A Metal Artifact Reduction Method Using a Fully Convolutional Network in the Sinogram and Image Domains for Dental...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Dental Technology

Background:

• Dental computed tomography (DCT) image quality is compromised by metal artifacts from implants.
• These artifacts, caused by beam hardening and streak effects, reduce diagnostic accuracy.
• Existing interpolation-based methods are often insufficient for DCT due to high X-ray attenuation from teeth and metals.

Purpose of the Study:

• To develop and evaluate an effective metal artifact reduction (MAR) method for DCT.
• To investigate a fully convolutional network (FCN)-based approach in both sinogram and image domains.
• To compare the proposed MAR method against existing techniques for artifact reduction.

Main Methods:

• A novel MAR method utilizing a fully convolutional network (FCN) was developed.
• The method involves metal trace segmentation, sinogram domain restoration, and image domain restoration with metal re-insertion.
• Computational simulations and experimental validation were performed to assess image quality.

Main Results:

• The proposed FCN-based MAR method significantly reduced metal artifacts in DCT images.
• The method demonstrated superior performance in reducing streak artifacts compared to normalized MAR and a sinogram-domain deep learning algorithm.
• No contrast anomalies were introduced by the proposed artifact reduction technique.

Conclusions:

• The developed FCN-based MAR method is effective for improving DCT image quality.
• This technique offers better artifact reduction performance than existing algorithms, enhancing clinical utility.
• The proposed approach successfully mitigates metal artifacts without compromising image contrast.