Metal Artifacts Reduction in CT Scans using Convolutional Neural Network with Ground Truth Elimination.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Computational Pathology

Background:

• Metal implants are prevalent in patients, leading to significant streak artifacts in Computed Tomography (CT) scans.
• These artifacts severely degrade image quality, potentially compromising accurate clinical diagnosis.
• Existing supervised learning methods require artifact-free images for training, which are unavailable in real-world clinical scenarios.

Purpose of the Study:

• To develop a Convolutional Neural Network (CNN) based method for reducing metal streak artifacts in hip CT scans.
• To eliminate the necessity of artifact-free images during the model training phase for clinical applicability.
• To enhance the diagnostic utility of CT scans in patients with metal implants.

Main Methods:

• A novel CNN architecture was designed to address metal artifact reduction (MAR) in CT images.
• The model was trained using CT scans from anatomical regions near the hip, circumventing the need for clean hip implant data.
• The proposed method focuses on artifact suppression while preserving crucial anatomical details.

Main Results:

• The developed method successfully suppressed streak artifacts in corrupted CT hip scans.
• Significant improvements in overall image quality were achieved post-artifact reduction.
• The method demonstrated preservation of surrounding tissue details, crucial for diagnostic interpretation.
• The approach yielded artifact-free images with high fidelity on clinical data from multiple patients.

Conclusions:

• The proposed CNN-based MAR method effectively reduces metal artifacts in hip CT scans.
• This technique offers a clinically viable solution by not requiring artifact-free training data.
• The method enhances CT image quality and preserves diagnostic information in the presence of metal implants.