Abstract: |
We propose a real-time volumetric imaging framework that reconstructs cone-beam CT (CBCT) images using surface and single-angle X-ray projections to enable tumor motion tracking during image-guided radiation therapy. Optical surface imaging (OSI) captures high-frequency surface topography of the patient on the treatment couch and serves as a surrogate for intra-fractional tumor motion. However, OSI lacks internal anatomical visualization, limiting its accuracy in localizing internal tumors, where surface motion often poorly correlates with tumor motion. To address this, the proposed framework integrates high-frequency surface imaging with low-frequency single-angle X-ray projections from the on-board CBCT system, minimizing imaging dose. A patient-specific generative model—termed physics-integrated consistency-refinement denoising diffusion probabilistic model (PC-DDPM)—is developed to synthesize Optical Surface-Derived CBCT (OSD-CBCT) volumes. This model leverages prior knowledge of patient-specific anatomy and respiratory motion patterns from four-dimensional CT (4DCT) acquired during treatment planning. A geometric transformation module (GTM) extracts volumetric anatomical information from single-angle X-ray data, which, combined with OSI input, guides the DDPM through a physics-integrated cycle-consistency refinement process to produce high-quality OSD-CBCT images throughout treatment delivery. A simulation study using data from 56 lung cancer patients demonstrated that the framework generates high-fidelity volumetric reconstructions with accurate tumor localization, validated by intensity-, structure-, visual-, and clinically based evaluations. This work highlights the potential of the proposed framework to enable real-time, low-dose volumetric imaging for precise tumor tracking, advancing image-guided techniques for motion-sensitive radiation therapy and interventional procedures. © 2025 |