Truncation artifact reduction in cone-beam computed tomography scans using conditional generative diffusion models

Abstract

Cone-beam computed tomography (CBCT) is commonly used in diagnostic imaging owing to its superior spatial resolution and lower radiation dose relative to conventional computed tomography (CT). A limitation of CBCT systems, however, is their inherently restricted field of view, which induces truncation artifacts, thereby reducing image quality and diagnostic accuracy. This study implemented conditional generative diffusion models (CGDMs) to generate the truncated regions within CBCT projection data to facilitate truncation artifact reduction. This approach capitalizes on the capability of CGDMs to model intricate data distributions and produce realistic projection completions. Compared to the established reference method (RTK padding), our approach achieves a superior improvement of 12 dB in PSNR and a gain of 0.15 in SSIM, indicating significantly higher fidelity to the original projection data. Furthermore, in the evaluation of the subsequently reconstructed volumetric images, the proposed method attains an average increase of 9 dB in PSNR and 0.10 in SSIM, accompanied by visibly enhanced image uniformity and improved low-contrast visibility. CGDM also achieves statistically significant advantages over GAN-based methods (PSNR/SSIM) in ROI reconstruction while fully preserving critical, diagnostically relevant features. Conditional generative diffusion models prove effective in reconstructing the missing projection information resulting from truncation, which substantially mitigates truncation artifacts, leading to notable improvements in the reconstructed image uniformity and the discernibility of low-contrast structures.