ESTRO 2023 - Abstract Book

S1399

Digital Posters

ESTRO 2023

Scientific Center named after Loginov MHD, Radiation methods of diagnostics and treatment, Moscow, Russian Federation; 3 M.V. Lomonosov Moscow State University, Accelerator’s physics and radiation medicine, Moscow, Russian Federation; 4 Federal State Autonomous Institution «N. N. Burdenko National Medical Research Center of Neurosurgery» of the Ministry of Health of the Russian Federation,, Radiotherapy and radiosurgery, Moscow, Russian Federation Purpose or Objective The significance of diffusion magnetic resonance imaging (dMRI) in planning of intracranial pathology radiotherapy (RT) is growing. dMRI are generally acquired with echo planar imaging (EPI) that is prone to spatial distortions related to off resonance effects. It is important to correct these distortions to provide a representation of the anatomy as accurate as possible. dMRI requires post-processing, in which the description of diffusion in a voxel is a non-trivial task. The objective of this study is development of algorithms for pre- and post-processing, analysis and visualization of dMRI for planning of intracranial targets RT. Materials and Methods 15 patients (12 with glioblastoma and 3 with Parkinson’s disease) were investigated. The protocol included: T1-, T2 weighted images, FLAIR, T1 with contrast enhancement. Diffusion tensor imaging (DTI) was performed on a 3T GE Signa HDxt scanner using an SE EPI pulse sequence: TR/TE, ms - 15350/88.9, = 0, 1000 mm2/s, matrix - 256 × 256; slice thickness/gap, mm — 2.5/2.5; field of view, cm/voxel , mm2 — 24/0.9375 × 0.9375; directions of diffusion gradient /number of excitations – 33/2. Proposed dMRI processing consists of: distortion correction by deformable image registration with anatomical series based on autocorrelation of local structure; mathematical correction of the echo intensity variation due to the spatial RF-field inhomogeneity based on the Markov Random Field modeling and optimization by iterated conditional modes (ICM); brain segmentation; multimodal tumor clustering using the k-means algorithm; probabilistic fiber tracking by the Hough transform. Results The software allows to split the image into separate b-value and extrapolate to any user-specified value. "Brain masking" of dMRI was developed using a deformable surface model. Within the framework of proposed distortion correction algorithm maximum voxel dimension was 1,03 ± 0,12 mm. The tools make it possible to build quantitative maps of diffusion parameters, analyze brain regions, and perform multimodal target clustering (median dice score = 0.82 ± 0,16). Nonparametric distribution of the observed MR echo intensities and neighborhood tissue correlations is used to correct dMRI signal inhomogeneities and segment MR-volume on clusters: white matter (WM), grey matter (GM) and cerebrospinal fluid (CSF). Brain white matter tracts were visualized by probabilistic algorithm using the Hough transform. The tractogram was filtered to discard streamlines that terminate inside the WM or that enter the CSF. In the context of functional radiosurgery, it is feasible to reduce the volume of the internal capsule receiving 12 Gy by 40%. Conclusion The proposed algorithms for dMRI processing can improve the efficiency and accuracy of diagnosis and RT of brain pathology. Segmentation data make it possible to reduce the clinical target volume, minimizing the dose on the normal tissues. White matter fiber tractography by Hough transform allows to effectively optimize the dose to organs at risk. 1 University College London, Medical Physics and Biomedical Engineering, London, United Kingdom; 2 University College London Hospitals NHS Foundation Trust, Radiotherapy, London, United Kingdom; 3 University College London Hospitals NHS Foundation Trust,, Radiotherapy, London, United Kingdom Purpose or Objective Synthetic CTs (synCTs) that match the anatomy of daily CBCTs have important applications in adaptive radiotherapy workflows. It is challenging to validate and compare the quality of synCT images generated with different popular methodologies, such as deformable image registration (DIR) or artificial intelligence (AI). In this work we propose an automated tissue segmentation method to assess the structural consistency between matched CBCT and synCTs. Materials and Methods We propose to evaluate the structural similarity between matched synCTs and CBCTs using segmentations of key tissue types within the body: soft tissue, skeleton, gastrointestinal (GI) air, and lungs. We automated the segmentation of these tissues on CT/CBCT using an AI-based method. CT/CBCT scans and semi-automatic (ground-truth) segmentations of all tissue types from 63 patients aged 2 to 22 historically irradiated in the thoracic-abdominal-pelvic region were used. We trained a patch-based 3D UNet (96 × 96 × 48) with 4 downsampling/upsampling stages (16, 32, 64, 128, 256 channels at each of the levels respectively) with two residual blocs at each resolution level. The training dataset (N=50) consisted of CBCT/CTs aiming for the network to perform well on both CBCT and CT-like images. The network was then used to segment an independent dataset (N=13): planning CT, daily CBCTs and two synCT images matched to each daily CBCT (Fig 1). One synCT was generated by deformably registering the CT to CBCT (synCT DIR); the other was generated using an in-house cycleGAN network that converted CBCT to CT (synCT AI). The structural similarity was evaluated using Dice Similarity Coefficient (DSC), the Hausdorff Distance (HD) and mean pixel intensity (HU). PO-1689 An automated strategy to evaluate the structure consistency of CBCT-to-CT synthesis A. Szmul 1 , S. Taylor 1 , I. Moreira 1 , P. Lim 2 , J. Cantwell 2 , D. D'Souza 2 , S. Moinuddin 2 , M. Gaze 2 , J. Gains 3 , C. Veiga 1