SkeDiff: Skeleton 3D CT Diffusion Reconstruction using 2D X-ray.

Area of Science:

• Medical Imaging
• Computer Vision
• Orthopedics

Background:

• 2D X-ray imaging is accessible but limited in visualizing skeletal structures.
• 3D CT imaging offers comprehensive visualization but involves higher radiation exposure.
• Bridging the gap between 2D X-ray and 3D CT is crucial for improved orthopedic diagnostics.

Purpose of the Study:

• To develop SkeDiff, an algorithm for reconstructing 3D CT images from 2D X-ray projections.
• To enhance 3D skeletal visualization for orthopedic diagnostics.
• To integrate scoliosis classification into the 3D reconstruction process.

Main Methods:

• SkeDiff utilizes a 3D diffusion model ($DM_{3DL}$) guided by 2D priors extracted via a cross-dimensional conditional encoder ($E_{Cond}$).
• The encoder employs a CNN-Mamba hybrid architecture for advanced feature extraction.
• A 3D UKAN diffusion backbone with Kolmogorov-Arnold Networks (KAN) improves feature representation.
• A diffusion-based scoliosis classifier ($D_{SC}$) is incorporated for simultaneous classification.

Main Results:

• SkeDiff successfully reconstructs 3D CT images of the skeleton from orthogonal 2D X-ray projections.
• The algorithm demonstrates superior performance compared to existing methods on spine, hip, and knee datasets.
• The integrated scoliosis classifier functions effectively during the 3D reconstruction.

Conclusions:

• SkeDiff offers a promising solution for generating high-quality 3D skeletal reconstructions from 2D X-rays.
• The method has the potential to improve orthopedic diagnostic capabilities.
• This approach could lead to more comprehensive and efficient skeletal imaging analysis.