Simulation of low-dose cone beam CT for paediatric image-guided proton beam therapy

Abstract

Purpose or Objective CBCT is used in image-guided proton beam therapy (IGPT) to reduce positioning errors and detect anatomical changes. In children, CBCT is often used sparingly due to dose concerns. In IGPT, anatomical changes are of particular importance due to potential range effects. Lower dose CBCT would allow for more frequent imaging – ensuring accurate treatment delivery while reducing overall dose burden. In this work, we present a novel offline simulation and reconstruction method for low-dose CBCT, solving for unique reconstruction problems in a commercial IGPT system, and present a preliminary visual grading analysis (VGA) study on simulated low-dose patient images. Material and Methods All CBCT images were acquired with the Varian Probeam (Varian Medical Systems, Palo Alto, USA) on-board imaging system. The signal-noise relationship was measured in Catphan projection images at varying exposures to inform the amount of non-uniform Gaussian noise to be added to original projection data to simulate lower exposures. Signal-to-noise ratio (SNR) was measured in simulated Catphan images and compared to the real acquisitions for validation. 3 different dose levels were simulated for the 5 images of patients treated for brain tumours with the ProBeam system – 100%, 30%, and 10% of the clinical standard (1.8 mGy CTDI). The developed reconstruction method corrected the projection images for exposure and beam hardening – using angle-dependent corrections to correct for shifts of the bowtie filter during gantry rotation. Filtered backprojection reconstruction was performed using The Reconstruction Toolkit (RTK) (Rit et al. 2014). VGA was performed on the reconstructed images by 3 experienced observers, grading visibility of bone, ventricle, air cavities, and external contour. Results Simulated and acquired Catphan projection images had good agreement in SNR – demonstrating adequate simulation of quantum noise due to lower mAs exposure. Figure 1 shows two patient images reconstructed at 100%, 30% and 10% of clinical exposure. Figure 2 shows results of the VGA on all patient images. Visibility of bone and external contour consistently highly rated even at the lowest exposure; air cavities scored similarly but their visibility was slighted degraded at lower doses; ventricle visibility was consistently moderate to poor, and depended on the specific patient – likely due to variable ventricle size - some were not visible even at 100% exposure. Conclusion An offline reconstruction method was developed that successfully emulated the image quality of the ProBeam system. Low dose CBCT images were effectively simulated and will enable the investigation of optimal image acquisition parameters using existing scans. Visual assessment of the generated images showed that bone and soft tissue contrast remain even in the lowest exposures simulated. Future work will use these methods to analyse registration uncertainty and quantify the dosimetric benefits of using low exposure 3D rather than 2D imaging for set up in IGPT.