EUFormer: Learning Driven 3D Spine Deformity Assessment with Orthogonal Optical Images

Area of Science:

• Medical Imaging
• Orthopedics
• Computer Vision

Background:

• Current adolescent idiopathic scoliosis (AIS) screening relies on subjective physical exams or radiation-exposed radiographic methods.
• There is a need for accurate, non-invasive methods for AIS diagnosis and severity grading.

Purpose of the Study:

• To develop and validate a novel pipeline for accurate AIS severity grading using RGB images.
• To introduce an Efficient U-shape transFormer (EUFormer) for precise 2D spine curve generation.

Main Methods:

• A pipeline utilizing posteroanterior (PA) and lateral (LAT) RGB images to generate spine curve maps.
• Reconstruction of a 3D spine curve from 2D maps for AIS severity grading.
• Implementation of the Efficient U-shape transFormer (EUFormer) for accurate and efficient 2D spine curve generation.

Main Results:

• The proposed EUFormer demonstrated superior performance in spine curve generation compared to classical U-shape models.
• The 3D spine curve reconstruction provided a more accurate assessment of AIS severity than 2D curves.
• The pipeline successfully generated consecutive spine curves from both PA and LAT views.

Conclusions:

• The developed pipeline offers a radiation-free and accurate method for AIS severity grading.
• Utilizing a 3D spine curve reconstruction enhances the precision of AIS severity assessment.
• EUFormer shows significant potential for improving spinal imaging analysis in clinical settings.