ESTRO 2022 - Abstract Book

S1498

Abstract book

ESTRO 2022

Ten historic patients treated with 50.4 Gy/28 fractions for cervical cancer were selected (5 with CBCTs from Truebeam, 5 with CBCTs from Ethos). For each patient, an initial (scheduled) Ethos 12-field IMRT plan was auto-optimised using a custom template. Ten treatment sessions (sampled throughout the course) were simulated. For each fraction, the Ethos system contoured “influencer” structures (bladder, bowel, rectum and uterus) and propagated target structures on the CBCT, these were edited by either a radiation oncologist or radiation therapist. The Ethos system then recalculated the scheduled plan (SP) and a newly-optimised ART plan (AP) based on the new contours. Margins followed local protocols throughout. Each patient course was simulated twice, once selecting the SP and once the AP each time. Ethos provides dose accumulation over the course of a treatment based on propagation of the delivered dose during each delivered fraction. A comparison of dose delivered using the SP (representing current practice) against the AP was performed using Wilcoxon signed rank test. Reported dose values have been scaled up from the simulated 10 deliveries to 28 fractions for direct comparison to clinical protocols. The time taken for each simulated session was noted. Results The mean CTV d min was increased by 6% (range -0.8 to 24.6) using oART (p<0.01). The mean CTV d max was reduced by 0.7% (p=0.04). Overall, CTV coverage was improved with oART, though not statistically significant at all levels, e.g. D 99% increased by 0.5 Gy using oART (p=0.06), whereas the D 99.5% increased by 1.4 Gy (p=0.02). As shown in Table 1, all OAR differences which were statistically significant were in favour of oART. The average time taken for each simulated session was 21 minutes (range 12-34). Conclusion Whilst overall there was some small benefit seen for oART (increased CTV coverage and reduced OAR doses) there was large inter-patient variation in the benefit of oART. This indicates that even for a traditionally mobile target such as the cervix-uterus complex, criteria for patient selection is required to determine if oART will provide a significant benefit. Further work investigating the dosimetric impact of CBCT-guided oART in cervical cancer in the live setting is ongoing. Improved accuracy of oART should allow margin reduction, leading to further reductions in OAR dose. 1 Royal North Shore Hospital, Northern Sydney Cancer Centre, Sydney, Australia; 2 The University of Sydney, ACRF Image X Institute, Faculty of Medicine and Health, Sydney, Australia Purpose or Objective Accurate fiducial marker segmentation is essential for kV-guided intra-fraction motion management to enable stereotactic ablative radiotherapy of the pancreas. We developed a compact convolutional neural network (CNN) model with 4 layers for marker segmentation in the prostate with excellent sensitivity (99.0%) and specificity (98.9%). Deep learning techniques don’t require additional learning imaging, prior marker properties (such as shape or orientation) and they are applicable to kV images. In this study, we further develop our CNN model for marker tracking applied to pancreatic cancer patient data. Materials and Methods We evaluated a CNN with 6 layers and a transfer learning approach from pretrained compact CNN. Training data from the ethics approved SPAN-C Trial for pancreas SABR was utilised. The training dataset contained both cone beam computed tomography (CBCT) projections and kV triggered images acquired during treatment (a total of 23 fractions of 7 patients) for pancreas patients with implanted fiducial markers. Data augmentation was also performed for subimages which contained markers. The total dataset had 1.3 million subimages. The CNN with 6 layers was trained on the full dataset and the transfer learning approach was trained with 32.3% of the full dataset. Cross validation based early stopping was employed to avoid overfitting for both. The performance of each model was tested on unseen CBCT and kV images from 5 fractions of 2 patients. The sensitivity, specificity, the Receiver Operating Characteristic (ROC) curve and the Area Under the ROC (AUC) plot were evaluated. The root-mean-square error (RMSE) was calculated for the centroid of the markers predicted by the CNN models, relative to the manually segmented ground truth. Results The sensitivity and specificity of the fully trained CNN was 98.4% and 99.0%, respectively, while the transfer learning model had 94.3% and 99.3%, respectively. The AUC of the fully trained model and that of the transfer learning model was 0.9887 and 0.9889, respectively. The mean RMSE of the fully trained CNN was 0.20 ± 0.03 mm and 0.35 ± 0.05 mm in x and y directions (of kV image), respectively, while the transfer learning had 0.15 ± 0.02 mm and 0.35 ± 0.04 mm in x and y directions, respectively. Conclusion A deep learning approach was implemented to classify implanted fiducial markers in pancreatic cancer patient data. The accuracy of marker position prediction by the CNN models from the ground truth was submillimeter as required for stereotactic ablative radiotherapy of the pancreas. PO-1698 Evaluation of deep learning based fiducial markers segmentation in pancreatic cancer patients A.M. Ahmed 1 , A. Mylonas 2 , M. Gargett 1 , D. Chrystall 1 , A. Briggs 1 , D.T. Nguyen 2 , P. Keall 2 , A. Kneebone 1 , G. Hruby 1 , J. Booth 1

PO-1699 Clinical experience with expiration gated 10MV stereotactic lung radiotherapy

I. Remmerts de Vries 1 , M. Dahele 1 , T. Rosario 1 , B. Slotman 1 , W. Verbakel 1