MARR: Decoupling Policy, Execution, and Calibration for All-in-One Medical Image Restoration
Authors: Haobin Chen, Ao Chang, Heqin Zhu, Rundong Wang, Ting Liu, Shaohua Kevin Zhou
Organizations: School of Biomedical Engineering, Division of Life Sciences and Medicine, University of Science and Technology of China (USTC), Hefei, Anhui, 230026, China · Medical Imaging, Robotics, Analytic Computing & Learning (MIRACLE) Lab, YRD-RIGHT, Suzhou Institute for Advanced Research, USTC, Suzhou, Jiangsu, 215123, China · Biomedical Basic Research Center (BBRC) of Jiangsu, Suzhou, Jiangsu, 215123, China · Jiangsu Key Laboratory of Multimodal Digital Twin Technology, Suzhou, Jiangsu, 215123, China · State Key Laboratory of Precision and Intelligent Chemistry, Hefei, Anhui, 230026, China
All-in-one medical image restoration seeks to recover heterogeneous clinical images with a single model, but PET, CT, and MRI differ substantially in degradation statistics, anatomical contrast, and output-space bias. A fully shared network can entangle modality-specific residual errors, whereas separate modality-specific networks sacrifice the practical advantages of unified deployment. We therefore recast all-in-one restoration as a question of where limited adaptation should be placed: policy selection, feature execution, or output calibration. We propose MARR, a compact restoration framework that constrains multi-modality adaptation into degradation-aware policy routing, modality-private residual execution, and image-domain residual correction without requiring degradation labels or separate modality-specific models. The policy branch forms a routing prompt from input statistics, latent content, and modality identity, and uses it only as a control signal. Prompt-gated modality-private adapters then perform lightweight residual refinement at intermediate decoder stages, while zero-initialized modality-specific output heads calibrate the final image-domain residual without perturbing the initial shared prediction. On an all-in-one PET, CT, and MRI restoration benchmark, MARR outperforms thirteen methods re-trained under the same protocol, achieving PSNR values of 37.34 dB, 33.85 dB, and 32.09 dB on PET, CT, and MRI, respectively, and the best modality-average PSNR of 34.43 dB. The code is publicly available at https://github.com/CHB-learner/MARR.
Figures & tables
Fig. 1: Overall architecture of MARR. The framework separates all-in-one restoration into three roles: degradation-aware prompt policy estimation for decision making, prompt-gated modality-private adaptation for execution, and zero-initialized output calibration for correction. Solid arrows denote restoration data flow and dashed arrows denote prompt/control flow.
Fig. 2: Prompt-gated modality-private adapter. The policy prompt zp predicts a modality gate that controls a lightweight DWConv-GELU-PWConv residual branch for the current decoder feature F .
Method
PET Synthesis
CT Denoising
MRI Super-Resolution
Avg.
PSNR ↑
SSIM ↑
RMSE ↓
PSNR ↑
SSIM ↑
RMSE ↓
PSNR ↑
SSIM ↑
RMSE ↓
PSNR ↑
SSIM ↑
RMSE ↓
NAFNet
35.15 *
0.9259 *
0.1105 *
33.39 *
0.9164 *
8.7455 *
29.83 *
0.9098 *
37.3700 *
32.79 *
0.9173 *
15.4086 *
Restormer
35.36 *
0.9276 *
0.1080 *
33.54 *
0.9177 *
8.5975 *
30.09 *
0.9136 *
36.3138 *
33.00 *
0.9197 *
15.0064 *
SwinIR
35.89 *
0.9318 *
0.1009 *
33.47 *
0.9161 *
8.6713 *
29.85 *
0.9107 *
37.1287 *
33.07 *
0.9195 *
15.3003 *
DaPT
35.53 *
0.9288 *
0.1051 *
33.45 *
0.9161 *
8.6889 *
30.08 *
0.9134 *
36.3602 *
33.02 *
0.9194 *
15.0514 *
RAT
35.46 *
0.9284 *
0.1071 *
33.52 *
0.9174 *
8.6280 *
29.92 *
0.9129 *
36.9398 *
32.96 *
0.9196 *
15.2250 *
TABLE I: All-in-One medical image restoration results. Avg. denotes the arithmetic mean over PET, CT, and MRI.
Group
Variant
Params (M)
PET Synthesis
CT Denoising
MRI Super-Resolution
PSNR ↑
SSIM ↑
RMSE ↓
PSNR ↑
SSIM ↑
RMSE ↓
PSNR ↑
SSIM ↑
RMSE ↓
Core modules
Output calibration only
39.67
37.2644
0.9482
0.0857
33.7453
0.9185
8.4099
32.0037
0.9392
29.1782
Private adapter only
39.75
37.2558
0.9481
0.0858
33.7476
0.9185
8.4080
32.0156
0.9394
29.1335
MARR
39.75
37.3400
0.9485
0.0849
33.8500
0.9199
8.3469
32.0900
0.9399
29.0116
Extended designs
+ frequency adapter
39.81
37.2405
0.9481
0.0861
33.7436
0.9184
8.4114
32.0075
0.9393
29.1621
+ frequency prompt pool
39.70
37.2498
0.9481
0.0859
33.7406
0.9185
8.4140
32.0167
0.9394
29.1392
TABLE II: Ablation study results of MARR. Bold indicates the best result in each metric column.
