Quantitative phantom-based comparison of commercial CT metal artifact reduction algorithms

Metal artifacts in computed tomography (CT) significantly compromise image quality and CT number (CTN) accuracy, which may influence downstream applications such as radiation therapy workflows. Although commercial metal artifact reduction (MAR) algorithms are widely used, their performance varies across vendors and clinical scenarios, and standardized methods for quantitative comparison remain limited.

This study quantitatively compares four commercial MAR algorithms using standardized metrics and a novel analytical approach.

A custom-designed resin phantom with interchangeable inserts (solid water, aluminum, titanium, and stainless steel) was scanned using four CT systems: Canon, GE, Philips, and Siemens. Each system's proprietary MAR algorithm was applied to obtain artifact-reduced image sets. Quantitative analysis employed complementary methods: (1) circular profile analysis measuring mean CTN deviations at concentric distances from metal objects; (2) geometric accuracy assessment of metal objects; and (3) volumetric artifact characterization using analysis of severe artifact voxels and composite error metric (M-error index) that integrates both artifact extent and intensity. Color maps and accumulated histograms of ΔCTN values between pre- and post-MAR images were used to characterized artifact patterns and MAR performance.

All four MAR methods showed minimal |ΔCTN| at mid (2.80 cm) and far (3.80 cm) distances but showed pronounced increases in |ΔCTN| near the metal interface at the closest radius (1.27 cm). Smart-MAR exhibited the highest CTN deviation for aluminum (16.83 HU), while O-MAR showed the largest deviations for titanium and stainless steel (42.40 and 42.60 HU). Quantitative evaluation showed that single-energy metal artifact reduction (SEMAR) and iterative metal artifact reduction (iMAR) achieved the most consistent geometric accuracy (<5%) across all metals. O-MAR showed the highest accuracy for titanium (approximately 0.36%) but larger deviations for aluminum (approximately 9%), while Smart-MAR exhibited the greatest errors exceeding 19% for both titanium and stainless steel. SEMAR, Smart-MAR, and iMAR showed lower M Error values (0.00-0.83) than O-MAR, which showed significantly higher values (1.15-2.45).

SEMAR and iMAR achieved the most consistent geometric accuracy (<5% error) and artifact suppression across all metals, while O-MAR demonstrated material-dependent performance with substantial residual artifacts near metal interfaces. These phantom-based benchmarks provide vendor-neutral performance data on MAR performance under controlled settings.