X-ray Source Motion Deblurring Framework for Fast Scan in Digital Breast Tomosynthesis.

OBJECTIVE: Accelerate wide-angle digital breast tomosynthesis (DBT) by reducing X-ray source motion blur while accounting for ripple artifacts inherent to limited-angle acquisition. METHODS: We model the in-plane point-spread function (PSF) as a depth-dependent 1D kernel and perform non-blind, slice-wise post-reconstruction deblurring. The framework has two components. A High-Attenuation Artifact Reduction (HAR) module segments high-attenuation (HA) regions and suppresses their ripple artifacts. A Ripple Artifact-Considered Deblurring (RAD) module alternates analytical data-fitting with a convolutional neural network (CNN)-based regularizer. RAD takes multiple initial deblurred estimates to implicitly handle ripple artifacts arising from soft tissue and avoid secondary ringing while restoring in-plane sharpness. RESULTS: On both numerical and physical phantom data, the method visually enhances lesion visibility and preserves textures without introducing ringing artifacts, while quantitatively showing promising results across evaluation metrics. CONCLUSION: A ripple-aware deblurring pipeline enables faster wide-angle DBT by allowing higher tube speeds without compromising image quality. SIGNIFICANCE: The proposed approach offers a practical path to shorten compression time and improve clinical throughput by jointly suppressing HA-induced artifacts and source motion blur while maintaining perceptual fidelity.