Next-generation nonstop gated CBCT for respiratory gating lung SBRT - Stereotactic body radiation therapy (SBRT) is a highly precise, non-invasive treatment for early-stage lung cancer, offering excellent local control rates, a favorable toxicity profile, and a curative option for patients. Effective management of respiratory motion is critical due to the high dose and precision required. Respiratory gating (RG) enables patients to breathe freely during treatment by delivering radiation within a specified portion of the respiratory cycle (“gating window”), thereby reducing the planning target volume and minimizing radiation exposure to nearby organs at risk (OARs). Because pretreatment imaging verification is essential, free-breathing gated cone-beam computed tomography (gCBCT) is commonly prescribed. Compared with standard three-dimensional (3D) CBCT, gCBCT offers reduced motion artifacts and improved visualization of tumors and OARs. Free-breathing gCBCT acquisition on a C-arm linear accelerator (LINAC) is a lengthy procedure, and therefore increases patient imaging time and discomfort as well as the risk of movement or breathing pattern drift, potentially compromising treatment accuracy. The issue is exacerbated because most lung SBRT patients undergo 3, 4, 5, or 8 fractions with 1-2 gCBCT scans/fraction. For patients with multiple tumors requiring multi-isocenters, the total number of scans can double or triple, further extending imaging time while limiting clinical throughput. Meanwhile, additional kV projections are acquired during gantry accelerations in gCBCT, resulting in higher imaging dose than in 3D CBCT. There is an urgent, unmet need for a more time- and dose-efficient gCBCT technique. We propose a next-generation imaging paradigm, nonstop gated CBCT (ngCBCT), which allows continuous gantry rotation (1 min/scan) with the kV X-ray beam activated only during the predefined gating window of the respiratory cycle. By restricting X-ray exposure to the gating window, ngCBCT significantly reduces imaging dose (by > 40%). However, ngCBCT produces highly non-uniform and under-sampled projections, posing a significantly more complex reconstruction challenge than in traditional sparse-view CBCT. To our knowledge, only limited research has been conducted to address this issue. We will develop, for the first time, a powerful and efficient deep learning-based dual-domain reconstruction network (DDRN) for ngCBCT. We hypothesize that ngCBCT, paired with the DDRN framework, will overcome limitations of current clinical gCBCT by substantially reducing imaging time and imaging dose while preserving image quality. The specific aims are: 1) develop, optimize, and evaluate a DDRN for ngCBCT using emulated patient data, 2) implement ngCBCT acquisitions on the LINAC and validate DDRN using motion phantom data, and 3) conduct pilot studies to evaluate ngCBCT performance in patients. We expect that the successful completion of the above aims will enable the clinical translation of ngCBCT to improve patient experience, treatment outcome, and efficiency of RG-SBRT. This novel acquisition is also critical for expanding the adoption of RG techniques in other tumor sites affected by respiratory motion, e.g., the pancreas and liver.
