Uncertainty-guided Generation of Dark-field Radiographs

TL;DR

The study tackles the scarcity of X-ray dark-field data by proposing an uncertainty-guided, progressive GAN that translates attenuation chest X-rays into dark-field images while modeling both aleatoric and epistemic uncertainties. It introduces per-pixel generalized Gaussian uncertainty with parameters $α_{ij}$ and $β_{ij}$ and uses Monte Carlo dropout to capture model uncertainty, coupled with a residual consistency loss to better represent texture. The framework is trained in three progressively refined stages and evaluated on a paired in-house dataset, with out-of-domain generalization tested on NIH Chest X-ray data; results show progressive improvements in MSE, PSNR, and SSIM, and uncertainty maps offer interpretability and reliability. The work demonstrates realistic dark-field synthesis with uncertainty-aware diagnostics, enabling scalable data generation and laying groundwork for uncertainty-aware clinical deployment in lung imaging.

Abstract

X-ray dark-field radiography provides complementary diagnostic information to conventional attenuation imaging by visualizing microstructural tissue changes through small-angle scattering. However, the limited availability of such data poses challenges for developing robust deep learning models. In this work, we present the first framework for generating dark-field images directly from standard attenuation chest X-rays using an Uncertainty-Guided Progressive Generative Adversarial Network. The model incorporates both aleatoric and epistemic uncertainty to improve interpretability and reliability. Experiments demonstrate high structural fidelity of the generated images, with consistent improvement of quantitative metrics across stages. Furthermore, out-of-distribution evaluation confirms that the proposed model generalizes well. Our results indicate that uncertainty-guided generative modeling enables realistic dark-field image synthesis and provides a reliable foundation for future clinical applications.

Figures (5)

• Figure 1: (a) Measurement setup for the grating-based dark-field scanner. The subject stands in the scanner while the grating assembly scans from bottom to top. (b) Attenuation image and (c) dark-field image of a healthy subject, jointly acquired in a single scan.
• Figure 2: Proposed Uncertainty-Guided Progressive GAN framework, illustrated for model training at stage three. At this stage, the previous layers are frozen. The aleatoric uncertainty estimates are used as attention maps to guide the model's weights for refinement in the subsequent stage. Dropout is incorporated to the generators at each stage.
• Figure 3: Comparison of attenuation, real dark-field, and generated dark-field images for two patients.
• Figure 4: Results and corresponding uncertainty estimates across the three training stages for a single patient. All images are cropped to the lung region to enhance visualization.
• Figure 5: Attenuation and generated dark-field images along with their corresponding uncertainty estimates for the out-of-distribution NIH Chest X-ray dataset, shown for 3 patients.