Transferable Optimization Network for Cross-Domain Image Reconstruction

Yunmei Chen; Chi Ding; Xiaojing Ye

cs.CVMar 8, 2026

Transferable Optimization Network for Cross-Domain Image Reconstruction

Authors: Yunmei Chen, Chi Ding, Xiaojing Ye

Organizations: Department of Mathematics, University of Florida, Gainesville, FL 32611, USA. · Department of Mathematics and Statistics, Georgia State University, Atlanta, GA 30303, USA.

Abstract

We develop a novel transfer learning framework to tackle the challenge of limited training data in image reconstruction problems. The proposed framework consists of two training steps, both of which are formed as bi-level optimizations. In the first step, we train a powerful universal feature-extractor that is capable of learning important knowledge from large, heterogeneous data sets in various domains. In the second step, we train a task-specific domain-adapter for a new target domain or task with only a limited amount of data available for training. Then the composition of the adapter and the universal feature-extractor effectively explores feature which serve as an important component of image regularization for the new domains, and this leads to high-quality reconstruction despite the data limitation issue. We apply this framework to reconstruct under-sampled MR images with limited data by using a collection of diverse data samples from different domains, such as images of other anatomies, measurements of various sampling ratios, and even different image modalities, including natural images. Experimental results demonstrate a promising transfer learning capability of the proposed method.

Figures & tables

Panel of Figure 1 — Figure 1: The flowchart (top) of the $t$ th iteration of Algorithm 1 , the $\mathbf{u}$ -module (middle), and the $\mathbf{v}$ -module (bottom). Each round node represents a variable. Each rectangular node represents a module, mapping, or condition check. The identity mapping is denoted by Id.

Method	PSNR (dB)				SSIM (%)
Method	Brain	Knee	$\text{Cardiac}^{\dagger}$	$\text{Prostate}^{\dagger}$	Brain	Knee	$\text{Cardiac}^{\dagger}$	$\text{Prostate}^{\dagger}$
DnCn Schlemper et al. (2018)	36.02	32.17	33.91	31.32	94.36	80.50	91.66	85.59
LDA Chen et al. (2021)	37.38	32.38	34.78	32.92	95.42	80.63	93.16	89.10
UNet Arshad et al. (2021)	34.05	31.78	32.05	30.74	92.33	79.72	88.50	84.49
U-MRI Liu et al. (2021)	36.40	32.30	34.85	31.69	94.67	80.67	92.88	86.88
Meta Bian et al. (2021)	36.32	32.11	35.61	32.23	94.67	79.81	93.95	87.77

Table 1: Average PSNR and SSIM of reconstructed images on four different anatomies in Section 5.2.1 . The anatomies marked by

\dagger

(cardiac and prostate) have small training datasets. A fixed Cartesian mask of 20% sampling ratio is used to undersample the

k

-spaces of all images.

Panel of Figure 2 — Figure 2: Comparison of cross-anatomy TL reconstruction results using 20% Cartesian sampling ratio in Section 5.2.1 . Top two rows show two instances of reconstructed cardiac images. Bottom rows show two instances of reconstructed prostate images. Blue squares zoom in the corresponding small squares in the images.

Method	PSNR (dB)					SSIM (%)
Method	$10\%$	$20\%$	$30\%$	$15\%^{\dagger}$	$25\%^{\dagger}$	$10\%$	$20\%$	$30\%$	$15\%^{\dagger}$	$25\%^{\dagger}$
DnCn	29.85	35.85	39.29	31.93	35.75	86.82	94.37	96.55	89.92	94.18
LDA	31.27	37.11	40.67	33.26	37.80	89.36	95.36	97.17	92.13	95.83
UNet	26.93	29.16	31.20	28.03	29.97	76.17	79.55	83.49	77.63	81.11
U-MRI	30.60	36.25	39.42	33.58	37.33	86.07	94.54	96.49	91.91	95.22
Meta	27.63	32.97	35.73	30.72	34.44	82.36	91.22	94.06	87.86	92.87

Table 2: Average PSNR and SSIM of reconstructed images with sampling ratios 10%, 15%, 20%, 25%, and 30% in Section 5.2.2 . The sampling ratios marked by

\dagger

(15% and 25%) have small training datasets. For each sampling ratio, the same Cartesian mask is used to undersample the

k

-spaces of all images.

Panel of Figure 3 — Figure 3: Comparison of cross-sampling-ratio TL reconstruction results in Section 5.2.2 . Top two rows show two instances of reconstructed images under 15% sampling ratio. Bottom rows show two instances of reconstructed images under 25% sampling ratio. Blue squares zoom in the corresponding small squares in the images.

Method	PSNR (dB)				SSIM (%)
Method	ImageNet	CIFAR-10	$\text{fastMRI}^{\dagger}$	$\text{Stanford2D}^{\dagger}$	ImageNet	CIFAR-10	$\text{fastMRI}^{\dagger}$	$\text{Stanford2D}^{\dagger}$
DnCn	33.90	38.68	31.90	36.29	92.87	97.09	84.97	92.33
LDA	35.37	41.33	33.42	37.85	94.26	97.99	87.30	94.23
UNet	31.36	36.27	30.88	35.33	89.94	96.40	82.58	91.00
U-MRI	34.33	39.72	30.50	34.48	93.37	97.66	80.96	89.08
Meta	32.81	38.11	30.93	35.04	90.95	96. 26	82.35	90.47

Table 3: Average PSNR and SSIM of reconstructed images on different modality domains in Section 5.2.3 . The modalities marked by

\dagger

(fastMRI and Stanford2D) have small training datasets.

