicate pseudo-CTs are accurate enough to replace current planning CTs for dose calculation, and evaluation of NTCP/ TCP. And it demonstrates the great potential for MRI to replace CT in the process of head simulation and treatment planning. However, pseudo-CT can’t replace the analysis of radiomics, which also provides a direction for the improvement of pseudo-CT generation technology. Purpose or Objective Convolutional Neural Networks (CNN) can process imaging data with applications that span multiple medical disciplines. CNNs are considered to be less sensitive to data characteristics than traditional machine learning methods, however, the pre-processing of data can significantly impact the CNN performance. A common example of such pre-processing is applying a Hounsfield Unit (HU) window level and width to CT images to exclude artifacts and limit the range of tissues captured. In this work, we aim to explore the impact of windowing CT data for a CNN that predicts distant metastasis in head and neck (H&N) cancer patients. Materials and Methods We included 435 patients diagnosed with H&N cancer from two distinct publicly available datasets. One dataset, comprising 298 patients from 4 different institutions, was used for training and validation, leaving the other dataset, from a cohort with 137 patients, exclusively for testing. The gross tumor volume (GTV) was extracted from the pre-treatment CT scans using the mask delineations defined by an expert, cropped around the GTV region, resampled to a uniform pixel spacing (1 x 1 mm3), and windowed. For this last step, we considered different windowing parameters to evaluate the impact on the model’s performance. Firstly, a window based on a previous study on CNN models for H&N cancer outcome prediction, with a level of 0 Hounsfield units (HU) and a width of 1000 HU. Furthermore, a window with a level of 0 HU and a width of 500 HU to assess the effect of employing a narrower window. Lastly, we incorporated a window with a level of 125 HU and a width of 350 HU based on the expected interval of the Hounsfield scale for the tissues in the H&N region (e.g., mucosal, soft tissue). The CNN structure was adapted from a previous study and optimized for the prediction of 2-year distant metastasis. The model was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) metric. Results The model had the best performance with a window level of 125 HU and a width of 350 HU. This model achieved an average AUC of 0.88, 0.87, and 0.85 in the training, validation, and testing sets. The use of a wider window resulted in a lower AUC across the three subsets of data. Applying the windows with a width of 1000 HU and 500 HU resulted in an average AUC of 0.83 and 0.86 in the training set, 0.81 and 0.83 in the validation set, and 0.80 and 0.78 in the testing set. Significant differences (Kruskal-Wallis, p<0.05) were found for the validation and testing sets when comparing the AUCs of the best- performing window to the AUCs of models trained using the other two windows. Conclusion A CNN may attenuate differences in the pre-processing of imaging data due to its flexibility. However, pre-processing can still significantly impact a model's performance and generalizability. Mainly the windowing of CT scans using parameters based on the HUs of relevant tissues can positively impact the model's prediction ability. PO-2105 Impact of windowing CT scans on the performance of a CNN for head and neck cancer prognosis P. Mateus 1 , I. Bermejo 1 , A. Dekker 1 1 Maastricht University, Radiotherapy, Maastricht, The Netherlands