Metal artifact reduction for practical dental computed tomography by improving interpolation-based reconstruction...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Dental Radiology

Background:

• Metal artifacts are a common issue in dental computed tomography (CT) leading to image corruption.
• Existing metal artifact reduction (MAR) techniques often introduce new artifacts or deform structures due to interpolation inaccuracies.

Purpose of the Study:

• To develop a novel deep learning-based strategy for effective metal artifact reduction in dental CT.
• To improve the accuracy of reconstructed images by minimizing artifacts caused by metal implants.

Main Methods:

• A three-step MAR method was developed, starting with coarse reconstructions from linearly interpolated data.
• A deep learning network was trained on simulated data to correct interpolation errors and recover nonmetal region information.
• Region of interest (ROI) reconstruction of metal regions was incorporated, with the network generalizing from simulated to real-world data.

Main Results:

• The deep learning network successfully reduced artifacts and accurately recovered tooth structures in both simulated and real patient datasets.
• Significant improvements were observed in relative root mean square error and structure similarity index measures.
• Experienced dentists provided positive evaluations of the method's effectiveness.

Conclusions:

• A transferable deep learning strategy was established for metal artifact reduction in practical dental CT systems.
• The method reframes MAR as an interpolation-artifact reduction problem, enabling effective learning from simulations.
• The proposed approach is easily implemented, computationally efficient, and demonstrates practical applicability.