Deep learning-based metal artifact reduction in CT for total knee arthroplasty
Jimin Lee1,2,3, Hee-Dong Chae4,5, Hyungjoo Cho1
1Program in Biomedical Radiation Sciences, Department of Transdisciplinary Studies, Graduate School of Convergence Science and Technology, Seoul National University, 1 Gwanak-ro, Gwanak-gu, Seoul, 08826, Republic of Korea.
Scientific Reports
|November 12, 2025
Summary
A new deep learning technique, KMAR-Net, significantly improves metal artifact reduction (MAR) in CT scans after total knee arthroplasty (TKA). This AI-powered method enhances image quality for better postoperative evaluation.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence in Medicine
Background:
- Metal artifacts in CT scans pose challenges for evaluating patients with total knee arthroplasty (TKA).
- Existing metal artifact reduction (MAR) methods have limitations in clinical practice.
Purpose of the Study:
- To develop and evaluate a deep learning (DL)-based MAR technique, KMAR-Net, specifically for postoperative TKA CT scans.
- To compare the performance of KMAR-Net against conventional MAR methods.
Main Methods:
- Developed KMAR-Net using simulated CT images generated via sinogram handling.
- Quantitative analysis involved measuring artifact area, mean attenuation, and standard deviation.
- Qualitative analysis used visual grading for artifact severity, bone conspicuity, and soft tissue visualization.
- A phantom study validated KMAR-Net's performance under controlled conditions.
Main Results:
- KMAR-Net demonstrated superior reduction in artifact area, mean attenuation, and standard deviation compared to the projection-completion method.
- Qualitative analysis showed KMAR-Net outperformed O-MAR in overall artifact reduction and soft tissue evaluation.
- One reader found KMAR-Net superior for bone conspicuity (P=0.080), while the second reader strongly agreed (P<0.001).
Conclusions:
- Deep learning-based KMAR-Net offers significant advantages for metal artifact reduction in postoperative TKA CT imaging.
- KMAR-Net provides superior image quality compared to conventional MAR techniques, aiding clinical assessment.
