An accurate, straightforward computer vision algorithm for optimal tumor-feeding visualization in cone-beam computed...

Area of Science:

• Medical imaging
• Interventional radiology
• Computer vision

Background:

• Standardized 3D volume rendering techniques (VRT) and embolization guidance aid in visualizing tumor-feeding arteries.
• Current vessel tracking software lacks automatic angle recommendations, necessitating manual adjustments by operators.
• Manual angle adjustments disrupt procedures and require time-consuming re-scrubbing.

Purpose of the Study:

• To develop a computer vision algorithm for automatically recommending optimal rotation/angles in VRT.
• To maximize the visualization of tumor-feeding arteries, specifically for hepatocellular carcinoma.
• To improve procedural efficiency for interventional radiologists.

Main Methods:

• The algorithm processes post-embolization guidance frames from 3D VRT.
• It rotates the VRT in 5° intervals, optimizing one axis while rotating another.
• The algorithm segments the embolization guidance line and identifies optimal views by maximizing line length and convex hull area.

Main Results:

• The algorithm was developed and iterated using 19 patient cases and expert feedback.
• In a 50-patient validation set, the top 4 suggested views/angles always included a sufficient view for embolization.
• This achieved 100% retrieval relevance according to an experienced interventional radiologist.

Conclusions:

• Traditional computer vision effectively achieves sufficient view/angle selection for hepatic artery embolization.
• The proposed technique is faster and more explainable than deep learning methods.
• Further studies with more patients and technical iterations are recommended to enhance procedural efficiency.