Artifact correction in low-dose dental CT imaging using Wasserstein generative adversarial networks

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Dental computed tomography (CT) poses health risks due to high-dose x-ray radiation.
• Low-dose CT (LDCT) is a focus, but downsampling can introduce noise and artifacts, compromising image quality.
• Artifact correction is crucial for reliable LDCT diagnostics.

Purpose of the Study:

• To propose a deep learning-based artifact correction method for downsampling CT reconstruction in dental imaging.
• To develop an algorithm that effectively removes noise and artifacts from low-dose dental CT scans.
• To improve the diagnostic quality of dental CT images obtained with reduced radiation exposure.

Main Methods:

• Utilized clinical dental CT data with low-dose artifacts as input.
• Trained a generative adversarial network (GAN) with Wasserstein distance (WGAN) and mean squared error (MSE) loss, termed m-WGAN.
• Employed high-quality normal-dose CT data for supervised training to remove artifacts.

Main Results:

• The proposed m-WGAN algorithm effectively removed low-dose artifacts from dental CT scans.
• Demonstrated superior efficiency in noise removal compared to existing approaches.
• Quantitative and qualitative analyses showed m-WGAN outperformed general GAN and CNNs in artifact correction and texture preservation.

Conclusions:

• The m-WGAN method significantly improves dental CT image quality as a postprocessing technique.
• This represents the first deep learning architecture applied to artifact correction for commercial cone-beam dental CT.
• The validated artifact correction performance offers a promising new direction for low-dose dental CT research.