Deep Learning Reconstruction with and without Metal Artifact Reduction in head CT from two different vendors

Abstract

Abstract Introduction: Patients evaluated with diseases in the head are often performed with non-contrast head computed tomography (CT) as first-line imaging. Several reconstruction algorithms have been introduced over the past years to overcome challenges with image noise and artifacts. Deep learning reconstruction (DLR) has been introduced to improve image quality while maintaining lower radiation doses. Metal implants can introduce metal artifacts, and therefore, metal artifact reduction (MAR) algorithms are applied to overcome challenges with artifacts from metal implants. Purpose: This study aims to compare the effect of DLR with and without MAR from two vendors (Canon/GE) in head CT images with titanium and steel aneurysm clips using standard and low-dose protocols. Methods: Two skulls filled with gelatin and positioned with two different aneurysm clips; titanium and steel, scanned using two vendor-specific MAR software applications. Visual grading analysis (VGA) was done by three radiologists, and objective analyses were performed measuring contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR). Results: DLR applied with MAR reduced artifacts and improved image quality in images with steel aneurysm clips. Applying MAR on titanium resulted in decreased CNR using standard head CT protocol. Low-dose head CT protocol showed higher CNR for images with Canon and not with GE. VGA for titanium with MAR showed lower scores for images with GE, but higher scores were observed for Canon. Conclusion: Applying MAR on DLR images has a different effect on steel and titanium implants. The best effect with DLR and MAR was seen in images with steel aneurysm clips. These results may occur due to MAR introducing new artifacts when applied to titanium implants. Implications for practice: The results underscore that knowledge about implant material is crucial when applying MAR in CT.