ESTRO 2020 Abstract book

S201 ESTRO 2020

experienced oncology clinician, who scored each structure either pass or fail, dependent on conformity to the GS. The structures were then imported into CERR (Computational Environment for Radiotherapy Research) software to calculate conformity against the GS using the Jaccard Index, Dice Similarity Coefficient, van’t Riet Index, Hausdorff Distance, Geographical Miss Index and Disconcordance Index. The CI results and corresponding pass/fail score for each structure were imported into the Matlab Classification Learner Application in order for ML algorithms to decipher patterns between CI data (predictors) and the pass/fail score (response), building trained models that predict conformity of future structures based on their CI values. All preinstalled ML algorithms were applied to the CI data and trained models were produced. A 5-fold cross validation technique, which resamples the data to train and validate each model, was applied to determine the Predictive Accuracy (PA) of each model. The model with the highest PA was selected and exported as a script to implement in CERR. Results Figure 1 shows that different algorithms produced trained models with different PAs that vary by structure type. By separating all contours by structure type, trained models with higher PAs can be produced. Stomach produced models with the highest PA and liver GTV produced models with the lowest PA.

boundaries due to low contrast, resulting in larger variation in outlining and of the scoring of a pass/fail. These models can reduce the time required for benchmark case reviews, but cannot replace a clinical review entirely. Further work is required to refine and test these models before they can be used routinely.

Proffered Papers: Proffered papers 20: Ensuring precision and accuracy in RT

OC-0349 Inter-observer variation of image registration in an MR-only versus CT/MRI-based workflow M.A. Boon 1 , J. Visser 1 , F. Beeksma 1 , A.A. Goedhart 2 , M.M.C. Bijveld 3 , K.A. Hinnen 1 , K.N. Goudschaal 1 , Z. Van Kesteren 1 1 Amsterdam UMC, Radiation Oncology, Amsterdam, The Netherlands ; 2 Vrije Universiteit Amsterdam, Medical Natural Sciences, Amsterdam, The Netherlands ; 3 Catharina hospital, Radiation Oncology, Eindhoven, The Netherlands Purpose or Objective In current prostate cancer radiotherapy MRI is used for target definition whilst CT is needed for treatment planning. Co-registration of CT and MRI introduces a systematic error in the treatment, which might be reduced by introducing an MR-only workflow for target definition. In an MR-only workflow an additional sequence is required and registrations between MRI sequences are needed because of inter-sequence movement within the patient. The systematic error can be reduced by removing the time between MRI and CT acquisition. The aim of this study was to quantify the inter-observer registration error (IOE) of a CT/MR-based and an MR-only workflow for prostate cancer radiotherapy. Material and Methods Twenty prostate cancer patients, treated with external beam radiotherapy at our institute, were included in this study after giving informed consent. CT and MRI data were acquired on a RT-couch top in RT position, with less than two hours between the CT and MRI. The MRI scan consisted of a transversal T2-weighted (TRA) image for target definition and a dedicated acquisition for gold fiducial identification (BTFE SPAIR). In addition, as part of this study, an extra (mDixon) scan for pseudo-CT (MRCAT) generation was acquired. It was possible to identify the four gold fiducials on the water-weighted reconstruction of the mDixon sequence. Seven experienced observers executed two manually registrations based on gold fiducials. Each observer registered the CT, via the BTFE SPAIR to the T2W image (CT-TRA) and the mDixon sequence was registered to the T2W image (MRCAT-TRA). The registration yielded rotations and translations defined with the center of mass of the fiducials as rotation point. The IOE of the CT-TRA registration was compared to the MRCAT-TRA registration on patient and on group level. The IOE was quantified by the inter quartile range (IQR) of the registration results per patient, and by the standard deviation (SD) of the registration results of the group. To be able to pool the data between patients, the mean of the registration values per patient was set to zero. Significance of difference in IOE on patient level was tested using a paired two-sided Wilcoxon signed rank test. A non-parametric Levene’s test was used to test the difference in IOE for the whole group. Results

Conclusion When contours are separated by structure type, models with higher PAs are seen. This suggests the suitability of algorithm model is dependent on structure type. The liver GTV models produced the lowest PAs. This could be due to difficulties in delineating the liver GTV