Yufang Cai

Yufang Cai

Auto-generated profile, based on publicly available information

Yufang Cai

Ministry of Education, Key Lab of Optoelectronic Technology and Systems, Chongqing University, Chongqing, China

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
2 October 2025 in IEEE Transactions on Computational Imaging
Verified
Journal Article
Closed access
QWQi Wang
Qi Wang
YCYufang Cai
Yufang Cai
HYHaijun Yu
Haijun Yu
FLFenglin Liu
Fenglin Liu
WWWeiwen Wu
Weiwen Wu
Limited data computed tomography (LDCT) plays a critical role in accelerating the scanning process and reducing radiation exposure for patients. However, LDCT reconstruction is inherently an ill-posed inverse problem, often resulting in pronounced edge artifacts and the loss of fine structural details. In recent years, score-based generative models (SGMs) have shown great promise in LDCT reconstruction by alleviating the ill-posedness and enabling high-fidelity image recovery in the case of noise-free condition. However, in practical CT systems, measurement data is often contaminated by noise. The coexistence of noise and limited data presents significant challenges for SGM-based image reconstruction methods. To address this challenge, this study proposes a Model-Informed Stable Diffusion (MISD) model which integrates a sampling process with a generative prior in the image-space module and a physics prior in the projection-space module. The projection-space module incorporates physical information to establish a noise suppression mechanism, effectively reducing the impact of noise. At the same time, the image-space module uses a generative model to progressively reconstruct clear structures and features from an initial state characterized by pure noise. Together, these two modules form a cohesive mathematical framework, utilizing iterative optimization to gradually minimize the effects of noise and artifacts. Experimental results show that the MISD method consistently achieves higher quantitative metrics and recovers finer structural details than other state-of-the-art reconstruction techniques, both on simulated and real datasets.

17 March 2025 in Journal of X-Ray Science and Technology
Verified
Journal Article
Closed access
YXYarui Xi
Yarui Xi
YCYufang Cai
Yufang Cai
GZGuorong Zhu
Guorong Zhu
HYHaijun Yu
Haijun Yu
WYWei Yuan
Wei Yuan
FLFenglin Liu
Fenglin Liu
Background: Orthogonal translation computed laminography (OTCL) has great potential for tiny fault detection in laminated structure thin-plate parts. It offers a larger magnification ratio but generates limited projection data, which would result in aliasing artifacts in the reconstructed image. Objective: One way to minimize these artifacts is to use prior information, such as the piecewise constant property and prior image information. This work was inspired by the adaptive-weighted high order total variation (awHOTV) model, which is known for its ability to protect edge and detail information. Meanwhile, the laminated structure thin-plate parts are printed using computer-aided design (CAD) images, which provide structural information. Methods: To create a reliable CAD information beforehand, we adopted a two-in-one estimation method. Therefore, combining the CAD information with the awHOTV model, we propose an improved adaptive weighted higher-order TV (Cpi-awHOTV) model based on the CAD prior and use the adaptive steepest descent projection onto convex set (ASD-POCS) algorithm to solve the imaging model. Results: To evaluate the performance of our algorithm, we compared it with existing filtered back projection (FBP), simultaneous algebraic reconstruction technique (SART), total variation (TV), adaptive-weighted TV (awTV), and high order TV (HOTV)algorithms on phantom1 and phantom2 with various scanning angle ranges. Additionally, we used the phantom2 as the CAD prior in real data experiments. The results show that, the Cpi-awHOTV algorithm can obtain high-quality reconstructed images and better quantitative evaluation indicators. Conclusions: Visual inspection and quantitative analysis of reconstructed images demonstrate that the Cpi-awHOTV algorithm effectively protects edge information, and reduces aliasing artifacts due to interference from adjacent slice structures.

1 September 2025 in Pattern Recognition
Verified
Journal Article
Closed access
ZLZexin Liu
Zexin Liu
QLQi Li
Qi Li
JWJunyao Wang
Junyao Wang
TDTingyuan Deng
Tingyuan Deng
RZRifeng Zhou
Rifeng Zhou
FLFenglin Liu
Fenglin Liu

1 September 2026 in IEEE Transactions on Radiation and Plasma Medical Sciences
Verified
Journal Article
Closed access
QWQi Wang
Qi Wang
WWWeiwen Wu
Weiwen Wu
FLFenglin Liu
Fenglin Liu
YCYufang Cai
Yufang Cai
YLYueh Z Lee
Yueh Z Lee
Ge WangGe Wang
Ge Wang
Stationary head CT (sHCT) plays a critical role in detecting intracranial lesions. However, its fixed gantry restricts projection angles and undermines reconstruction accuracy. Recent studies have shown that score-based generative models (SGMs) can restore high-quality images from limited projections in two-dimensional (2D) limited-angle CT. Extending SGMs to three-dimensional (3D) sHCT is still challenging, as achieving high-resolution images without incurring excessive computational cost remains an open problem. To address this challenge, this study proposes a mutually orthogonal plane-based optical-flow-regulated (MOPO) diffusion model that effectively balances sampling efficiency and reconstruction accuracy. Specifically, we establish a 3D artifact-aware framework that incorporates a cross-plane attention mechanism to explicitly model and mitigate inter-slice artifact propagation. Furthermore, we design a mutually orthogonal plane-based (MOP) accelerated sampler capable of reducing the sampling process to as few as 10–20 steps. This approach achieves up to a 160× speedup while maintaining high reconstruction accuracy. To ensure global consistency, an anisotropic optical flow regularization is introduced, constructing non-rigid displacement fields between adjacent slices for structural coherence. Experimental results demonstrate that the MOPO-based framework outperforms existing methods in artifact suppression, detail preservation, and computational efficiency, providing a novel solution for high-fidelity and accelerated sHCT reconstruction.