Predicting 3D forearm fracture angles from bipolar X-ray images with rotational bone pose estimation

Why is it challenging to reconstruct 3D fracture angle from biplanar X-ray images?

Background

• X-ray is the primary choice for diagnosing forearm fractures.
• Its 2D projection nature obscures vital details, such as precise fracture angles and posture variations.
• When patients undergo follow-up X-rays to assess forearm fractures, replicating the exact arm positioning from previous images is challenging.
• It is difficult for clinicians to discern whether fracture angulation changes reflect clinical progression or are mere artifacts of different positioning.
• Such discrepancies can hinder precise diagnosis, surgical preparation, and post-treatment evaluation by physician

• X-ray is the primary choice for diagnosing forearm fractures.
• Its 2D projection nature obscures vital details, such as precise fracture angles and posture variations.
• When patients undergo follow-up X-rays to assess forearm fractures, replicating the exact arm positioning from previous images is challenging.
• It is difficult for clinicians to discern whether fracture angulation changes reflect clinical progression or are mere artifacts of different positioning.
• Such discrepancies can hinder precise diagnosis, surgical preparation, and post-treatment evaluation by physicianpower

Solution

We take the bone posture estimation into reconsideration during 3D reconstruction.

Method Pipeline

• 2D fracture measurement from biplanar X-ray images.
• Bone pose estimation network.
• 3D reconstruction

How to train the Bone pose estimation network?

• Segmented 20 CT images and manually created bone fractures; 40 healthy bones + 40 fractured bones.

• Generated DR images by projecting 30 bones in various directions.
• Simulated X-ray images for pose estimation training.