The Value of Deep Learning-based Reconstruction Algorithms for Improving the Quality of Pulmonary Vein CT Images
Objective:To investigate the improvement of image quality and diagnostic confidence in CT images of the left atrium and pulmonary veins with deep learning image reconstruction(DLIR)com-pared to conventional adaptive iterative reconstruction algorithms(ASiR-V).Methods:Thirty-one pa-tients with CT imaging of the left atrium and pulmonary veins were retrospectively included,and their raw data were reconstructed using five reconstruction algorithms:filtered back projection(FBP),30％ASiR-V,70％ASiR-V,DLIR-M(medium)and DLIR-H(high),respectively,to measure CT values of the left atrium and calculate the corresponding background noise(Standard Deviation,SD)values,signal-to-noise ratio(SNR)values and contrast-to-noise ratio(CNR)values were calculated.The image quality of each of the five sets of reconstructed images was evaluated by two radiologists who had worked for many years.Results:Objective evaluation:The differences in SD,CNR and SNR values in the left atrium were statistically significant(P＜0.001)in all five groups.DLIR-H was the best,with no significant differences between FBP and 30％ASiR-V,DLIR-M and 70％ASiR-V.The SD values decreased and the SNR and CNR values increased with increasing DLIR reconstruction grade.Subjective evaluation:The subjective evaluation of image quality by the two radiologists was consistent(kappa value of 0.814),with DLIR-H showing the best subjective image quality score.Conclusion:DLIR is more effective in noise reduction and has better image quality than the FBP and ASiR-V algorithms.
deep learningiterative reconstructionfiltered back-projectionimage quality
NETL
NSTL
万方数据
