ESTRO 2024 - Abstract Book

S3873

Physics - Image acquisition and processing

ESTRO 2024

A total of 10 iterations were needed to identify the optimal network parameters: the total training time was 58 hours, while the generation time was 45 seconds for each case. Figure 1 reports an example of generated sCT image, together with the relative dose distribution: on the test patients analysed the MAE was 63.3±16.7 HU while the ME was 6.16±12.86 HU. Mean gamma passing rates for the three tolerance criteria analysed (1%/1mm,2%/2mm and 3%/3mm) were 85.8±5.4%, 94.7±5.2% and 98.0±1.8% respectively. All the evaluated DVH parameters were in accordance with 1% for PTV and OARs.

Conclusion:

The proposed GAN network can effectively generate sCT images in the brain, starting from 0.35T MRI with generation time compliant for online adaptive procedure. The image and dose level of accuracy obtained is sufficient to safely calculate complex dose distributions, also in the case of high dose per fraction. If confirmed on a larger cohort of patients, this approach can pave the way towards faster and more efficient MR-only RT workflow, removing the CT simulation by the clinical workflow.

Keywords: Synthetic CT, Artificial Intelligence, MRI

1834

Mini-Oral

Blood oxygenation level dependent (BOLD) MRI of the prostate on an MR Linac

Asher Ezekiel 1 , Martin Swinton 2 , David L. Buckley 1,3 , Ananya Choudhury 2,4 , Cynthia L. Eccles 4,5 , Peter Hoskin 2,4 , Michael Dubec 1,4 , Damien McHugh 1,4 1 The Christie NHS Foundation Trust, Christie Medical Physics and Engineering, Manchester, United Kingdom. 2 The Christie NHS Foundation Trust, Clinical Oncology, Manchester, United Kingdom. 3 University of Leeds, Biomedical Imaging, Leeds, United Kingdom. 4 The University of Manchester, Division of Cancer Sciences, Manchester, United Kingdom. 5 The Christie NHS Foundation Trust, Radiotherapy Services, Manchester, United Kingdom