SU‐D‐116‐05: Automatic Detection of Ring and Streak Artifacts in Routine CT QC Images

Purpose: To develop an image analysis algorithm for effectively detecting ring and streak artifacts in CT water phantom images to improve the review of a large number of daily QC images. Methods: Daily QC water phantom images were acquired and transferred to a computer server. The images were analyzed using a computer program developed for ring and streak artifact detection. For ring artifact detection, an average HU along circles of different radii is computed. A statistical bias correction accounts for different levels of noise depending on the circle radius. A score is assigned based on the maximum difference of the HU averages between circles. For streak artifact detection, HU averages are computed along parallel lines at different angles. The noise level bias due to different line lengths in the phantom is corrected. Artifacts are scored using the maximum difference between HU averages along all lines. GPU computation is used for real time computation. Scores are accessed via a web interface, where a QC technologist reviews 700–1000 routine daily QC scores sorted by severity and visually inspects the highest scored images. Thresholds of scores can be defined form service actions depending on severity of the artifacts. An email notification is sent automatically when a score is above a threshold value. Results: The program of automatic ring artifact detection has been in use for our daily CT QC of 21 CT scanners for over one year. Visual evaluations of QC images for artifacts confirmed that this automatic system detected ring artifacts and functioned as expected. Streak artifact detection has been implemented more recently and requires validation. This program has also been extended to process large‐phantom images acquired on a weekly basis. Conclusion: Automatically detecting ring and streak artifacts in CT QC images is practically feasible, reduces manpower and provides consistent results.