Comprehensive Summary
This study evaluated three deep learning (DL) models designed to reduce image artifacts caused by spinal implants in lumbar CT scans. Multi-energy CT scans from 93 patients were used to train and test models built at 70 keV, 100 keV, and a combination of both. Image quality was assessed using peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Findings showed that the multi-energy model (Modelmix) achieved the highest performance in PSNR and SSIM. Modelmix produced clearer images with fewer artifacts across a broad range of CT energy levels, while single-energy models performed well only within narrow ranges near their respective training levels.
Outcomes and Implications
CT imaging is commonly used by orthopedic surgeons to guide clinical decisions; however, the presence of metal artifacts can strongly reduce image quality. The proposed model could enhance image clarity across multiple energy levels, improving clinical interpretability and reducing the need for repeat scans. Despite its strong performance, this model remains limited by a lack of external validation and testing across a wider energy spectrum. Further research is needed before integrating this model into clinical imaging techniques.