Optimizing neurointerventional procedures: an algorithm for embolization coil detection and automated collimation to...

Area of Science:

• Medical Imaging
• Interventional Radiology
• Artificial Intelligence in Medicine

Background:

• Radiation dose and time monitoring are critical in radiological interventions, particularly neurointerventions like aneurysm coiling.
• Current methods may not fully optimize image quality or minimize patient dose during these procedures.

Purpose of the Study:

• To develop and evaluate an algorithm for detecting embolization coils in medical images.
• To enhance the efficiency and safety of neurointerventional procedures through automated collimation.
• To optimize image quality while minimizing patient radiation dose.

Main Methods:

• Deep learning models (Faster R-CNN with ResNet-50 FPN, RetinaNet) were employed.
• A classical blob detection approach was used as a benchmark.
• Fivefold cross-validation was performed for model evaluation.

Main Results:

• The top-performing model achieved a mean Average Precision at 75% overlap (mAP@75) of 0.84 on validation data and 0.94 on test data.
• Simulations demonstrated reduced dose area product and minimized scatter radiation through automatic collimation.
• The algorithm improved X-ray angiography image quality during narrow collimation, streamlining the process for physicians.

Conclusions:

• This is the first reported approach for successful embolization coil detection, with potential for integration into X-ray angiography systems.
• The developed method can be extended to detect other medical objects in interventional procedures.
• The algorithm enhances procedural safety, efficiency, and image quality while reducing radiation exposure.