Influence Of Iterative Reconstruction Algorithm On The Stability Of FFR_CT Measurements Based On Machine Learning:A Reproducible Study

Objective The FFR_CT value is measured by iterative reconstruction algorithm with different iterative intensity based on machine learning,and the fluctuation law and influencing factors are analyzed through the change of the measured value,the influence of iterative reconstruction intensity on the stability of SAFIRE value is clarified,and the iterative reconstruction level with the best image quality and the lowest image noise is discussed,which provides a theoretical basis for optimizing the FFR_CT reconstruction algorithm level in coronary artery imaging.Methods A total of 50 patients who completed CCTA examination in the affiliated Hospital of Inner Mongolia Medical University from September to October 2021 were collected retrospectively.Image data are reconstructed with FBP and SAFIRE.Five different intensity factors are used to reconstruct the SAFIRE iterative reconstruction sequence,and finally six data sets are obtained.According to the opacity of vascular boundary,discontinuity of structure,outline and overall image impression,the two observers independently evaluated the image quality and image impression of each coronary artery segment by 4-point method.The region of interest（ROI）was measured at the root of the aorta and the adipose tissue around the coronary artery at the opening of the left main trunk for three times.The average values of three times were taken to record image noise（SD）,CT value,signal-to-noise ratio（SNR）and contrast-to-noise ratio（CNR）.150 blood vessels of 50 patients were uploaded to Shu Kun software for FFR_CT analysis.Epicardial coronary artery FFR_CT≤0.80 is defined as coronary artery stenosis with significant hemodynamics.The objective noise and subjective image quality scores of the 6 groups were compared by analysis of variance（one-way ANOVA）.First of all,the homogeneity test of variance was carried out,and the pairwise comparison between the two groups was made by using the method of minimum significant difference（least significant difference,LSD）or Dunnett T3 method（uneven variance）.Pearson correlation coefficient/Spearman rank correlation coefficient（r）was used to evaluate the correlation between factors and FFR_CT values,as well as the correlation between factors and intensity.The ROC curve was drawn to quantify the influencing factors.The boundary of FFR_CT value was0.8,the result of FFR_CT≥0.8 was set as 1 and<0.8 was set as 0.The indexes of ROC curve were obtained,including area under curve（area under the curve,AUC）,tangent point and P value.Results The two different reconstruction methods produced different FFR_CT values,but did not achieve a statistically significant difference.In 31（62%）of the 50 patients in this study,at the vessel level,the FFR_CT values of 41（27.3%）vessels showed very different results in the SAFIRE sequence than in the FBP sequence.Compared with FBP,the subjective image quality and objective image quality of iterative reconstruction sequence show statistically significant differences.In terms of subjective image quality,among the six groups of images obtained by different reconstruction methods,the image quality score increased successively with the increase of iterative intensity（r=0.33）.The difference of image quality score between FBP（SAFIRE0）and SAFIRE1,SAFIRE2,SAFIRE3,SAFIRE4,SAFIRE5 was statistically significant,which further confirmed that the iterative reconstruction algorithm could improve the strong performance of image quality.However,there was no significant difference in image quality scores of SAFIRE3,SAFIRE4 and SAFIRE5 among the groups.In terms of objective image quality,among the 6 groups of images obtained by different reconstruction methods,with the increase of iterative intensity,the image noise decreased in turn（r=-0.844）,CNR and SNR increased in turn（r=0.718、r=0.661）.From the comparison between groups,there were significant differences in image noise,CNR and SNR between FBP（SAFIRE0）and SAFIRE2,SAFIRE3,SAFIRE4 and SAFIRE5,but there was no significant difference in image noise,CNR and SNR between SAFIRE1 and FBP.This shows that FBP and SAFIRE1 have the same ability to reduce image noise and improve image contrast,but the difference in reducing noise is not significant.Compared with FBP,the application of iterative reconstruction technology can reduce image noise,improve image contrast and improve image quality.In addition,the subjective image score of SAFIRE5 grade showed a similar"plastic-like"appearance,and the FFR_CT value also showed a downward trend.In the ROC curve,the maximum difference of image quality score was statistically significant,and the efficiency was higher（AUC=0.872,P<0.001）.The results show that the maximum difference of subjective image quality score is the main factor affecting the change of FFR_CT value.Conclusion This study shows that there is no statistically significant difference in the FFR_CT values generated by FBP and iterative reconstruction algorithms at different iterative intensity levels,but there are still some patients whose FFR_CT values have changed,resulting in inconsistent qualitative results,resulting in Diagnostic error classification.Through our research,we find that the use of different SAFIRE iterative intensity in CCTA will lead to significant statistical changes in subjective and objective image quality indicators.It is precisely because different iterative intensity levels cause changes in subjective image quality,resulting in differences in subjective image quality scores,resulting in misclassification of FFR_CT qualitative diagnosis.