ESTRO 2024 - Abstract Book

S3890

Physics - Image acquisition and processing

ESTRO 2024

radiotherapy (MRgRT) for moving target delineation. This study aims to compare the target volumes delineated on 3D T1w/T2w MR images to those on 4DCT by using a motion phantom.

Material/Methods:

All CT and MR scans were performed with a CT scanner and a 1.5T MR-Linac, respectively. A MR-conditional motion management quality assurance (QA) phantom was used to simulate different breathing patterns, namely static, sinusoidal model (cycle time of 3.5s, amplitude of 7 mm in lateral, 1.3 mm in anterior-posterior (AP), and 10 mm in superior-inferior (SI) directions ) and cos 6 (x) model (cycle time of 4s, amplitude of 7 mm in lateral, 2 mm in AP, and 20 mm in SI directions). Respiratory gating for scanner (RGSC) was used for 4DCT reconstructions. T1-weighted FFE sequence (TE: 4.5 ms and TR:7.6 ms) and T2-weighted TSE sequence (TE:206 ms and TR: 2100 ms) were selected. 4DCT and MR scans were repeated three times and five times, respectively. The 4DCT data were sorted into 10 respiratory phase bins. The phantom's gross tumor volume (GTV) in each phase was manually delineated. The internal target volume (ITV) per each 4DCT scan ITV allphase was the union of all the phase GTVs. Ground truth ITV GT was generated from ITV allphase contoured from three sets of 4DCT per motion pattern using a Simultaneous Truth And Performance Level Estimation (STAPLE) algorithm. ITV MIP was also contoured on the maximum intensity projection (MIP) CT images for comparison. ITV T1 /ITV T2 was delineated based on the hyper-/hypo-intensity signal of phantom target in T1w/T2w MR images. Structures were compared to ground truth structures using the Dice similarity coefficient (DSC), Hausdorff distance (HD), Jaccard score (JS), and Mean distance to agreement (MDA). Figure 1 and 2 show the contoured structures on CT/T1w/T2w images with static and moving phantom, respectively. With the sinusoidal motion, the mean GTV volume of 4DCT phases (total 30 for each motion) was 19.42 ±0.81 cc. The mean GTV of static CT was 22.01 cc, larger than static T1w (21.85 cc) and static T2w (20.06 cc) images by 0.16 and 1.95 cc, respectively. The average volume of ITV allphase determined by 4DCTs was 33.00 ±0.95 cc, indicating a satisfying reproducibility of 4DCT. The derived ground truth ITV GT (33.55 cc) was 0.56 cc (1.70%) larger than ITV MIP (32.99 ±0.88 cc). The volume of ITV T1 (26.73 cc) and ITV T2 (28.92) decreased by 20.33% and 13.80%, respectively, compared with ITV GT . Table 1 lists the calculated DSC, JS, HD and MDA between contoured structures. In comparison with ITV GT , ITV MIP achieved the highest mean DSC of 0.97 and JS of 0.94, and the minimum HD of 2.04 mm and MDA of 0.26 mm. ITV T1 and ITV T2 also obtained high DSC (>0.89) and JS (>0.80), and moderate HD (<4.82 mm) and MDA (<0.82). ITV T1 had comparable DSC, JS, HD, and MDA with ITV T2 . Results:

Table 1: Mean and standard deviation of calculated DSC, JS, HD and MDA between structures.