Non-ideal measurement computed tomography (NICT), which lowers radiation at the cost of image quality, is expanding the clinical use of CT. Although unified models have shown promise in NICT enhancement, most methods require paired data, which is an impractical demand due to inevitable organ motion. Unsupervised approaches attempt to overcome this limitation, but their assumption of homogeneous noise neglects the variability of scanning protocols, leading to poor generalization and potential model collapse. We further observe that distinct scanning protocols, which correspond to different physical imaging processes, produce discrete sub-manifolds in the feature space, contradicting these assumptions and limiting their effectiveness. To address this, we propose an Uncertainty-Guided Manifold Smoothing (UMS) framework to bridge the gaps between sub-manifolds. A classifier in UMS identifies sub-manifolds and predicts uncertainty scores, which guide the generation of diverse samples across the entire manifold. By leveraging the classifier's capability, UMS effectively fills the gaps between discrete sub-manifolds, and promotes a continuous and dense feature space. Due to the complexity of the global manifold, it's hard to directly model it. Therefore, we propose to dynamically incorporate the global- and sub-manifold-specific features. Specifically, we design a global- and sub-manifold-driven architecture guided by the classifier, which enables dynamic adaptation to subdomain variations. This dynamic mechanism improves the network's capacity to capture both shared and domain-specific features, thereby improving reconstruction performance. Extensive experiments on public datasets are conducted to validate the effectiveness of our method across different generation paradigms.
Figures & tables
Figure 1. Visualization of reconstructed results under different scanning protocols.
Figure 2. The overview of the proposed framework.
Figure 3. Generation examples of the uncertainty-guided diffusion model.
Geometrical Parameters β
Scanning Protocol
Photon Count α
View Number
Angle Range
LDCT
1.25e4
512
[0∘,360∘]
SVCT
1.25e8
60
[0∘,360∘]
LACT
1.25e8
512
[0∘,125∘]
Table 1. Scanning protocols used to simulate different NICT images.
Figure 4. Reconstruction results of different algorithms under varying sampling strategies. The display window for rows 1, 3, and 5 is [-805, 145] HU, while for other rows, it is [-265, 285] HU.
LDCT
SVCT
LACT
Average
Algorithm
PSNR
SSIM
LPIPS
NoiseSD
PSNR
SSIM
LPIPS
NoiseSD
PSNR
SSIM
LPIPS
NoiseSD
PSNR
SSIM
LPIPS
NoiseSD
CycleGAN
38.25
94.57
0.0306
0.0120
36.01
91.44
0.0444
0.0157
31.97
90.93
0.0522
0.0253
35.41
92.31
0.0424
0.0176
CUT
36.21
90.84
0.0654
0.0151
33.37
81.92
0.1196
0.0211
29.42
83.48
0.0766
0.0344
33.00
85.41
0.0543
0.0235
AttentionGAN
39.59
95.83
0.0341
0.0108
36.70
90.91
0.0647
0.0148
31.01
91.06
0.0480
0.0258
35.76
92.60
0.0489
0.0171
SwitchableCycleGAN
37.26
92.92
0.0493
0.0129
33.38
84.39
0.1057
0.0206
27.40
85.45
0.1000
0.0414
32.68
87.58
0.0850
0.0249
SRC
35.52
88.16
0.0899
0.0164
32.07
79.05
0.1556
0.0244
29.09
83.92
0.0696
0.0351
32.22
83.71
0.1050
0.0253
Table 2. Quantitative results of PSNR(dB) ↑ , SSIM(%) ↑ , LPIPS ↓ and NoiseSD ↓ for different algorithms under varying sampling strategies
Known Domain
Unkown Domain
Known Domain
Unkown Domain
SVCT
LACT
LDCT
LDCT
LACT
SVCT
Algorithm
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
CycleGAN
35.11
89.10
31.57
90.11
38.65
94.44
36.76
90.93
31.04
89.63
31.71
78.15
CUT
32.25
77.11
28.69
81.03
36.48
91.75
35.92
87.26
28.21
81.61
31.18
73.33
SRC
30.95
65.84
28.60
82.20
34.93
82.65
34.93
82.42
28.59
81.91
30.36
70.77
UNSB
33.11
83.33
28.51
83.38
36.57
94.10
35.64
91.87
27.69
82.40
32.03
84.31
Table 3. Reconstruction results of different algorithms for domain generalization
Figure 5. Reconstruction results of different algorithms on natural image dataset.
Hazy
Rainy
Average
Algorithm
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
CycleGAN
29.18
95.31
25.31
88.97
27.24
92.14
CUT
24.02
87.44
22.99
79.09
23.50
83.26
SRC
22.80
88.07
20.32
73.55
21.56
80.81
UNSB
22.70
83.33
21.31
72.65
22.00
77.99
CSUD
23.13
82.11
30.30
92.81
26.71
87.46
Table 4. Quantitative results for different algorithms on natural image dataset
AttentionGAN
SRC
UNSB
CSUD
Ours
ACC ( ↑ )
72.16%
76.13%
66.43%
61.69%
77.32%
Table 5. Quantitative results of the downstream task.
