Removing ring artifacts in CBCT images via Transformer with unidirectional vertical gradient loss

Area of Science:

• Medical Imaging
• Image Processing
• Artificial Intelligence

Background:

• Cone-beam computed tomography (CBCT) is vital for disease detection but suffers from ring artifacts due to detector inconsistencies.
• These artifacts compromise diagnostic accuracy, necessitating effective correction methods.
• Existing artifact removal techniques have limitations that this study addresses.

Purpose of the Study:

• To develop a Transformer-based model for robust ring artifact correction in CBCT images.
• To leverage the global and local feature modeling capabilities of Transformers for artifact removal.
• To enhance diagnostic image quality by effectively eliminating artifacts without significant detail loss.

Main Methods:

• A Transformer model was designed incorporating dual-domain information fusion (image and polar domains).
• A specialized loss function was developed, including an unidirectional gradient loss in the polar domain.
• The method operates in the image domain for direct artifact-free output prediction and detail preservation.

Main Results:

• Extensive experiments validated the proposed method's effectiveness, outperforming classical and state-of-the-art techniques.
• The method achieved superior image quality and evaluation metrics in artifact correction.
• Ablation studies confirmed the contribution of individual components of the proposed approach.

Conclusions:

• The Transformer-based method significantly enhances ring artifact correction in CBCT images.
• Dual-domain fusion and a customized loss function enable effective artifact removal and detail preservation.
• The approach offers a promising solution for improving the reliability of CBCT imaging.