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

Coronary artery disease (CAD) is the most important cardiovascular disease threatening human health [1]. Type 2 diabetes mellitus (T2DM) is significantly associated with increased risk of CAD, and morbidity and mortality of patients with CAD is considerably higher in the presence of diabetes [2‐6]. The treatment strategies for obstructive CAD vary depending on whether the lesion is hemodynamically significant. Therefore, early screening and prediction of high-risk CAD with haemodynamically significant stenosis in T2DM patients can reduce the occurrence of major adverse cardiovascular events (MACE), which has important clinical value. Fractional flow reserve (FFR) is the gold standard for diagnosing hemodynamically significant coronary stenosis, while it is difficult to access due to high costs and potential risks, limiting its widespread use [7‐10]. Coronary CT angiography (CCTA) derived fractional flow reserve (FFR_CT), has highly consistent assessment in myocardial ischemia compared with invasive FFR [11‐13], without extra image acquisition and taking adenosine, showing great potential in the diagnosis of functional myocardial ischemia.

Inflammatory response of the coronary arteries has been shown to affect the formation and differentiation of pericoronary adipose tissue (PCAT) by releasing cytokines to prevent the lipid accumulation, which can be indicated by fat attenuation index (FAI) [14]. Meanwhile, chronic atherosclerosis and vascular inflammation of coronary artery can trigger permanent changes in the perivascular space, including fibrosis and microvascular remodeling [15, 16], which can be captured by radiomics with high-throughput extraction of quantitative features [17].Recently, radiomic signatures of PCAT have been proved to have important value for predicting hemodynamic significance of coronary stenosis [18‐20], however above studies mainly focused on radiomics analysis based CCTA. The use of iodine contrast agents intravenously might lead to the risk of allergic reaction, renal impairment and microcirculation disorders. Hence, this study aimed to develop a radiomics nomogram of PCAT based on non-contrast CT to predict haemodynamically significant coronary stenosis in patients with T2DM.

Local institutional review board and the ethics committee approved this study, and the requirement to acquire informed consent was waived.

In this study, a non-contrast CT-based radiomics nomogram of PCAT by integrating radiomics features and CACS was established and validated. Our results demonstrated the radiomics nomogram had excellent predictive performance for haemodynamically significant coronary stenosis and could be a promising noninvasive method to optimize risk stratification and guide treatment strategies in T2DM patients.

It is of especially crucial to find new and efficient methods to predict high-risk CAD with myocardial ischemia both in time and extent, on account of that CAD in T2DM patients is often asymptomatic and undiagnosed until acute myocardial infarction occurs [23] and early detection of high-risk CAD can provide an opportunity for early intervention and prevent MACE. FFRCT is widely used for prediction of function myocardial ischemia, influence on treatment decision making and prognostic evaluation of clinical outcome, however the technology is not suitable for screening of high-risk CAD especially for asymptomatic because of relative low incidence, low MACE rate, high cost, adverse side effects and high radiation exposure [24, 25]. Our study provides a novel noninvasive, practical and economical tool for predicting haemodynamically significant coronary stenosis in patients with T2DM, with low radiation dose and iodine avoiding, especially suitable for silent CAD. In our study, the radiomics nomogram achieved accurate prediction with AUC of 0.83 in the validation cohort. The median estimated effective dose of no-contrast CT in the present study is 0.67 mSv, which is consistent with our goal to screen high-risk CAD with a relative lower radiation dose, iodine avoiding and costs reduction.

