The Predictive Value Of Liver Tissue Radiomics Signature Combined With Coronary Multi-Parameter Model Based On CCTA For Acute Coronary Syndrome

Objects: To evaluate the predictive value of liver tissue radiomics signature combined with coronary multi-parameter model based on coronary computed tomography angiography(CCTA)for future acute coronary syndrome(ACS)within 3 years in patients with suspected coronary heart disease.Methods: Patients with suspected coronary heart disease who received CCTA in our hospital from January 2013 to December 2020 were retrospectively collected.All enrolled patients had complete clinical data and follow-up records for more than 3 years.Patients with ACS during 3 years follow-up after CCTA examination were selected as ACS group(n=104).Patients without ACS during 3 years follow-up after CCTA examination were 1:1matched to ACS group according to age,sex,body mass index and traditional cardiovascular risk factors,then selected as control group(n=104).All patients were randomly divided into training and testing datasets with a ratio of 8:2.Region of interests(ROIs)with the diameter of about 2cm were manually delineated in the maximum liver layer scanned by CCTA image in Intellispace Discovery of Philips Healthcare.The best liver radiomics features were extracted and screened to construct Liver Radiomics Model(Model 1)based on automatic machine learning.The quantitative and qualitative characteristics of culprit plaques in ACS group and the plaques with the highest degree of stenosis in the control group were evaluated,and Fat attenuation index(FAI)around the plaques was measured.The best predictors were screened by the method of automatic feature selection to construct the Coronary Multi-parameter Model(Model 2).The Integrated Model(Model 3)was constructed based on the average of Model 1 and Model2.Receiver operating characteristic(ROC)curves were drawn to compare the predictive ability of different models for future ACS within 3 years in patients with suspected coronary heart disease,and the Area under the curve(AUC),sensitivity,specificity and accuracy were calculated.Decision curves were drawn to compare the clinical application value of different models.All statistical analyses were performed with R software(version3.5.1;http://www.Rproject.org).P-values < 0.05 were considered as statistically significant.Results: ROC curves showed that Liver Radiomics Model(Model 1)(AUC=0.692[95%CI:0.531-0.852])had slightly lower predictive efficacy than Coronary Multiparameter Model(Model 2)(AUC=0.749 [95%CI:0.600-0.898])in predicting ACS events within 3 years in patients with suspected coronary heart disease,and there was no statistical difference in the predictive efficacy between the two models(P=0.59).Compared to Model2,Integrated Model(Model 3)(AUC=0.771 [95%CI:0.627-0.915])had slightly improved predictive performance,but there was no statistical difference between the two models(P=0.79).Decision curves showed that the clinical benefit of Integrated Model was higher than that of Liver Radiomics Model and Coronary Multi-parameter Model.Conclusion: Liver radiomics features based on CCTA could provide additional predictive information for the occurrence of future ACS.The Integrated model based on Liver Radiomics model and Coronary Multi-parameter model had a good predictive ability for the occurrence of future ACS within 3 years in patients with suspected coronary heart disease,which could provide certain incremental predictive value for Coronary Multiparameter model and contribute to risk stratification for ACS patients.