Panel of Figure 4 — Figure 4: Comparison of cross-modality reconstruction results in Section 5.2.3 . Knowledge is transferred from natural image datasets ImageNet and CIFAR-10 to medical image reconstruction. Top two rows show two instances of reconstructed images in the fastMRI dataset. Bottom rows show two instances of reconstructed images in the Stanford2D dataset. Blue squares zoom in the corresponding small squares in the images.

Panel of Figure 5 — Figure 5: Reconstructed images, zoom-in views, and pointwise absolute errors obtained by the compared methods on the cross-anatomy experiment (top three rows), cross-sampling-ratio experiment (middle three rows), and cross-modality experiment (bottom three rows).

Panel of Figure 6 — Figure 6: CKA (left column) and NCC (right column) between $\mathbf{g}_{i}$ trained by LDA on $D_{i}$ and $\mathbf{g}$ trained by U-LDA versus phase number $T$ . Top and bottom rows show the results on brain ( $\mathbf{x}_{1}$ ) and knee ( $\mathbf{x}_{2}$ ) images, respectively.

Panel of Figure 7 — Figure 7: Average PSNR of reconstructed images with three different target domain training sample sizes (5, 40, and 100). Each bar represents the average PSNR and standard deviation (black interval) obtained by the corresponding method over five tests.

Methods			Results (PSNR)
Adapter	Init	AUG	Cross-Anatomy	Cross-Sampling-Rate	Cross-Modality
$\times$	$\times$	$\times$	32.92	33.26	33.42
$\checkmark$	$\times$	$\times$	32.84	33.48	33.45
$\checkmark$	$\checkmark$	$\times$	33.10	33.68	33.47
$\checkmark$	$\checkmark$	$\checkmark$	33.31	34.90	33.58

Table 4: PSNR of images reconstructed by U-LDA with (

\checkmark

) or without (

\times

) adapters (Adapter), initialization trick of

\mathbf{g}

(Init), and data augmentation (AUG).

$\alpha$	0.1	0.01	0.001
PSNR	36.45	37.88	37.69

Table 5: Average PSNR of images reconstructed by U-LDA using different values of SSIM weight parameter

\alpha

$\delta$	0.0001	0.001	0.01	0.1
PSNR	27.32	37.88	37.79	37.43

Table 6: Average PSNR of images reconstructed by U-LDA using different values of smoothing level

\delta

in the smoothed ReLU activation.

Explore similar work

Open World MRI Reconstruction with Bias-Calibrated Adaptation
Mar 13, 2026Jiyao Liu, Shangqi Gao, Lihao Liu +5Image Reconstruction Domain Adaptation
Fudan University, Shanghai, China · University of Cambridge, Cambridge, United Kingdom · Shanghai Artificial Intelligence Laboratory, Shanghai, China
Bridging the Inter-Domain Gap through Low-Level Features for Cross-Modal Medical Image Segmentation
May 17, 2025Pengfei Lyu, Pak-Hei Yeung, Jing Xia +5Cross-Modal Learning Cross-Modality Medical Image Segmentation
Northeastern University, Shenyang, China · Nanyang Technological University, Singapore
Data-Efficient Networks for Multi-Contrast MRI Reconstruction based on a Generalized Content/Style Prior
Sep 2, 2026Chinmay Rao, Efe Ilıcak, Matthias J. P. van Osch +5Accelerated MRI Reconstruction Magnetic Resonance Imaging
Department of Radiology, Leiden University Medical Center, Leiden, The Netherlands · Philips Innovative Technologies, Hamburg, Germany · Philips Cardiologs, Paris, France +1

Mar 13, 2026J/K move · Enter open · S save

Open World MRI Reconstruction with Bias-Calibrated Adaptation

Jiyao Liu, Shangqi Gao, Lihao Liu, Junzhi Ning+4

Fudan University, Shanghai, China · University of Cambridge, Cambridge, United Kingdom · Shanghai Artificial Intelligence Laboratory, Shanghai, China

Image Reconstruction Domain Adaptation Accelerated MRI Reconstruction Diffusion-Based Inverse Problems

Open paper AskPDF arXiv

Real-world MRI reconstruction systems face the open-world challenge: test data from unseen imaging centers, anatomical structures, or acquisition protocols can differ drastically from training data, causing severe performance degradation. Existing methods struggle with this challenge. To address this, we propose BiasRecon, a bias-calibrated adaptation framework grounded in the minimal intervention principle: preserve what transfers, calibrate what does not. Concretely, BiasRecon formulates open-world adaptation as an alternating optimization framework that jointly optimizes three components: (1) frequency-guided prior calibration that introduces layer-wise calibration variables to selectively modulate frequency-specific features of the pre-trained score network via self-supervised k-space signals, (2) score-based denoising that leverages the calibrated generative prior for high-fidelity image reconstruction, and (3) adaptive regularization that employs Stein's Unbiased Risk Estimator to dynamically balance the prior-measurement trade-off, matching test-time noise characteristics without requiring ground truth. By intervening minimally and precisely through this alternating scheme, BiasRecon achieves robust adaptation with fewer than 100 tunable parameters. Extensive experiments across four datasets demonstrate state-of-the-art performance on open-world reconstruction tasks.

Abstract

Explore similar work

Open World MRI Reconstruction with Bias-Calibrated Adaptation

Bridging the Inter-Domain Gap through Low-Level Features for Cross-Modal Medical Image Segmentation

Data-Efficient Networks for Multi-Contrast MRI Reconstruction based on a Generalized Content/Style Prior