Non-contrast CT-based radiomics nomogram of pericoronary adipose tissue for predicting haemodynamically significant coronary stenosis in patients with type 2 diabetes

Abstract

Background: Type 2 diabetes mellitus (T2DM) patients have a higher incidence of coronary artery disease than the general population. The aim of this study was to develop a radiomics nomogram of pericoronary adipose tissue (PCAT) based on non-contrast CT to predict haemodynamically significant coronary stenosis in T2DM patients.

Methods: The study enrolled 215 T2DM patients who underwent non-contrast CT and coronary computed tomography angiography (CCTA). CCTA derived fractional flow reserve (FFR_CT) ≤ 0.80 was defined as hemodynamically significant stenosis.1691 radiomics features were extracted from PCAT on non-contrast CT. Minimum redundancy maximum relevance (mRMR) and least absolute shrinkage and selection operator (LASSO) were used to select useful radiomics features to construct Radscore. Logistic regression was applied to select significant factors among Radscore, fat attenuation index (FAI) and coronary artery calcium score (CACS) to construct radiomics nomogram.

Results: Radscore [odds ratio (OR) = 2.84; P < 0.001] and CACS (OR = 1.00; P = 0.023) were identified as independent predictors to construct the radiomics nomogram. The radiomics nomogram showed excellent performance [training cohort: area under the curve (AUC) = 0.81; 95% CI: 0.76-0.86; validation cohort: AUC = 0.83; 95%CI: 0.76-0.90] to predict haemodynamically significant coronary stenosis in patients with T2DM. Decision curve analysis demonstrated high clinical value of the radiomics nomogram.

Conclusion: The non-contrast CT-based radiomics nomogram of PCAT could effectively predict haemodynamically significant coronary stenosis in patients with T2DM, which might be a potential noninvasive tool for screening of high-risk patients.

Keywords: CCTA derived fractional flow reserve; Haemodynamically significant coronary stenosis; Nomogram; Non-contrast CT; Pericoronary adipose tissue; Radiomics.

Fig. 1

Example case. The left anterior descending artery contained mixed plaques with functional ischemia: 3-dimensional (volume rendering) (A, B), curved planar reconstruction (C) and coronary computed tomography angiography derived fractional flow reserve [the measurement location (yellow marker and white arrow): 2 cm distal to the stenosis in plaque artery]. (D)

Fig. 2

The workflow for FAI achievement and radiomics nomogram development

Fig. 3

Feature selection process. The y axis represented LASSO coefficient profiles of the radiomics features and the lower x-axis indicated the log lambda (λ) (A). Sixteen radiomic features were selected to calculate radiomics score (B). The regression coefficients in the selected radiomics features (C)

Fig. 4

Distribution of the Radscore in the training and validation cohorts indicated that Radscore had an excellent ability to predict haemodynamically significant coronary stenosis in patients with type 2 diabetes. Coronary arteries without functional ischemia (blue); Coronary arteries with functional ischemia (yellow)

Fig. 5

AUCs of CACS, Radscore and radiomics nomogram for predicting haemodynamically significant coronary stenosis in patients with type 2 diabetes in each cohort. AUC, area under receiver operating characteristic curve

Fig. 6

The nomogram was constructed with the Radscore and CACS (A). Calibration curves of the radiomics nomogram in the training (B) and validation (C) cohorts. Decision curves indicated that radiomics nomogram and Radscore basically had higher clinical application value than CACS when risk threshold between 0.1 and 0.8 (D)