ESTRO 2020 Abstract book

S197 ESTRO 2020

RO group. AI-assistance improved both accuracy and efficiency for the junior RO group, which could facilitate implementation of institutional contouring standards and improvement of practice consistency. OC-0344 Automatic contouring of diffusion-weighted MRI from an MR-Linac using a convolutional neural network O. Gurney-Champion 1 , J. Kieselmann 1 , W. Kee 2 , B. Ng- Cheng-Hin 3 , K. Newbold 2 , K. Harrington 3 , U. Oelfke 1 1 The Institute of Cancer Research and The Royal Marsden NHS Foundation Trust, The Joint Department of Physics, London, United Kingdom ; 2 The Royal Marsden NHS Foundation Trust, Clinical Oncology, London, United Kingdom ; 3 The Institute of Cancer Research and The Royal Marsden NHS Foundation Trust, Targeted Therapy team, London, United Kingdom Purpose or Objective The MR-Linac allows for easy daily acquisition of diffusion- weighted imaging (DWI). Retrieving quantitative MRI information (the apparent diffusion coefficient; ADC) from DWI involves contouring regions of interest. This contouring is time-consuming, particularly with daily repeated DWI, and is prone to manual errors. Therefore, we trained a deep convolutional neural network that automatically and systematically contoured metastasized lymph nodes on DWI of head and neck cancer (HNC) patients. Material and Methods DWI images from 48 HNC patients with a total of 65 metastasized lymph nodes were obtained on a diagnostic 1.5T MR-scanner throughout radiotherapy (RT) treatment (18 with induction chemotherapy; 30 RT-only). A radiation oncologist delineated these lymph nodes on the b=50 s/mm 2 diffusion-weighted source images. To assess the inter-observer variability, a second radiation oncologist also contoured the lymph nodes for five of the images. We implemented a 3D U-net in python using Keras and Tensorflow (Dice loss function; 20% dropout; batch normalisation; 64 base features; 4 pooling layers; 22 convolutional layers; learning rate: 2e-4; automated stopping criteria with latency of 20 epochs). We added a voxel-wise bias-layer [Dalca et al. IEEE-CVPR 2018] before each convolution. We envision a workflow in which a clinician can receive contours and ADC information by clicking on the lymph node. In this workflow, a bounding box (64×64×32 voxels) is placed centred at the selected voxel and used as input for the U-net. Such clicks were simulated and training was done using the resulting bounding boxed b=50 s/mm 2 DWI image and the corresponding manual contours. The performance was evaluated using 8-fold cross- validation (over patients) and calculating the Dice similarity coefficient (DSC) and absolute change in median ADC between the manual and the learnt contours. Finally, the performance of the network was tested on an independent test dataset with 4 DWI images (3 patients) including 8 involved lymph nodes obtained with a substantially different imaging protocol on a 1.5T MR- Linac.

Proffered Papers: Proffered papers 19: Auto-contouring using deep learning

OC-0343 Artificial Intelligence Based Auto-Contouring of CTV for Cervical Cancer C. Ma 1 , J. Zhou 1 , X. Xu 1 , J. Guo 1 , M. Han 2 , Y. Gao 2 , Z. Wang 3 , J. Zhou 3 1 The First Affiliated Hospital of Soochow University, Dept. of Radiotherapy & Oncology, Suzhou, China ; 2 United Imaging Intelligence, Dept. of UII SH Algo1, Shanghai, China ; 3 Shanghai United Imaging Healthcare, Dept. of RT BU APPL, Shanghai, China Purpose or Objective To validate an artificial intelligence (AI)-based auto- contouring tool of CTVs for cervical cancer, and to evaluate its potential in improving contouring accuracy and efficiency. Material and Methods 533 CT image sets were collected from patients who received external beam radiotherapy for cervical cancer from January 2013 to June 2019. The CTVs were delineated by ROs according to institutional consensus, and set as reference contours after reviewed by experts. For definitive cases, delineate the pelvic lymphatic drainage area as dCTV1, and the paraventricular area as dCTV2; for postoperative cases, delineate the pelvic lymphatic drainage area as pCTV1. An AI-based auto-contouring tool was constructed for dCTV1, dCTV2 and pCTV1 and tested, respectively. The training/validation/testing cohorts contained 157/20/23 cases for dCTV1/dCTV2, and 270/30/33 cases for pCTV1. Dice coefficient (DC) and mean surface distance (MSD) were calculated for AI contours in testing cohorts. In addition, 4 definitive and 6 postoperative cases were randomly selected from the testing cohort and manually contoured by 3 groups of ROs: junior, intermediate, and senior, with 3 ROs per group. DC were calculated for ROs’ manual contours of each group and compared against those of AI. One month later, the junior RO group re-contoured the same 10 cases by modifying AI contours (AI-assisted junior contours) to evaluate whether AI-assistance improves accuracy. Manual contouring time and AI-assisted contouring time (including both algorithm and modification time) were recorded for efficiency comparison. Results For dCTV1, dCTV2 and pCTV1, DC of AI contours in the testing cohort were 0.88±0.03, 0.70±0.09, 0.86±0.03, respectively; and MSD were 1.32±0.48 mm, 2.42±1.62 mm, 1.15±0.38 mm, respectively. Comparison of DC between AI and ROs’ manual contours of each group showed there were no significant difference between AI and all RO groups in dCTV1, and AI outperformed the junior group for dCTV2 (median DC: 0.75 vs 0.56, P <0.01) and all groups for pCTV1(median DC: 0.75 vs 0.83/0.83/0.83, P <0.05 for all). The junior group performed AI-assisted contouring for dCTV2 and pCTV1 to determine whether AI-assistance brings improvement. Median DC was improved from 0.56 to 0.72 (30.0%, P <0.05) for dCTV2 and from 0.83 to 0.86 (3.6%, P <0.05) for pCTV1. Mean contouring time reduced from 19.0 minutes to 10.1 minutes (46.7%) for dCTV2 and from 43.6 minutes to 14.7 minutes (66.2%) for pCTV1. Conclusion Our study validated an AI-based auto-contouring tool of CTVs for cervical cancer. AI-generated contours showed high accuracy for all three CTVs. The similarity between AI and reference contours was comparable to that of senior