Impact of deep learning image reconstructions on coronary artery calcium scoring quantification and cardiovascular risk classification

ZHANG Hailun; ZHANG Qiushuang; DING Jianrong; PAN Jingli

doi:10.20258/j.cnki.1006-9011.2025.12.008

Abstract：Objective To investigate the impact of deep learning image reconstruction algorithm(DLIR)on coronary artery calcification score(CACS)quantification and cardiovascular risk classification.Methods A total of 120 patients with sus-pected coronary artery disease underwent coronary CT angiography for CACS assessment.Image reconstruction was performed us-ing four algorithms:the filtered back projection(FBP)algorithm and the low,medium,and high-intensity DLIR(DLIR-L,DLIR-M,and DLIR-H)algorithms.The reconstructed images were analyzed to compare the CT values of the aortic root and its standard deviation(SD),signal-to-noise ratio(SNR),contrast-to-noise ratio(CNR)and maximum CT value of coronary artery calcification(CAC)plaques.Furthermore,the consistency of cardiovascular event risk classification between the DLIR algo-rithms and the FBP algorithm was evaluated.Results There was no statistically significant difference in the CT values of the aortic root among the four reconstruction algorithms(P>0.05).Compared with the FBP algorithm,as DLIR intensity increased,the SD value and the maximum CT value of the calcification gradually decreased,while,the SNR and the CNR increased(both P<0.05).Pairwise comparisons of SD,SNR,CNR,and maximum CT value of the CAC among the four algorithms showed statisti-cally significant(all P<0.05).Compared with the FBP algorithm,the Agatston scores,calcification volume,and calcification mass progressively decreased with increasing DLIR intensity(all P<0.05).With the exception of the Agatston score comparison between the DLIR-L and DLIR-M and the volume comparison between the FBP and DLIR-L,all other pairwise comparisons of Agatston scores,volume,and mass showed statistically significant(all P<0.05).Compared with the FBP algorithm,the DLIR-L algorithm misclassified 5 cases,the DLIR-M algorithm misclassified 10 cases,and the DLIR-H algorithm misclassified 10 cases.Conclusion Compared with the FBP algorithm,the DLIR algorithm progressively improves image quality as the inten-sity increases but underestimates the Agatston scores,calcification volume,and calcification mass,which may result in a lower risk class for cardiovascular events.Therefore,the DLIR algorithm should be used with caution in the quantitative assessment of CACS in the clinical setting.