ESTRO 2021 Abstract Book

S1376

ESTRO 2021

Conclusion Segmentations provided by the 2D and 3D U-net gave good overlap (median DSC 0.74-0.79) with manual delineations, with DSCs similar to those reported in interobserver agreement studies. The 3D models did not improve segmentations relative to the more time and memory-efficient 2D models. Increasing 2D model complexity did not increase performance, whereas including higher variance artificial data did. Thus, a simple 2D U-net model with augmented data appears to be sufficient. PO-1656 Quantitative evaluation of DWI sequence in a low T magnetic resonance guided radiotherapy system L. Placidi 1 , M. Nardini 2 , D. Cusumano 2 , L. Boldrini 2 , G. Chiloiro 2 , V. Valentini 2 , L. Indovina 2 1 Fondazione Policlinico Universitario A. Gemelli IRCCS, Radiation Oncology , Rome, Italy; 2 Fondazione Policlinico Universitario A. Gemelli IRCCS, Radiation Oncology, Rome, Italy Purpose or Objective Magnetic Resonance guided Radiotherapy (MRgRT) offers the possibility to monitor the tumour and OARs positions during the radiotherapy treatment (RT). Additionally, it also allows to online adapt the RT treatment plan based on the daily anatomy, optimizing the dose distribution while the patient is still on couch. Nevertheless, an important improvement would be to identify a specific biological state by quantitative imaging biomarkers. Apparent diffusion coefficient (ADC) map is a measure of the magnitude of diffusion within tissue and is commonly clinically calculated using MRI with diffusion weighted imaging (DWI). The aim of this study is to quantitatively evaluate the quality of DWI on a low T (0.35T) MRgRT system to identify the optimal setting in terms of image quality and time elapse. Materials and Methods Two different phantoms have been employed in the study. The first is a spherical homogeneous phantom (SHP) filled with NiSO 4 solution. The second phantom is a home-made cylindrical phantom (HMCP) with inserts containing five different concentrations of Metylcellulose (1%, 5%, 10%, 20% and 30%). For both phantoms, DWI images have been acquired considering different b-values and averages number, only in the longitudinal diffusion direction. ADC maps have been consequently computed and imaging signal-to-noise ratio (SNR), uniformity, stability and repeatability have been evaluated considering the different sequence parameters. The obtained ADC results have been also compared with the refence value obtained in a 1.5 T diagnostic MR system. Results As regards to the SHP, homogeneity higher than 85% has been achieved by all the tested b-values for a number of averages higher than 10, as depicted in figure 1.