Investigate the Feasibility of Using CBCT to Assess the Dose Validation for Spot-Scanning Proton Arc (SPArc) Therapy for Advanced Staged Lung Cancer Treatment

Abstract

<h3>Purpose/Objective(s)</h3> Spot-scanning proton arc therapy has been developed to improve the dose conformity and plan quality in varieties of disease indications, including advanced staged lung cancer treatment. Recently, the artificial Intelligent based synthetic CT method has been introduced in the proton clinic to assess the plan's robustness daily using CBCT. This study investigated the dose validation accuracy of using AI based synthetic-CT for SPArc in comparison with conventional IMPT delivery. <h3>Materials/Methods</h3> Twenty-seven advanced stage lung cancer patients' CT/CBCT data sets were used for training. A Generative-Adversarial-Network (GAN) model, including one generator with Residual-Unet (Res-GAN) architecture and one discriminator using a multi-layer convolutional neural net (CNN), were trained from scratch to predict synthetic CT from CBCT, with random 3D B-spline deformation as data augmentation. Another six patients with the same day re-plan CT (reCT) and CBCT were retrospectively selected for the dose validation comparison. Synthetic CT (synCT) was predicted from CBCT and compared to reCT in terms of Mean Absolute Error (MAE). Both IMPT 2-3 field and SPArc plans were generated on the reCT using the same robust optimization parameters (± 3mm setup and ± 3.5% range uncertainty) as well as planning objectives. Then, both plans were recalculated on the synCT from the same day CBCT. 3D Gamma Index was used to quantitatively assess the dose distribution discrepancy between the synCT and reCT for both IMPT and SPArc techniques. <h3>Results</h3> The result show that the image similarity between the synCT and reCT is very close with average MAE of 32.6 ± 7.5 HU. Overall, SPArc plans have superior 3D gamma index passing rate compared the multi-field IMPT plan in terms of 1%1mm; 2%2mm, 3%/3mm (<i>P</i> < 0.05) as shown in the table. In other words, SPArc effectively mitigated the dosimetric impact from the proton range differences or CT HU error in the synthetic CT. <h3>Conclusion</h3> Deep-learning CBCT-based synthetic CT could be used in dose validation not only for IMPT but also for SPArc even though SPArc plan has better dose conformity to the target. This finding paves the road map to the CBCT-based adaptation platform for SPArc adaptive therapy of lung cancer patient.