Automated Metal Artifact Segmentation for Extended CT In Radiation Therapy Planning: A Feasibility Study
Abstract
Purpose
The increase of prostheses in patients throughout the years has resulted in different strategies to alleviate streaking artifacts in CT scans that obscure nearby anatomy and complicate treatment planning. Manual density overrides can be time-intensive while also lacking standardization. Fortunately, our institution uses an extended CT scale which allows high-density metals such as titanium to be observed at true HU values. We developed and evaluated the feasibility of a standardized algorithm that contours artifacts and metals while automating density overrides.
Methods
The algorithm employs multi-planar analysis across axial, coronal, and sagittal projections to identify metals using adaptive thresholding from extended CT datasets. In each slice, a 16-line star HU profile is generated stemming from the prosthesis centroid. Full-width at 75% maximum thresholding calculates thresholds at 75% of peak HU values averaged across profiles, allowing generalization across prosthesis types and metal compositions. Preliminary segmentation uses sequential nested Boolean operations for dark artifacts, bright artifacts, bone, and metals. Two discrimination methods distinguish bone and tissue from artifacts: distance-based segmentation using radial proximity, local variance, and HU filtering; star profile-based discrimination analyzing the 16 profiles for peak width, smoothness, and gradient characteristics.
Results
Automated metal detection and segmentation were successful across varying implant geometries and compositions, and adaptive thresholding provided consistent metal boundary detection independent of prosthesis material. Both discrimination methods offer user-adjustable thresholds for clinical flexibility. Furthermore, star profile discrimination distinguishes bone and tissue from artifacts based on characteristic peak widths, though further refinement is needed. Both methods reduce manual override time from an hour to under 3 minutes, demonstrating improved efficiency and standardization.
Conclusion
This feasibility study demonstrates two automated analytical approaches for metal artifact characterization with the potential for fully automated density overrides, improving clinical workflow efficiency. Future work will establish electron density override protocols and validate dosimetric accuracy.