Diagnostic Value of Stent Images using Ultra-High Resolution Deep Learning Algorithms in Coronary CT Angiography

Recently, with the advancement of artificial intelligence image reconstruction technology, we are attempting to evaluate the accuracy of coronary artery stent lumen by developing ultra-high-resolution deep learning image reconstruction technology based on the most accurate 0.25mm multi-detector available. This study was divided into patient studies and phantom studies. Images were reconstructed in the same patient using four different methods: filter correction reverse projection, repetitive reconstruction, deep learning reconstruction, and artificial intelligence ultra-high resolution reconstruction. To objectively evaluate image quality, four parameters, image noise, CT density, signal-to-noise ratio, and contrast-to-noise ratio, were analyzed for each dataset. The phantom study installed a latex Nelaton sterilized catheter to express the coronary artery in the dummy chest phantom. Subsequently, a coronary stent (2.5mm and 3.5mm in diameter) was inserted into each catheter. Images were obtained using various reconstruction techniques (FBP, IR, DLR, and UHR-DLR) for each CT manufacturer. Stent clarity was evaluated using edge rise distance (ERD) and edge rise slope (ERS). We showed that the average Signal-to-Noise Ratio (SNR) and Contrast-to-Noise Ratio (CNR) of coronary arteries reconstructed with Ultra-High Resolution Deep Learning Reconstruction (UHR-DLR) improved by 29.2% and 36.4% compared to those reconstructed with DLR, respectively. Image noise was reduced by about 21.4% in UHR-DLR. The ERS average value of the coronary stent phantom measured by the CT manufacturers was highest in the UHR-DLR of Aquilion ONE Prism (Canon), and the ERD value is statistically significant for each CT manufacturer. Compared to FBP, HIR, and DLR, UHR-DLR not only improved image clarity and reduced image noise and blooming artifacts but also clearly showed the stent support structure and stent lumen.