Deep learning denoising and deep learning-augmented contrast enhancement improve ultra-high-resolution photon-counting-detector CCTA image quality.
Authors
Affiliations (5)
Affiliations (5)
- Department of Neuroradiology, University Medical Center Mainz, Johannes Gutenberg University, Mainz, Germany.
- Department of Radiology, University Medical Center Mainz, Johannes Gutenberg University, Mainz, Germany.
- Department of Cardiology, University Medical Center Mainz, Johannes Gutenberg-University, Mainz, Germany.
- Division of Cardiovascular Imaging, Department of Radiology and Radiological Science, Medical University of South Carolina, Charleston, SC, USA.
- Department of Radiology, University Medical Center Mainz, Johannes Gutenberg University, Mainz, Germany. [email protected].
Abstract
Ultrahigh-resolution (UHR) coronary computed tomography angiography (CCTA) using a photon-counting detector (PCD) provides excellent spatial resolution but is limited by increased image noise. We evaluated whether deep learning (DL)-based denoising and contrast enhancement can further improve UHR-CCTA image quality. UHR-CCTA scans from 25 patients with coronary artery disease were retrospectively analyzed. Images were reconstructed at 0.2-mm slice thickness without and with quantum iterative reconstruction (QIR, Q<sub>off</sub>-Q<sub>4</sub>), and subsequently processed using a DL tool for denoising and increasing iodine contrast enhancement. Subjective image quality (sharpness, noise, vessel wall delineation, and overall quality) was assessed by three radiologists using a 5-point Likert scale. Objective analysis included signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and coronary wall sharpness. Differences between image sets were analyzed using Wilcoxon signed-rank tests. DL-based CCTA demonstrated significantly superior subjective image quality across all parameters compared to standard reconstructions, with an overall median image quality ([interquartile interval, IQR]: 4 [3-5] versus 2 [2-3], p < 0.001). SNR and CNR were significantly higher for DL-based CCTA across all QIR levels (all p < 0.001). The highest values were observed for DL-based CCTAQ<sub>4</sub> versus UHR-CCTAQ<sub>4</sub> (SNR: 258.0 ± 147.5 versus 145.2 ± 50.1; CNR: 25.8 ± 17.1 versus 11.1 ± 4.5; both p < 0.001). Coronary wall sharpness was improved for DL-based CCTA for Q<sub>1-4</sub> (all p < 0.027). DL-based denoising and iodine contrast enhancement effectively reduce noise and improve anatomic detail in UHR-CCTA, providing incremental value beyond vendor-specific reconstruction, particularly at high QIR strength levels. Question Deep learning reduces noise and enhances contrast in ultrahigh-resolution coronary CT angiography. Findings Image quality improves beyond standard iterative reconstruction methods, and vessel wall sharpness and diagnostic clarity significantly increase. Relevance statement This study aimed to determine at which levels of vendor-specific iterative reconstruction, DL-based denoising and contrast enhancement perform best in UHR CCTA, providing a foundation for future investigations on how these technical improvements may translate into diagnostic benefits.