X-Recon: Learning-based Patient-specific High-Resolution CT Reconstruction from Orthogonal X-Ray Images

TL;DR

X-Recon presents a dual-view, GAN-based framework for ultra-sparse CT reconstruction from orthogonal chest X-rays, achieving high-resolution CT volumes (224×224×224) with improved anatomical fidelity and quantitative pneumothorax metrics. The MFusionRen-enabled generator and 3D CoordConv discriminator, guided by a ProST-driven multi-angle projection loss, effectively bridge 2D X-ray inputs to 3D CT outputs. Complementing this, PTX-Seg provides zero-shot segmentation of air-accumulated regions and lungs to enable accurate pleural occupancy quantification, achieving strong segmentation correlations with ground truth. Together, these components demonstrate high reconstruction quality, improved pneumothorax assessment, and potential clinical utility for radiation-safe, patient-specific chest diagnostics, with avenues for future domain adaptation and higher-resolution gains.

Abstract

Rapid and accurate diagnosis of pneumothorax, utilizing chest X-ray and computed tomography (CT), is crucial for assisted diagnosis. Chest X-ray is commonly used for initial localization of pneumothorax, while CT ensures accurate quantification. However, CT scans involve high radiation doses and can be costly. To achieve precise quantitative diagnosis while minimizing radiation exposure, we proposed X-Recon, a CT ultra-sparse reconstruction network based on ortho-lateral chest X-ray images. X-Recon integrates generative adversarial networks (GANs), including a generator with a multi-scale fusion rendering module and a discriminator enhanced by 3D coordinate convolutional layers, designed to facilitate CT reconstruction. To improve precision, a projective spatial transformer is utilized to incorporate multi-angle projection loss. Additionally, we proposed PTX-Seg, a zero-shot pneumothorax segmentation algorithm, combining image processing techniques with deep-learning models for the segmentation of air-accumulated regions and lung structures. Experiments on a large-scale dataset demonstrate its superiority over existing approaches. X-Recon achieved a significantly higher reconstruction resolution with a higher average spatial resolution and a lower average slice thickness. The reconstruction metrics achieved state-of-the-art performance in terms of several metrics including peak signal-to-noise ratio. The zero-shot segmentation algorithm, PTX-Seg, also demonstrated high segmentation precision for the air-accumulated region, the left lung, and the right lung. Moreover, the consistency analysis for the pneumothorax chest occupancy ratio between reconstructed CT and original CT obtained a high correlation coefficient. Code will be available at: https://github.com/wangyunpengbio/X-Recon

Figures (8)

• Figure 1: Patient-specific high-resolution CT reconstruction from orthogonal X-rays. $I_{Pa}$ and $I_{La}$ represent posteroanterior and lateral X-ray images and $V_{Rec}$ represents the volumetric CT image reconstructed by the network $\mathcal{F}_{\Theta}$.
• Figure 2: The overview architecture of X-Recon, the model takes posteroanterior and lateral X-ray images as inputs, and the generator produces the reconstructed CT image. The loss functions employed in the training process involve three main components: the reconstruction loss, the multi-angle digitally reconstructed radiograph loss, and the generative adversarial loss. C denotes concatenation, straight arrows with different colors represent distinct data flows, and red dashed arrows indicate the computation of corresponding loss functions between predictions and GTs.
• Figure 3: Details of the network structure of the X-Recon generator. It consists of two encoder-decoder subbranches as well as a decoder main branch and an output branch. Each encoder-decoder branch is used to process information from an independent viewpoint. The decoder main branch is used for information fusion and the output branch is used for CT overlay rendering.
• Figure 4: Illustration of the basic fusion block of MFusionRend Module: Fuse and Fuse-out block. They receive information from both decoders and perform information fusion. The Fuse-out block is an enhanced version of the Fuse block with an additional information output function.
• Figure 5: Visualization of X-Recon reconstruction results. The first two rows are samples from healthy subjects and the last two rows are samples from pneumothorax patients. The images are presented in three anatomical planes: transverse, sagittal, and coronal planes.