ESTRO 2024 - Abstract Book

S3961

Physics - Image acquisition and processing

ESTRO 2024

4. Shaw E, Kline R, Gillin M, Souhami L, Hirschfeld A, Dinapoli R, Martin L. Radiation Therapy Oncology Group: radiosurgery quality assurance guidelines. Int J Radiat Oncol Biol Phys. 1993 Dec 1;27(5):1231-9. doi: 10.1016/0360 3016(93)90548-a. PMID: 8262852.

5. Paddick, Ian & Lippitz, Bodo. (2006). A simple dose gradient measurement tool to complement the conformity index. Journal of neurosurgery. 105 Suppl. 194-201. 10.3171/sup.2006.105.7.194

2993

Digital Poster

Evaluation of a DL based synthetic CT algorithm for pelvis and brain MR-Only radiotherapy

Aurélien Badey 1 , Enric Jaegle 1 , Maria-Elena Alayrach Biarnes 1 , Paul Martinez 1 , Yann Lauzin 1 , Pauline Mazars 1 , Marianne Zinutti 1 , Oussama Saadi 1 , Hugo Le Brun 1 , Véronique Bodez 1 , Laure Vieillevigne 2 , Catherine Khamphan 1 1 Institut Sainte Catherine, Medical Physics, Avignon, France. 2 Oncopole Claudius Regaud, Institut Universitaire du Cancer de Toulouse (IUCT), Medical Physics, Toulouse, France

Purpose/Objective:

Magnetic Resonance (MR) images in radiotherapy are used to define contours, in particular to provide accurate spatial information on target volumes and Organs at Risk (OAR). Computed Tomography (CT) is essential for dose calculation using available electron or mass densities. The solution provided by Syngo.Via (VB60, Siemens Healthineers, Germany) offers two deep learning (DL) models (cGAN type) for pelvis and brain to create synthetic CT images (sCT) (or pseudo CT) for MR-only radiotherapy. This tool is evaluated in terms of HU accuracy and dosimetry.

Material/Methods:

Twenty patients for each cohort, brain (10 normo-fractionated and 10 stereotactic prescription schemes) and pelvis (10 prostate (60Gy/20 fractions) and 10 cervix (55Gy/45Gy/25 fractions with lymph nodes irradiation (+/- N+ boost)), were included. For each patient, anatomical sequences (3D T1 Flair and Space for brain; 3D T2 Turbo Spin Echo for pelvis) and a dosimetric sequence (3D T1 Vibe Dixon) were acquired on an MRI SOLA 1.5T (Siemens, Germany). These patients also received a CT scan. Dose calculation was carried out for the treatment using AAA v16.1 algorithm (Eclipse, VMS, USA). A comparison was made between CT (reference) and sCT images in terms of HU accuracy with Mean Error (ME) and Mean Absolute Error (MAE) metrics for body contour, soft and bone tissues separately and other organs of interest. The sCT was rigidly registered for evaluations except for ME and MAE evaluation for pelvis cohort to take anatomical variations into account (deformable registration). Dosimetric comparisons were made by computing differences observed between dose planned and recalculated on sCT for selected OAR ( Dmean ), and target volumes ( D95%, D2% ). Dose matrices comparison obtained on CT and sCT was also performed with 3D gamma index calculation ( global, 2%/2mm, threshold : 10% ).

Results: