Toward automatic C-arm positioning for standard projections in orthopedic surgery

Area of Science:

• Medical imaging
• Artificial intelligence in surgery
• Orthopedic surgery

Background:

• Intra-operative fluoroscopy with mobile C-arms is crucial in orthopedic surgery.
• Accurate, standardized C-arm projections are essential but manual positioning is time-consuming and error-prone.
• Reducing radiation exposure and improving accuracy in C-arm positioning are key challenges.

Purpose of the Study:

• To develop an automated C-arm positioning procedure guided by deep learning.
• To reduce time, radiation exposure, and errors in orthopedic surgical procedures.
• To create a system that can guide and eventually automate C-arm positioning.

Main Methods:

• A convolutional neural network regression model predicts C-arm pose updates from X-ray images.
• The model learns anatomical hints from in silico simulations using 3D CT scans.
• The approach does not require patient-specific pre-operative data or additional equipment.

Main Results:

• The method was validated on lumbar spine and proximal femur anatomies.
• Iterative pose error reduction was achieved, improving accuracy to desired standard poses.
• The approach generalized to real X-ray data from cadaveric hip joints without retraining.

Conclusions:

• Deep learning-based automated C-arm positioning trained on simulations is feasible.
• The proposed 2-stage approach significantly improves accuracy on synthetic and real X-ray images.
• Simulation-based learning translates to acceptable performance in clinical settings.