According to statistics, cardiovascular diseases are a leading cause of death worldwide. Coronary computed tomography angiography (CCTA) can clearly display the coronary arteries, accurately detect coronary plaques, and properly assess coronary lesions. With a negative predictive value close to 99% for coronary artery disease, CCTA has become an indispensable diagnostic tool for patients with cardiac symptoms.
The motion and deformation of the coronary arteries are uneven, and motion artifacts occur when the speed of movement exceeds the time resolution of the CT scanner. Motion artifacts can blur the coronary artery images, making it impossible to assess or accurately diagnose coronary stenosis. Since the motion direction of the mid-right coronary artery (mRCA) is perpendicular to the scanning plane, the in-plane motion is the highest, and the motion artifacts are the most severe. Current methods to reduce CCTA motion artifacts include using beta-blockers to control heart rate (HR), shortening image acquisition time, applying single heartbeat scanning, developing ECG editing, multi-segment reconstruction, and software solutions for intravascular motion correction algorithms. Although these methods can prevent and correct motion artifacts, the effectiveness of these methods is somewhat affected by heart rate, which limits their clinical applicability to some extent.
Generative Adversarial Networks (GANs) are an innovative neural network that has been introduced into the field of medical imaging. GAN consists of a generator network and a discriminator network. These two networks compete with each other to optimize network parameters and generate new images.
Recently, a study published in the journal European Radiology used GAN to correct motion artifacts representing the maximum motion of the coronary segment in mRCA and evaluated the image quality and diagnostic performance of GAN-generated images, providing technical support for improving CCTA image quality.
This study included 313 patients who underwent CCTA scans, each with paired motion-affected and motion-free reference images at different R-R intervals within the same cardiac cycle, along with an additional 53 CCTA cases compared to invasive coronary angiography (ICA) images. Pix2pix is an image-to-image translation GAN trained on motion-affected and motion-free reference pairs, generating motion-free images from motion-affected ones. Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM), Dice Similarity Coefficient (DSC), and Hausdorff Distance (HD) were calculated to assess the quality of GAN-generated images.
At the image level, the median PSNR, SSIM, DSC, and HD of GAN-generated images were 26.1 (interquartile range: 24.4-27.5), 0.860 (0.830-0.882), 0.783 (0.714-0.825), and 4.47 (3.00-4.47), significantly better than the motion-affected images (P <0.001). At the patient level, the results for image quality were similar. The motion artifact mitigation score (4 points vs. 1 point, p < 0.001) and overall image quality score (4 points vs. 1 point, p < 0.001) of GAN-generated images were improved compared to motion-affected images. In patients compared with ICA, the accuracy of GAN-generated images in identifying no stenosis, <50%, and ≥50% stenosis was 81%, 85%, and 70%, respectively, higher than the motion-affected images’ 66%, 72%, and 68%.
Figure shows GAN-generated images compared to standard invasive coronary angiography diagnostic performance images. A 65-year-old man suspected of having coronary artery disease underwent CCTA and invasive coronary angiography. The heart rate during CT scanning was 55 bpm; heart rate variability was 4 bpm; the optimal phase was 75%. A and C axial CCTA images show calcified plaques in the mid-right coronary artery, corresponding to two 50% stenoses shown in invasive coronary angiography. b. The motion-affected CT images were blurred, making it difficult to determine the degree of stenosis. After GAN correction, the GAN-generated images could determine the degree of lumen stenosis. GAN, Generative Adversarial Network.This study developed and tested a GAN algorithm to correct images with motion artifacts in CCTA. Quantitative assessments and subjective evaluations showed that GAN-generated images significantly reduced motion artifacts and markedly improved image quality and diagnostic accuracy.
Original article source: Lu Zhang, Beibei Jiang, Qiang Chen, et al. Motion artifact removal in coronary CT angiography based on generative adversarial networks.
For authorized reprints and tips, please contact Meis Medical Administrator.
These days, due to changes in the public account’s push rules, the timeline has been disrupted. If you haven’t starred us or often shared “Looking”, our pushes may be drowned in your vast subscription list.
The editor is particularly afraid of losing you, so we need you to mark Cardiovascular New Frontiers as a star, and each time you finish reading an article, pleaselight up “Looking” orshare it, so we can hold on tight and not get lost!
↓Click below “Read Original“, download the Meis Medical APP!