Click here to

Session: Advances in CT II [Return to Session]

Development of a Generalizable, Vendor-Neutral Metal Object Insertion Framework to Simulate Metal Devices and Image Artifacts in Interventional CT

C Favazza*, L Ren, W Cao, A Missert, A Ferrero, Mayo Clinic, Rochester, MN


TU-D930-IePD-F8-2 (Tuesday, 7/12/2022) 9:30 AM - 10:00 AM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 8

Purpose: To assess variability of metal artifacts different CT scanners and develop a generalizable framework to simulate the presence of metallic devices encountered during CT-guided procedures.

Methods: Three different needles were inserted into a water phantom and scanned with two different scanner models: Siemens Edge and Philips Big Bore. Device locations and acquisition parameters were matched across both scanners. Metal artifact severity was assessed by calculating Structural Similarity Index (SSIM) between metal artifact and water images. Separate sets of scans were performed to create needle masks to simulate their presence in patient images. The following phantom combinations were scanned: (1) empty water tank, (2) empty water tank + needle, (3) full water tank + needle, (4) full water tank only. Artifact-free needle masks were created by subtracting the empty water tank image from the empty water tank + needle image. Masks of the needle with artifact were created by subtracting the full water tank image from the full water tank + needle image. These masks were then added to patient data in the image domain to simulate the presence of needles with and without artifact. Needle insertions derived from both vendors were applied to patient data from both vendors for qualitative comparison.

Results: Metal artifact images from the two scanner models yielded different SSIM scores, varying by up to 15%. Qualitative comparison of simulated needles and artifacts in patient data show similarity between like scanner models; whereas, cross-model simulations yielded noticeable differences.

Conclusion: Preliminary results demonstrate artifact appearance varies across scanner models and separate simulation models are needed to account for such differences. The presented image domain insertion method can capture these differences and simulate the presence of metal devices commonly used in CT-guided procedures—with and without the artifacts. This framework could be used to train metal artifact reduction algorithms.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email