Parametric-MAA: fast, object-centric avoidance of metal artifacts for intraoperative CBCT

Area of Science:

• Medical Imaging
• Computer-Aided Surgery
• Orthopedic Imaging

Background:

• Metal artifacts in intraoperative CBCT imaging hinder visualization of critical areas, especially in orthopedic and trauma surgery.
• Existing metal artifact avoidance (MAA) methods often fail clinically due to computational demands and focus on non-essential objects.

Purpose of the Study:

• Introduce a novel parametric metal artifact avoidance (P-MAA) method to address limitations of current MAA techniques.
• Develop a computationally efficient approach for trajectory optimization in CBCT imaging.

Main Methods:

• Utilize a deep learning model to detect keypoints in scout views for ellipsoidal modeling of relevant objects.
• Devise a computationally efficient scoring metric based on ellipsoidal representations for fast, CPU-based trajectory optimization.
• Train and validate the object localization model using simulated and real clinical data.

Main Results:

• The detection model achieved a mean average recall of 0.78, showing generalizability to clinical cases.
• The ellipsoid-based scoring method closely approximated raytracing results and proved effective in complex scenarios.
• The ellipsoid method demonstrated a 33-fold speed increase without requiring GPU acceleration.

Conclusions:

• The P-MAA approach offers a feasible solution for intraoperative CBCT metal artifact avoidance.
• Enables rapid trajectory optimization focused on clinically relevant structures.
• Represents a significant advancement toward practical clinical implementation of MAA.