ESTRO 2024 - Abstract Book

S3865

Physics - Image acquisition and processing

ESTRO 2024

References:

Yang, Bin, Hui Geng, Tien Yee Amy Chang, Mei Yan Tse, Wai Wang Lam, Chen-Yu Huang, Tungho Wu et al. "Clinical implementation of kVCT-guided tomotherapy with ClearRT." Physical and Engineering Sciences in Medicine 45, no. 3 (2022): 915-924.

1786

Digital Poster

A new method to quantify signal intensity changes in glioblastoma during radiotherapy on an MR-Linac

Philipp Wallimann 1 , Bertrand Pouymayou 1,2 , Michael Mayinger 1 , Sylwia Nowakowska 3 , Andreas Boss 3 , Matthias Guckenberger 1 , Stephanie Tanadini-Lang 1 , Nicolaus Andratschke 1 1 University Hospital Zurich, Department of Radiation Oncology, Zurich, Switzerland. 2 University Hospital Zurich, Department of Neuroradiology, Zurich, Switzerland. 3 University Hospital Zurich, Institute for Diagnostic and Interventional Radiology, Zurich, Switzerland

Purpose/Objective:

Magnetic Resonance Linear Accelerators (MR-Linacs) have enabled longitudinal patient imaging with enhanced soft tissue contrast during radiotherapy. This capability presents an opportunity for performing quantitative analyses of the acquired images, facilitating a comprehensive assessment of the evolving characteristics within both normal tissue and tumors during the course of radiotherapy. In this study we present a novel approach for quantifying changes in daily positioning images of glioblastoma patients undergoing fractionated radiotherapy on an MR-Linac.

Material/Methods:

The proposed method aims to identify significant longitudinal changes in MR images. It is based on a quantitative comparison of brain MR images, so we first preprocessed them as visualized in figure 1(a). We affinely registered all daily images of a patient to the baseline image using a TRSAA approach as implemented in the Advanced Normalization Tools (ANTS) library [1]. The brain was selected as the region of interest to ensure a high spatial correspondence of the normal brain tissue after registration. We performed an N4 intensity uniformity correction [2] on the images, as implemented in ANTS, and a custom intensity normalization method that normalizes the intensity of white matter to 0 and that of ventricular CSF to 100 [3]. After registration and normalization, a voxel wise comparison between the images was possible. A threshold of significance for these voxel wise comparisons was determined as shown in figure 1(b). We chose a significance level of 0.05 and corrected it for multiple testing by dividing it with the number of voxels in the brain. We then calculated the number of standard deviations that correspond to this two-sided significance level for a unit normal distribution. This value was multiplied with the magnitude of random intensity variation between images, which we determined by calculating the standard deviation of voxel wise daily intensity changes observed in the