A novel arterial redox-specific machine learning-derived radiomic signature of perivascular adipose tissue predicts cardiac mortality from routine CCTA

Abstract

Abstract Background Vascular oxidative stress is involved in inflammation and atherogenesis. Vascular inflammation induces spatial changes in perivascular adipose tissue (PVAT) composition, which can be detected by radiomic analysis of coronary computed tomography angiography (CCTA) images. Purpose To explore the association of arterial oxidative stress with long-term risk of major adverse cardiovascular events (MACE). To develop a radiomic signature to identify high oxidative stress non-invasively using CCTA. Finally, to assess the ability of this signature to predict future cardiac risk. Methods Arm 1 included 272 patients undergoing cardiac surgery. Segments of internal mammary artery (IMA) were used for ex-vivo quantification of NADPH-stimulated and Vas2870 (pan-NOX inhibitor) inhibitable superoxide production by lucigenin-enhanced chemiluminescence. Eighty-two of these patients with CCTA scans available constituted Arm 2. Peri-IMA PVAT was segmented and used to extract 1,616 radiomic features, which, after filtering (40 final features), were utilised to train extreme gradient boosting, a machine learning algorithm, to predict high arterial oxidative stress. Arm 3 included a nested cohort of 308 participants (41 suffering cardiac death and 267 matched controls) from the CRISP-CT (Cardiovascular RISk Prediction using Computed Tomography) study to externally validate the redox-specific signature developed in Arm 2 for cardiac risk prediction. Results Over a median follow-up of 40 months, 18 (6.6%) MACE (cardiovascular death, non-fatal myocardial infarction, and stroke) occurred in Arm 1. High arterial NADPH-stimulated superoxide was independently associated with MACE risk (Adj. HR[95% CI]: 1.61 [1.04–2.53] per SD, p=0.03, adjusted for age, sex, diabetes, hypertension, hyperlipidemia, smoking, obesity, and plasma TNFa). Unsupervised hierarchical clustering of radiomic features from peri-IMA PVAT segments in Arm 2 identified two distinct clusters (A) that differed in NADPH stimulated (p=0.01) and Vas2870 inhibitable (p=0.04) IMA superoxide (B), supporting the hypothesis that PVAT mapping can capture changes corresponding to differential levels of underlying vascular redox state. This prompted the development of a radiomic signature specific to PVAT alterations associated with high vascular oxidative stress, which was validated in Arm 3 (AUC:0.61, p=0.026, C). The novel signature was able to stratify cardiac risk in the validation set, independently of the Fat Attenuation Index, epicardial adipose tissue volume, high-risk plaque features, and obstructive CAD (Adj. HR [95% CI]:2.56 [1.35–4.87], p=0.004, D). Conclusion Increased arterial oxidative stress predicts cardiac risk in patients with advanced atherosclerosis. We present for the first time a novel, non-invasive CCTA imaging biomarker reflecting changes in vascular redox state by radiomic phenotyping of perivascular space, which stratifies cardiac risk beyond standard and newer risk assessment methods. Funding Acknowledgement Type of funding source: Foundation. Main funding source(s): British Heart Foundation, National Institute of Health Research, Oxford Biomedical Research Centre