Fig. 3: Visual comparisons on all-in-one medical image restoration. Each row shows LQ input, representative comparison methods, MARR, and HQ target with matched zoom-in regions.
Fig. 4: Mechanism diagnostics of MARR. (A) Policy prompt embeddings visualized by t-SNE form modality-aware clusters. (B) Adapter residual norms show stronger level-2 feature correction for CT and MRI. (C) Output calibration shifts are largest for PET, indicating complementary image-domain residual correction.
All-in-One medical image restoration (MedIR) aims to address diverse tasks across modalities and degradation types using a single universal model. Existing methods typically prioritize modeling inter-task heterogeneity (e.g., distinct data distributions and degradation types). However, they largely neglect the inherent homogeneity present in medical images, such as widely shared anatomical structures within and across modalities, which can be leveraged to ease model training and improve generalization. To this end, we propose UniH3, a novel framework that Unifies Hierarchical Homogeneity and Heterogeneity for all-in-one medical image restoration. Specifically, to comprehensively exploit homogeneity, we introduce a Hierarchical Homogeneity Memory (H2M) module that progressively distills intra- and inter-task homogeneity priors from high-quality images during training, and adaptively retrieves the most relevant priors tailored to the input for guided restoration. These retrieved priors are then injected into the restoration pipeline via an efficient Homogeneity-Guided Attention (HGA) mechanism. Furthermore, to comprehensively address heterogeneity, we design a Hierarchical Heterogeneity Balancer (H2B) that mitigates both inter- and intra-task conflicts during optimization, facilitating balanced and effective multi-task learning. Extensive experiments on two large-scale benchmarks, MedIR-2D-500K and MedIR-3D-3K, demonstrate that UniH3 achieves state-of-the-art performance on both all-in-one and single-task medical image restoration. We hope this work establishes a strong benchmark and advances the development of general-purpose medical image restoration models. Code is available at https://github.com/Yaziwel/UniH3.
Zhiwen Yang, Jiayin Li, Chengyu Liu +3
School of Biological Science and Medical Engineering, Beihang University, Beijing 100191, China · Department of Biomedical Engineering, Tsinghua University, Beijing 100084, China · Independent Researcher
All-in-one image restoration aims to handle diverse degradations within a unified framework. Existing methods commonly encode heterogeneous degradation conditions in a shared latent space, where degradation-related cues and scene content can remain entangled. We characterize the resulting challenge as dual ambiguity: semantic ambiguity in channel-wise modulation and spatial ambiguity in restoration responses, which can lead to content corruption and residual artifacts. To mitigate this issue, we propose DAR-Net, a Dual-Ambiguity Rectification Network for all-in-one image restoration. DAR-Net first introduces a Degradation Archetype Representation (DAR) module to construct a structured degradation state through simplex-constrained archetype mixture modeling. Based on this state, a Semantic Ambiguity Rectification (SeAR) module generates degradation-aware prompts to improve channel-wise conditioning in the decoder. A Spatial Ambiguity Rectification (SpAR) module further regularizes degradation-aware and complementary features toward orthogonal response subspaces, reducing spatial interference between removal and preservation cues. Extensive experiments on standard all-in-one restoration benchmarks show that DAR-Net achieves the best overall performance under both three-degradation and five-degradation settings, improving the average PSNR over the strongest competitor by 0.14 dB and 0.34 dB, respectively; it additionally shows superior performance on CDD-11 and WeatherBench.
Cencen Liu, Wen Yin, Dongyang Zhang +6
University of Electronic Science and Technology of China · University of Electronic Science and Technology of China, Chengdu, Sichuan, China · Jiigan Technology +1
All-in-one image restoration is a unified low-level vision task that aims to effectively recover high-quality images from inputs degraded by various types and levels of corruption using a single model. Recent works have achieved remarkable progress by learning degradation-adaptive prompts or network architectures. However, these methods typically apply a uniform restoration strategy across the entire image, neglecting the fact that different regions may suffer from distinct degradation types and varying degrees of severity. In contrast, we propose to perform restoration at the pixel level, thereby enabling more fine-grained and precise control over the restoration process. Specifically, we present MGN-AIR, a novel pixel-level restoration framework for all-in-one image restoration. Our approach first learns to estimate a pixel-level visual prompt. Then, it leverages both textual and visual prompts to provide global and local degradation cues, guiding the model on where to look and how to restore at each pixel. We conduct extensive experiments on multiple all-in-one image restoration benchmarks, covering a wide range of tasks including denoising, deraining, deblurring, dehazing, desnowing, and low-light enhancement. Experimental results demonstrate that our proposed method consistently and significantly outperforms existing approaches.