X2P-Net: Context-Aware 2D/3D Vertebra Localization

Area of Science:

• Medical Imaging
• Computer Vision
• Spine Surgery

Background:

• Accurate 3D vertebral localization from 2D X-rays is vital for minimally invasive spine surgery.
• Existing methods face challenges due to the ill-posed nature of 2D-to-3D localization and spinal column ambiguities.

Purpose of the Study:

• Introduce X2P-net, a prompt-guided, semantic context-enhanced 2D/3D vertebra detection framework.
• Enhance the accuracy and efficiency of vertebral localization in surgical settings.

Main Methods:

• Developed BrickFormer, a novel Transformer architecture with a dual-attention mechanism.
• Employed prompt-guided and semantic context enhancement for refined vertebral foreground extraction.
• Validated the framework on synthetic (BiSpineX) and animal (SheepSpineX) datasets.

Main Results:

• X2P-net demonstrated superior landmark localization performance compared to state-of-the-art methods.
• Achieved high accuracy on BiSpineX: 96.9% at 10mm, 98.8% at 20mm, 2.99mm mean error, 0.9923 AUC.
• Showcased excellent results on SheepSpineX: 98.4% at 10mm, 100.0% at 20mm, 1.08mm mean error, 0.9972 AUC.

Conclusions:

• The proposed X2P-net framework significantly improves 2D/3D vertebra detection and localization accuracy.
• BrickFormer's efficient context extraction contributes to the method's effectiveness.
• X2P-net offers a promising solution for enhancing surgical navigation in minimally invasive spine procedures.