Convolutional neural network-based metal and streak artifacts reduction in dental CT images with sparse-view sampling...

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Sparse-view sampling in dental cone-beam computed tomography (CBCT) reduces scan time and radiation dose.
• Deep learning methods improve image quality by reducing streak artifacts in sparse-view CT.
• Existing methods often fail to adequately address artifacts caused by metal implants.

Purpose of the Study:

• To propose a novel framework for reconstructing high-quality CBCT images from sparse-view data with metal implants.
• To simultaneously reduce metal artifacts and streak artifacts in sparse-view CT.
• To overcome the limitations of current deep learning techniques in handling metal artifacts.

Main Methods:

• A framework combining normalized metal artifact reduction (NMAR) with two convolutional neural networks (CNNs).
• The first CNN reduces initial artifacts and preserves details for NMAR.
• The second CNN further reduces streak artifacts post-NMAR processing.

Main Results:

• The proposed method effectively reduced both metal and streak artifacts.
• Anatomical structure details were preserved.
• Performance was validated using computer simulations with XCAT and clinical datasets.

Conclusions:

• The developed framework enables accurate CT image reconstruction in metal-inserted sparse-view CBCT.
• It successfully reduces artifacts from metal objects and sparse-view sampling.
• The method shows promise for feasible fast-scan dental CBCT imaging.