ESTRO 2024 - Abstract Book

S3822

Physics - Image acquisition and processing

ESTRO 2024

currently under intense investigation, due to their potential to improve range prediction accuracy (1,2). This study aims to characterize and compare the performance of clinical DECT and PCCT scanners against a proton CT (pCT) prototype for proton relative stopping power (RSP) estimation.

Material/Methods:

For the characterization of the imaging performances, a custom plastic phantom was imaged with the three scanners. The phantom, filled with water, contains five different cylindrical inserts: four made of homogeneous plastic materials ([0.99, 2.18] gm/cm3), and one heterogeneous wood insert (3). A PCCT and a DECT scan session were conducted at a NAEOTOM Alpha scanner and a SOMATOM Force scanner (both by Siemens Healthineers, Forchheim, Germany), respectively. For the PCCT measurements, the phantom was scanned at a tube voltage of 120 kVp and a normal-resolution mode. Virtual monoenergetic images were reconstructed at energies ranging from 40 to 190 keV in 10 keV increments. For DECT, the phantom was scanned at tube voltages of 90 kVp and 150 kVp with Sn filter. Additionally, tissue-equivalent inserts (Gammex 467) placed in a custom plastic phantom were scanned for calibration. For all xCT scans, a dose of 20 mGy CTDIvol was used. CT images were reconstructed by the scanners with the Qr36 kernel and a voxel size of (0.5×0.5×1) mm 3 , and subsequently converted into RSP voxel-by-voxel relying on calibration with the Gammex inserts based on a pair of low- and high-energy xCT images (4). The prototype pCT scanner’s acquisitions were performed at the experimental beam line of a clinical proton therapy facility at the nominal proton kinetic energy of 211.2 MeV (5,6). A distance-driven filtered back-projection algorithm (7), taking into account the single proton most likely path and the residual energy, was applied to reconstruct the phantom 3D RSP map. A Hann filter was used instead of the conventional rectangular window at Nyquist’s frequency. The 5×20 cm 2 FOV is reconstructed with a voxel size of (0.5×0.5×1) mm 3 . The dose estimation for the entire acquisition is about 11.6 mGy. Analyzing all the CT data, we compared the spatial resolution level at 10% of the modulation transfer function (MTF). Additionally, the noise level was investigated in homogeneous water slices. Finally, for evaluation of RSP accuracy, small rods of each insert were measured with a clinical multi-layer ionization chamber (MLIC, Giraffe, IBA) at clinical proton beam energies (3), to assess the reference RSP value of each insert. The mean value of the spatial resolution obtained for each insert is reported in Tab 1 for each scanner. With a Hann filter equal to 0.90, the pCT images show spatial resolution levels comparable with the xCT scanners. At the same dose level, the PCCT scanner results in noise level lower than DECT (Tab 1). The comparison with pCT should be considered as preliminary, due to the different dose levels and different reconstruction algorithms. As shown in Fig 1, the pCT scanner provided the most accurate RSP estimation, with values of accuracy below 1% and resulting in a mean absolute percentage error (MAPE) equal to (0.22 土 0.07)% (3). At the same time, the xCT scanners have higher MAPE values: (1.20 土 0.07)% and (1.67 土 0.07)% for DECT and PCCT, respectively. The main discrepancy is associated with the highest density material (teflon), which is not tissue equivalent, whereas the DECT/PCCT-based RSP estimation method is dependent on the calibration with tissue-equivalent materials. Results: