ESTRO 2022 - Abstract Book

S284

Abstract book

ESTRO 2022

L. O'Connor 1,2 , J.H. Choi 3,1 , J. Dowling 4 , H. Warren-Forward 2 , J. Martin 5,6 , P. Greer 1,3

1 Calvary Mater Hospital, Radiation Oncology, Newcastle, Australia; 2 University of Newcastle, School Health Sciences, Newcastle, Australia; 3 University of Newcastle, School of Mathematical and Physical Sciences, Newcastle, Australia; 4 Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australian E-Health Research Centre, Herston, Australia; 5 Calvary Mater Newcastle, Radiation Oncology, Newcastle, Australia; 6 University of Newcastle, School of Medicine and Public Health, Newcastle, Australia Purpose or Objective There are several means of synthetic computed tomography (sCT) generation for MRI-only planning in the literature. The aim of this research was to apply four of the most popular methods of sCT creation to facilitate MRI only radiation therapy treatment planning for male and female rectum, anal canal, cervix and endometrium neoplasms. The sCT methods were validated against conventional CT, with regards to Hounsfield unit (HU) estimation and plan dosimetry. Materials and Methods Paired MRI and CT scans of forty patients, treated for a range of pelvic malignancies were used for sCT generation and validation. Bulk density assignment, tissue class segmentation, hybrid atlas and deep learning sCT generation methods were applied to all 40 patients. Each sCT creation method used was based on successfully applied methods in the literature. Dosimetric accuracy was assessed by dose difference at reference point, dose volume histogram (DVH) parameters and 3D gamma dose comparison. Due to the non-parametric nature of the data, statistical significance was determined using a Mann-Whitney U-test with a significance level of 0.05. Hounsfield unit estimation was assessed by mean error and mean absolute error in HU value between each sCT and CT. Results There was no statistically significant dose difference to CT at the ICRU reference point for any of the sCT methods (Table 1). The median percentage dose difference between the CT and sCT was <1.0% for all sCT methods, with the lowest median percentage dose difference for the deep learning method at -0.03% (Interquartile Range (IQR) 0.13, -0.31) and the largest difference for the bulk density assignment at -0.73% (IQR -0.10, -1.01). The median DVH dose difference for all structures and parameters combined was less than 0.5% for all sCT methods (Figure 1). Table 1 ICRU median percentage dose difference and median DVH dose difference by sCT method

Figure 1 Percentage DVH dose difference by structure (each structure parameters combined) for each synthetic CT method. The mean 3D gamma dose agreement at 3%/2mm amongst all sCT methods was 99.8%. The highest agreement at 1%/1mm was 97.3% for the deep learning method and lowest was 93.6% for the bulk density method. The deep learning and hybrid atlas techniques gave the lowest difference to CT in mean error and mean absolute error in HU estimation for the whole body, bone and soft tissue estimations, reflecting the dosimetric results. Conclusion Bulk density assignment, tissue class segmentation, hybrid-atlas and deep learning methods of sCT generation all result in similarly high dosimetric agreement for MRI only planning of male and female cancers of the rectum, anal canal, cervix and endometrium. Choice of sCT generation technique can be guided by department resources available and image guidance considerations, with minimal impact on dosimetric accuracy

PD-0322 Artificial intelligence organ-at-risk dose prediction for high-risk prostate cancer IMRT