Focal spot mitigation with dynamic sampling conditions for high-resolution X-ray CT via diffusion priors

Abstract

High-resolution imaging in clinical X-ray CT is essential for accurate diagnosis, yet its performance is constrained by focal spot dynamics, which dominate image degradation under modern detector technology and low-dose acquisition protocols. We introduce dynamic focal spot diffusion CT (DFSD-CT), a reconstruction framework that jointly estimates both the tomographic image and the stochastic, time-varying focal spot profile. By leveraging a conditional forward operator integrated with a generative diffusion prior, DFSD-CT accounts for instantaneous and view-dependent source distortions, eliminating reliance on post hoc deblurring. Experiments with varied anatomical regions and challenging sampling conditions demonstrate improved modulation transfer function (MTF) and reconstruction fidelity by up to 3 dB with respect to state-of-the-art diffusion algorithms, narrowing the resolution gap between clinical and micro-CT imaging in low-dose settings, and thus offering a pathway toward enhanced diagnostic accuracy while maintaining safe radiation levels.