LDCT
SVCT
LACT
Average
Algorithm
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
CycleGAN
38.25
94.57
36.01
91.44
31.97
90.93
35.41
92.31
CycleGAN†
39.25
95.30
36.32
91.95
32.84
91.76
36.13
93.00
Δ
↑ 1.00
↑ 0.73
↑ 0.31
↑ 0.51
↑ 0.87
↑ 0.83
↑ 0.72
↑ 0.69
Switchable
37.26
92.92
33.38
84.39
27.40
85.45
32.68
87.58
Switchable†
38.60
94.16
34.84
88.64
27.99
86.41
33.81
89.73
Table 6. Effectiveness of the proposed uncertainty-guided manifold smoothing method
Dataset
LDCT
SVCT
LACT
Algorithm
LD
SV
LA
Syn
PSNR
SSIM
PSNR
SSIM
PSNR
SSIM
Single-LD
✓
✗
✗
✗
38.84
94.95
–
–
–
–
Single-SV
✗
✓
✗
✗
–
–
35.35
90.55
–
–
Single-LA
✗
✗
✓
✗
–
–
–
–
31.91
87.92
Multi-Vanilla
✓
✓
✓
✗
38.25
94.57
36.01
91.44
31.97
90.93
Multi-Vanilla⋆
✓
✓
✓
✗
38.51
94.19
36.22
92.11
32.51
90.36
Table 7. Ablation study on joint training and contribution parts
Figure 6. t-SNE visualization of real and synthetic data.
Generative models, particularly Diffusion Models (DM), have shown strong potential for Computed Tomography (CT) reconstruction serving as expressive priors for solving ill-posed inverse problems. However, diffusion-based reconstruction relies on Stochastic Differential Equations (SDEs) for forward diffusion and reverse denoising, where such stochasticity can interfere with repeated data consistency corrections in CT reconstruction. Since CT reconstruction is often time-critical in clinical and interventional scenarios, improving reconstruction efficiency is essential. In contrast, Flow Matching (FM) models sampling as a deterministic Ordinary Differential Equation (ODE), yielding smooth trajectories without stochastic noise injection. This deterministic formulation is naturally compatible with repeated data consistency operations. Furthermore, we observe that FM-predicted velocity fields exhibit strong correlations across adjacent steps. Motivated by this, we propose an FM-based CT reconstruction framework (FMCT) and an efficient variant (EFMCT) that reuses previously predicted velocity fields over consecutive steps to substantially reduce the number of Neural network Function Evaluations (NFEs), thereby improving inference efficiency. We provide theoretical analysis showing that the error introduced by velocity reuse is bounded when combined with data consistency operations. Extensive experiments demonstrate that FMCT/EFMCT achieve competitive reconstruction quality while significantly improving computational efficiency compared with diffusion-based methods. The codebase is open-sourced at https://github.com/EFMCT/EFMCT.
Jiayang Shi, Lincen Yang, Zhong Li +3
Centrum Wiskunde en Informatica · LIACS, Leiden University · Great Bay University +1
Learning-based medical image registration has matched the accuracy of conventional methods while offering superior computational efficiency. However, existing approaches suffer from poor generalization across diverse clinical scenarios, requiring the laborious development of multiple isolated networks for specific registration tasks, e.g., inter-/intra-subject registration or anatomical region-specific alignment, leading to cumbersome development pipelines. To overcome this limitation, we propose UniReg, the first conditional unified model for multi-scenario medical image registration, which combines the precision advantages of task-specific learning methods with the generalization of traditional optimization methods. Our key innovation is a unified registration framework that adaptively estimates deformation fields conditioned on: (1) anatomical structure priors, (2) registration type constraints (inter/intra-subject), and (3) instance-specific features, enabling effective alignment across heterogeneous CT and MR registration scenarios within a single model. Through comprehensive experiments on multiple CT/MR registration datasets, UniReg achieves superior average registration accuracy compared with current state-of-the-art learning-based methods while exhibiting strong cross-scenario generalization. Moreover, by replacing multiple isolated task-specific models with a compact unified model, UniReg substantially reduces the overall training burden in terms of total training cost and model redundancy.
Zi Li, Jianpeng Zhang, Tai Ma +7
Department of Electrical and Computer Engineering, The University of Hong Kong, Hong Kong SAR, China · DAMO Academy, Alibaba Group · The First Affiliated Hospital, Zhejiang University School of Medicine, Hangzhou, China +1
Sparse-view CT reduces radiation dose and scanning time by acquiring fewer projection views, but angular undersampling makes reconstruction severely ill-posed, causing streak artifacts, structural blurring, and loss of fine details. Existing supervised methods are often tied to specific sampling settings, whereas generative methods may introduce anatomically inconsistent hallucination-like structures under severe undersampling. We propose Lucid, a sparsity-adaptive, consistency-guided reconstruction framework based on a Flow Matching generative prior for sparse-view CT. Lucid is trained only on high-quality CT images to learn a continuous transport between a Gaussian distribution and the high-quality CT image distribution, independent of view sampling. During inference, the sampling sparsity level is explicitly incorporated to adapt the generative trajectory of a single pretrained model. Specifically, Lucid constructs a degradation-matched initial state by sparsity-weighted fusion of the sparse-view FBP image and Gaussian noise, performs sparsity-modulated Flow Matching updates, and applies projection-domain data-consistency correction after each prior update. Experiments under multiple sparse-view settings show that Lucid achieves stable reconstruction performance across different sampling densities, improves image quality and structural fidelity, and reduces the risk of hallucination-like structures in generative sparse-view CT reconstruction.
Jigang Duan, Jiayi Wang, Heran Wang +3
School of Mathematical Sciences, Capital Normal University, Beijing 100048, China · National Center for Applied Mathematics Beijing, Capital Normal University and Academy for multidisciplinary studies, Capital Normal University, Beijing, 100048, China