A New Frontier In Longitudinal Lung Functional Ventilation Imaging Via Dynamic CBCT with Spatiotemporal Gaussian Representation
Abstract
Purpose
We investigate the feasibility of using daily setup free-breathing cone-beam CT (CBCT) as a functional imaging modality to generate longitudinal ventilation maps throughout the treatment course, with the goal of detecting emerging lung function impairment during radiotherapy and enabling safer, function-guided thoracic treatment.
Methods
Daily setup free-breathing CBCT raw projection data from 10 thoracic cancer patients were exported from a conventional linear accelerator treatment console. For each fraction, projections acquired during a 1-minute CBCT scan were retrospectively binned into 10 respiratory phases. Due to the short acquisition, the acquired data are sparsely sampled in each respiratory phase. To reconstruct phase-resolved CBCT, we developed a spatiotemporal Gaussian neural representation based on a differentiable 4D Gaussian model. Each Gaussian element was parameterized by position, covariance, rotation, and density. A Gaussian deformation network, consisting of a HexPlane encoder and a multi-head decoder, predicted phase-resolved Gaussian deformations by minimizing L1 and structural similarity index measure (SSIM) losses between forward projection of phased CBCT and corresponding measured projections. To derive daily ventilation maps for each fraction, end-inhalation and end-exhalation phases were deformably registered using a validated deformable image registration (DIR) algorithm, and voxel-wise Jacobian determinants which indicate volumetric changes were calculated and considered as a surrogate for lung ventilation function.
Results
Daily 4D-CBCT images were successfully reconstructed from routine 1-minute scans using the proposed framework, without introducing additional clinical workflow complexity or extra imaging dose. Longitudinal ventilation response maps were generated for every treatment fraction across the entire course of radiotherapy as an indicator of lung function change.
Conclusion
This study demonstrates the potential of daily CBCT to serve as a functional imaging modality without added workflow burden or patient imaging dose. Longitudinal ventilation maps can be reliably generated, enabling future applications such as early radiation pneumonitis risk assessment and pulmonary function–preserving adaptive radiotherapy.