This retrospective analysis included 215 participants from 1st January to 31st July 2022, who had a mean age of (66.09 ± 11.72) years, including 123 males and 92 females. These data were collected for the purpose of this study and the participants underwent CAG and chest-CT plain film scanning at China National Nuclear Corporation 416 Hospital: Chengdu Medical College Second Affiliated Hospital. The total population was divided into two groups based on the results of CAG: the CAD group, comprising 97 cases with ≥ 50% stenosis of the coronary artery, and the non-CAD group, comprising 118 cases with coronary artery stenosis < 50% or no coronary artery stenosis. Among these participants, 63 males were in the CAD group, while 60 were in the non-CAD group, and 34 females were in the CAD group, with 58 in the non-CAD group.

Exclusion criteria: (1) participants without complete medical history, biochemical test results, and chest-CT data. (2) Participants with a history of cardiac surgery, including coronary artery bypass grafting and cardiac valvuloplasty. (3) Participants with a history of severe cardiac valvular disease, cardiomyopathy, and pericardial effusion. (4) Participants with acute infections. (5) Participants with autoimmune diseases.

Statement of assurance: (1) The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki. (3) The study protocol has been priorly approved by the Institution’s ethics committee on research on humans.

An ICT Brilliance 128-row 256-slice CT scanner (Philips, The Netherlands) was used for this study, employing a tube voltage of 120 kVp, a tube current of 220 mA, a pitch of 1.375, a scanning thickness of 2.5 mm, and an image matrix of 512 × 512. Both standard algorithmic and non-algorithmic reorganization of the images were performed, with a reorganization layer thickness and spacing between layers set at 1.25 mm each. The images were analyzed using a soft tissue window with a window width of 350 HU and a window position of 35 HU. The scanning process covered the region from the thoracic inlet to the bottom of the lungs, with the patient completing the scan while holding their breath after one deep inspiratory breath. Chest-CT scans with thin-layer reconstruction were performed in this study without ECG gating, and one week after the chest-CT, CAG was conducted.

CAG was conducted via the Seldinger puncture method with transradial artery access. Based on the American Heart Association Coronary Angiography Stenosis Evaluation Criteria [12], patients with ≥ 50% coronary stenosis were classified into the coronary CAD group, while those with < 50% coronary stenosis or no coronary stenosis formed the non-CAD group. Furthermore, the left ventricle was divided into subgroups based on gender, consisting of the male CAD group and non-CAD group, and the female CAD group and non-CAD group.

Age was calculated from the identity card number, while height, weight, and blood pressure measurements were taken by the cardiology nurse. The nurse also conducted tests for C-reactive protein (CRP), total cholesterol (TC), triglyceride (TG), fasting plasma glucose (FPG), glycosylated hemoglobin Alc (HbA1c), uric acid (UA), lipoprotein A (LpA), and apolipoprotein a (ApoA). In addition, the laboratory examined other biochemical indexes, including TC, TG, FPG, HbA1c, UA, LpA, and ApoA. Left ventricular ejection fraction (LVEF) was determined through echocardiography, while the degree of coronary artery stenosis was based on cardiologists’ post-CAI records and clinical diagnosis.

The enrolled patients underwent Philips 128-row ICT chest scanning, and axial images were selected at the level of the bifurcation of the main trunk of the thin left coronary artery and saved in DICOM (Digital Imaging and Communications in Medicine) format. ImageJ (Image J, version 1.3.4.11. National Institutes of Health, Bethesda, Maryland) software was used to set the image threshold from – 250 HU to – 50 HU, while carefully avoiding pericardial fat along the visible cardiac fibrous membrane [13]. The region of interest (ROI) was obtained by manually outlining the edge of the EAT [14] (Fig. 1). First-order features were outlined, measured by two associate radiologists, each with over 20 years of experience in radiology (Fig. 2) were classified into two groups based on their significance (Table 1): first-order features i.e., gray-scale histogram parameters(Mean, StdDev, Mode, IntDen, Median, Skew, Kurt, RawIntDen, AR, Solidity), and shape (Area, BSA index, Perim, Circ, Area%, Round). The were then averaged, and grayscale values, except Skew, were taken as absolute values. Due to the nature of the measurements being parameters of fat, many feature values are negative. However, the negative sign in adipose tissue has a different meaning compared to the algebraic negative sign. This discrepancy can influence calculations, potentially resulting in contrary statistical outcomes. Therefore, with the exception of skewness, all negative values have been converted to their absolute values. BSA index were calculated by dividing the EAT area by the BSA value, which was obtained using the “Chinese general formula” based on height and weight: BSA = 0.0061 × height(cm) + 0.0124 × weight(kg) – 0.0099 [15].

EAT left coronary bifurcation level outlining ROI. A Sixty-four -year-old male, LCX stenosis 60%; a EAT left-main coronary bifurcation level. B:Yellow line represents manually outlined EAT, red within yellow line is EAT area

Gray-scale histogram of the EAT ROI. A Sixty-four -year-old male with 60% LCX stenosis.Gray-scale histogram waveform peaks to the left. B Fifty-nine-year-old male coronary artery without stenosis, gray-scale histogram waveform peaks to the right (Threshold range of less than Min-250 is due to software value interval)

CT images of 40 patients were randomly selected for intra- and inter-group consistency assessment by two associate radiologists. In the intra-group consistency assessment, the first radiologist outlined the ROI and obtained the radiomics feature following specific steps. Afterward, the same method and steps were repeated within a 2-week interval, and the radiomics features of the 40 patients were extracted from both sets of data for intra-group correlation analysis. For the inter-group correlation assessment, the second radiologist outlined the CT images of the same 40 patients using the same methods and steps as the first radiologist. The radiomics feature obtained by the second radiologist were then compared with the ones extracted by the first radiologist for the initial analysis. The results showed a mean intra-group correlation coefficient (ICC) value of 0.93 (ranging from 0.785 to 1.000, p < 0.001) and a mean inter-group correlation coefficient of 0.91 (ranging from 0.732 to 1.000, p < 0.001).

SPSS 19.0 software was utilized for data analysis. For metric data that followed a normal or near-normal distribution, the mean ± standard deviation was used for expression, and a t test was employed to compare between groups in the study design. Meanwhile, count data was presented using frequency and percentage, and the comparison between groups was conducted using the Chi-square test. Furthermore, clinical indicators and textural parameter indexes that exhibited statistically significant differences in the inter-group comparisons were depicted on the subjects’ operating characteristic ROC. For the clinical indicators and radiomics parameter indicators with statistically significant differences between the groups, ROC curves were plotted, and the area under the ROC curve (AUC) was calculated. Statistical significance was set at p < 0.05.