Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Reconstruct Spine CT from Biplanar X-Rays via Diffusion Learning

Zhi Qiao, Xuhui Liu, Xiaopeng Wang, Runkun Liu, Xiantong Zhen, Pei Dong, Zhen Qian

TL;DR

The paper tackles reconstructing 3D spine CT from intraoperative biplanar X-rays when CT is unavailable. It introduces Diff2CT, a conditional diffusion model that leverages two orthogonal X-ray views as input, with X2NoiseNet converting 2D data into 3D-conditioned features and a projection-based loss enforcing 3D structural fidelity. On a lumbar pedicle screw dataset, Diff2CT achieves a higher perceptual quality (SSIM) and lower FID than state-of-the-art baselines, albeit with some trade-offs in voxel-wise metrics. This work demonstrates the potential of diffusion-based CT reconstruction for intraoperative guidance and paves the way for further improvements with real paired data and refined conditioning strategies.

Abstract

Intraoperative CT imaging serves as a crucial resource for surgical guidance; however, it may not always be readily accessible or practical to implement. In scenarios where CT imaging is not an option, reconstructing CT scans from X-rays can offer a viable alternative. In this paper, we introduce an innovative method for 3D CT reconstruction utilizing biplanar X-rays. Distinct from previous research that relies on conventional image generation techniques, our approach leverages a conditional diffusion process to tackle the task of reconstruction. More precisely, we employ a diffusion-based probabilistic model trained to produce 3D CT images based on orthogonal biplanar X-rays. To improve the structural integrity of the reconstructed images, we incorporate a novel projection loss function. Experimental results validate that our proposed method surpasses existing state-of-the-art benchmarks in both visual image quality and multiple evaluative metrics. Specifically, our technique achieves a higher Structural Similarity Index (SSIM) of 0.83, a relative increase of 10\%, and a lower Fréchet Inception Distance (FID) of 83.43, which represents a relative decrease of 25\%.

Reconstruct Spine CT from Biplanar X-Rays via Diffusion Learning

TL;DR

The paper tackles reconstructing 3D spine CT from intraoperative biplanar X-rays when CT is unavailable. It introduces Diff2CT, a conditional diffusion model that leverages two orthogonal X-ray views as input, with X2NoiseNet converting 2D data into 3D-conditioned features and a projection-based loss enforcing 3D structural fidelity. On a lumbar pedicle screw dataset, Diff2CT achieves a higher perceptual quality (SSIM) and lower FID than state-of-the-art baselines, albeit with some trade-offs in voxel-wise metrics. This work demonstrates the potential of diffusion-based CT reconstruction for intraoperative guidance and paves the way for further improvements with real paired data and refined conditioning strategies.

Abstract

Intraoperative CT imaging serves as a crucial resource for surgical guidance; however, it may not always be readily accessible or practical to implement. In scenarios where CT imaging is not an option, reconstructing CT scans from X-rays can offer a viable alternative. In this paper, we introduce an innovative method for 3D CT reconstruction utilizing biplanar X-rays. Distinct from previous research that relies on conventional image generation techniques, our approach leverages a conditional diffusion process to tackle the task of reconstruction. More precisely, we employ a diffusion-based probabilistic model trained to produce 3D CT images based on orthogonal biplanar X-rays. To improve the structural integrity of the reconstructed images, we incorporate a novel projection loss function. Experimental results validate that our proposed method surpasses existing state-of-the-art benchmarks in both visual image quality and multiple evaluative metrics. Specifically, our technique achieves a higher Structural Similarity Index (SSIM) of 0.83, a relative increase of 10\%, and a lower Fréchet Inception Distance (FID) of 83.43, which represents a relative decrease of 25\%.
Paper Structure (11 sections, 4 equations, 3 figures, 1 table)

This paper contains 11 sections, 4 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Proposed Method
  3. X2NoiseNet
  4. Loss Functions
  5. Experiments
  6. Datasets
  7. Settings
  8. Qualitative Results
  9. Quantitative Results
  10. Case Study
  11. Conclusion

Figures (3)

  • Figure 1: Representative CT reconstruction results shown in the mid-axial (1st row), mid-sagittal (2nd row), and mid-coronal (3rd row) views. Our method is compared with baseline methods (PSR, 3DCNN, Diff2CT) and the ground truth (GT).
  • Figure 2: Analysis of results (A lighter color indicates a larger SSIM value, and a larger diameter of the circle indicates a larger PSNR value)
  • Figure 3: A case for low resolution CT image with original pixel spacing of [0.9765,0.9765,0.8]. In preprocessing, we firstly resample all of CT images into a unified [2,2,2] pixel spacing.