Dynamic CBCT Imaging using Prior Model-Free Spatiotemporal Implicit Neural Representation (PMF-STINR)

Dynamic cone-beam computed tomography (CBCT) is desired in clinics to capture high-spatial-resolution, time-varying dynamic images, which enable accurate guidance for motion monitoring, patient setup, and adaptive planning of radiotherapy. However, dynamic CBCT reconstruction is a highly ill-posed spatiotemporal inverse problem, as each CBCT volume in the dynamic sequence is only captured by one or a few X-ray projections, due to the slow imaging acquisition speed and the fast anatomical motion (e.g., breathing). To address this challenge, we developed a machine learning-based technique (PMF-STINR) to reconstruct dynamic CBCTs. PMF-STINR employs a joint image reconstruction and registration approach to address the under-sampling challenge, enabling dynamic CBCT reconstruction from conventional 3D CBCT scans. Specifically, PMF-STINR uses spatial implicit neural representation (INR) to reconstruct a reference CBCT volume; and it applies temporal INR to represent the intra-scan dynamic motion with respect to the reference CBCT to yield dynamic CBCTs. To represent time-varying deformable motion, PMF-STINR couples the temporal INR with a B-spline-based, data-driven motion model. The spatial INR, the temporal INR, and the B-spline model of PMF-STINR are all learned on the fly during reconstruction, without using any patient-specific prior knowledge or motion sorting/binning. PMF-STINR was evaluated via digital phantom simulations and physical phantom measurements. It was also evaluated on a multi-institutional dataset of real patient cone-beam projections featuring various imaging protocols. The results showed that the "one-shot" learning-based PMF-STINR makes an accurate and robust reconstruction technique of dynamic CBCTs. It can capture regular/irregular motion with high temporal (~0.1s) resolution and sub-millimeter accuracy.