Plaque development and lumen stenosis are caused by vascular inflammation, inducing endothelial dysfunction and impaired vasodilation, which may reduce the distal flow reserve and result in functional ischemia [26, 27]. Previous research demonstrated that FAI could identify coronary inflammation by capturing changes in perivascular fat attenuation, and also had value in predicting hemodynamic significance of coronary stenosis with AUC of 0.83 [28, 29]. However, in this study, the predictive performance of FAI was insufficient to identify haemodynamically significant coronary stenosis. It was controversial whether FAI can identify haemodynamically significant coronary stenosis, and the present study was consistent with previous studies with AUC from 0.55–0.67 [18, 28‐30]. There might be related with following reasons: first, different CT acquisition parameters had influence on FAI values; second, we traced the proximal 40 mm segments of LAD and LCX and the proximal 10–50 mm segment of RCA, while other studies traced the specific stenosis in CCTA [29]; third, the FAI analysis in above studies was based on CCTA, while we measured FAI on non-contract CT; forth, FAI might be changed by treatments [16].

Haemodynamically significant coronary stenosis caused by vascular inflammation [26, 27], might lead to alterations in the microenvironment and tissue components within the PCAT. Radiomics can extract thousands of quantitative imaging features from medical image data, and construct prediction models by selecting the most valuable features [17]. In our study, the Radscore was an independent predictor for functional ischemia. It was derived from the sixteen most contributive radiomic features extracted from PACT, including seven first-order features and nine texture features, without shaped features. Only one feature was extracted from the original images, nine features from wavelet, one feature from LoG filtered images and five from nonlinear strength transformation. First-order features reflect the intensity features containing gray histogram information of PCAT, and the texture features reflect the heterogeneity of PCAT. From the radiomics model, we found that, the PCAT of the positive FFR_CT case showed high heterogeneity (ExponentialGLRLM_exponential_RunLengthNonUniformity, SquareGLSZM_square_SizeZoneNonUniformityNormalized and LaplacianGLSZM_log − sigma − 1 − 0 − mm − 3D_SizeZoneNonUniformity), and a small proportion of lower gray-level values (WaveletGLSZMwavelet − HLH_LowGrayLevelZoneEmphasis and WaveletGLSZMwavelet − LHH_LowGrayLevelZoneEmphasis). These features mainly reflected the confusion, complexity and variability of PCAT and were potentially captured by PCAT radiomics features instead of PCAT attenuation. Radiomics signature is not only reflect the acute inflammation, but also describes fibrosis and vascularity in adipose tissue induced by chronic coronary inflammation [16]. Moreover, fat radiomics signature reveals persistently structural changes in PVAT and a corresponding residual risk not confounded by medications taking or other acute processes. Hence, Radscore performed better than FAI to identify haemodynamically significant coronary stenosis in patients with T2DM.

Hyperglycemia, insulin resistance and excess fatty acids in T2DM enhance oxidative stress, destroy protein kinase C signaling and increase advanced glycation end products which result in vascular inflammation, vasoconstriction and atherogenesis [31, 32]. Complex mechanism leads to release of osteoprogenitor cells form the bone marrow into the circulation, promoting coronary intimal calcification [33, 34]. Consistent with above studies [28], coronary artery calcification in the present study was shown to be an independent predictor for haemodynamically significant coronary stenosis in T2DM. Radiomics analysis has higher predictive ability than CACS and is particularly important to screen high-risk patients without severe coronary calcification in early T2DM patients.

However, there are several limitations in this study. First, image acquisition was acquired from the same CT manufacturer in order to ensure the image uniformity, which needed to verify the generalization of the present findings in other manufacturers further. Second, a multi-center study with a larger sample count should be performed for further research, since this single-center retrospective study lacked external validation. Third, the sample size was insufficient, not allowing for subgroup study, especially for patients in “gray-zone lesions”. Fourth, part of LCXs were not analyzed because of anatomic variation and difficult delineation. Fifth, the models demonstrated higher predictive value in the validation cohort than in the training cohort. The reason for this result may be that the data set was not large enough, even though the training and test sets were randomly assigned, the distribution of the training and test sets were somewhat uneven. Another possible reason was that there might be underfitting in the training cohort and led to the lowering predictive value of the training cohort.

In conclusion, a radiomics nomogram of PCAT based on non-contrast CT has excellent performance for discriminating coronary functional ischemia, which may potentially  become a noninvasive and economical method for predicting and screening high-risk CAD in patients with T2DM.