ESTRO 2024 - Abstract Book

S4974

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

Purpose/Objective:

By combining radiotherapy and MR-imaging in one device, MR-Linacs have not only created new possibilities in terms of dose delivery including precise tumor tracking and adaptive treatment planning, but also recently in non-contrast enhanced functional lung imaging [1]. Despite all technical advances, radiation-induced pneumonitis (RP), usually diagnosed in CT-scans 2-3 months after treatment, is still a common complication of radiotherapy delivered to lung cancer patients [2]. In conventional radiotherapy, clinical parameters such as the mean lung dose (MLD) and the lung volume receiving more than 20 Gy (V20) as well as SPECT- and 4D-CT-based functional parameters have been analyzed for RP prediction [3,4]. However, no studies on similar investigations for MR-guided radiotherapy considering clinical and functional imaging parameters have been performed. In this work, we therefore aimed to show the potential of extracting functional parameters over the course of the treatment using non-contrast enhanced ventilation MRI to predict the occurrence of RP after 2-3 months. With this, identified patients at risk could receive closer monitoring to counteract the RP in the early developing stages. A total of 19 patients (11 female, 8 male) were treated at a 0.35 T MR-Linac (MRIdian, ViewRay Inc., Cleveland, Ohio) in 3-10 fractions (Fx) with stereotactic body radiotherapy (SBRT). 11 patients showed signs of RP such as ground-glass opacities and consolidations localized around the tumor in CT-scans performed after 2-3 months. Directly after each treatment Fx, each patient received an additional free-breathing 2D cine MR-scan in coronal orientation at the tumor position (240 images in about 1 min.) with an optimized balanced steady-state free precession sequence (TR/TE=2.42/1.02 ms, FOV=500×500×20 mm 3 , matrix=128×128, flip angle=70°, frame rate=3.68 images/s). Ventilation-weighted maps were generated from each of the acquired image series by an in-house pipeline. This consists of deformable registration to a fixed reference frame and segmentation of the image series, followed by filtering to remove non-ventilation contributions from the lung signal. Using the non-uniform Fourier transform, a Fourier spectrum was calculated pixel-wise for the ventilation (Vent) signal and a ventilation-weighted map was generated from the peak-value in the spectrum. The next step in the workflow was to register the Vent-maps with the target volumes and planned dose. For this, each Fx’s 3D MR-scan was first rigidly registered to the baseline planning scan. From each registered Fx-scan, a slice corresponding to the 2D image series was determined. Deformation fields extracted from the deformable registration between the image series and the selected slice were then applied to the Vent-maps for each Fx. Three different regions-of-interest were selected: The planning target volume (PTV), the V20 without the PTV and the whole tumor bearing lung without PTV (TLung). The mean Vent-map value was calculated in each region and normalized to the mean value in the non-tumor bearing lung. The relative difference of the mean Vent-map value of the last Fx with respect to its value in the first Fx was considered as metric and calculated for each region. In addition to the three functional parameters, two common clinical parameters (V20 and MLD) were included in a univariate analysis based on Receiver Operating Characteristic Curve (ROC) analysis using radiological findings indicating RP in the CT-scan 2-3 months after treatment as endpoint. Bootstrapping with 5000 samples was applied to calculate the median Area Under the Curve (AUC) and the 95% confidence interval. To probe a potential significant difference between the RP and non-RP group, the non-parametric Mann-Whitney U test (α=0.05) was used for the different parameters. Material/Methods:

Results: