LatXGen: Toward Radiation-Free and Accurate Quantitative Analysis of Sagittal Spinal Alignment via Cross-Modal...

Area of Science:

• Medical Imaging
• Orthopedics
• Artificial Intelligence

Background:

• Adolescent Idiopathic Scoliosis (AIS) requires assessing both coronal and sagittal spinal alignment.
• Current methods for sagittal alignment evaluation often involve ionizing radiation.
• Radiation-free assessment of sagittal spinal alignment in AIS is an underexplored area.

Purpose of the Study:

• To introduce LatXGen, a generative framework for synthesizing lateral spinal radiographs from RGBD images.
• To enable accurate, radiation-free estimation of sagittal spinal alignment in AIS.
• To address the challenges of inferring sagittal morphology and cross-modality translation.

Main Methods:

• LatXGen utilizes a dual-stage generative framework.
• It incorporates an attention-based Fast Fourier Convolution (FFC) module and a Spatial Deformation Network (SDN).
• A large-scale dataset of 3,264 paired RGBD and lateral radiograph images was created.

Main Results:

• LatXGen successfully synthesizes anatomically accurate lateral spinal radiographs.
• The framework demonstrates superior performance compared to existing GAN-based methods in visual fidelity and quantitative metrics.
• It effectively infers sagittal spinal morphology and performs cross-modality translation.

Conclusions:

• LatXGen offers a promising, radiation-free solution for sagittal spine assessment in AIS.
• This technology advances the comprehensive evaluation of AIS by enabling accurate sagittal alignment measurement.
• The developed framework and dataset contribute significantly to the field of medical imaging for spinal deformities.