Deep learning image reconstruction optimizes coronary artery calcium quantification.

OBJECTIVE

To investigate the effects of deep learning reconstruction (DLR) on the image quality and quantification of coronary artery calcium (CAC).

MATERIALS AND

METHODS

Patients who underwent calcium scoring and coronary CT angiography examinations were retrospectively collected. The images of calcium scoring were reconstructed using filtered back projection (FBP), hybrid iterative reconstruction (HIR) and DLR algorithms. The CT value, image noise, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of aortic root and left ventricle were compared in three algorithms. Two radiologists scored the subjective image quality using a four-point scale. The quantification of CAC (Agatson score, calcium volume and mass) in FBP, HIR and DLR were calculated by automatic software. The risk classification of CAC were evaluated according to the Agatston score.

RESULTS

In objective image quality, compared with FBP and HIR, DLR significantly reduced image noise and improved SNR (all p < 0.05) without changing the CT value of aortic root and left ventricle (all p > 0.05). DLR received significantly higher subjective scores (3.80 ± 0.40) than HIR (3.48 ± 0.50) and FBP (2.36 ± 0.48) (both p < 0.001). In calcium quantification, the Agatston score, calcium volume and mass were no significant difference among the three algorithms (all p > 0.05). In risk classification analysis, DLR reduced the number of reclassification compared with HIR.

CONCLUSION

DLR enhances the image quality and consistency of CAC quantification compared with FBP and HIR. Besides, DLR reduced risk reclassification relative to HIR.