Metal artefact reduction in the oral cavity using deep learning reconstruction algorithm in ultra-high-resolution...

Area of Science:

• Medical Imaging
• Radiology
• Dental Imaging

Background:

• Metal artefacts pose a significant challenge in dental CT imaging, potentially obscuring details and affecting diagnostic accuracy.
• Optimizing metal artefact reduction (MAR) algorithms is crucial for improving image quality in ultra-high-resolution CT (UHRCT) of the oral cavity.

Purpose of the Study:

• To evaluate the impact of acquisition and reconstruction parameters on the effectiveness of MAR algorithms in UHRCT for oral cavity applications.
• To determine the optimal settings for enhancing MAR performance in dental imaging.

Main Methods:

• A mandible tooth phantom was scanned using super-high-resolution, high-resolution (HR), and normal-resolution (NR) modes on a UHRCT scanner.
• Images were reconstructed using deep learning-based reconstruction (DLR) and hybrid iterative reconstruction (HIR) with MAR.
• Radiologists assessed metal artefact severity and lesion shape reproducibility; quantitative analysis included signal-to-artefact ratio (SAR), CT number accuracy, and image noise.

Main Results:

• High-resolution mode with DLR (HR_DLR) demonstrated significantly reduced metal artefact severity (4.6 ± 0.5 vs. 2.6 ± 0.5) and improved lesion shape reproducibility (4.5 ± 0.5 vs. 2.9 ± 1.1) compared to normal-resolution with HIR (NR_HIR).
• HR_DLR achieved a superior SAR (4.9 ± 0.4 vs. 2.1 ± 0.2), lower absolute percentage error in CT number (0.8% vs. 23.8%), and reduced image noise (15.7 ± 1.4 vs. 51.6 ± 15.3) compared to NR_HIR.
• All quantitative and qualitative improvements were statistically significant (p < 0.05).

Conclusions:

• The combination of HR mode and DLR significantly enhances the performance of MAR algorithms in UHRCT for the oral cavity.
• This optimized approach improves image quality, reduces artefacts, and increases diagnostic confidence in dental imaging.