In the routine evaluation of atherosclerosis, patients typically undergo both lipid profiling, including measurement of low-density lipoprotein cholesterol (LDL-C), and coronary computed tomography angiography (CCTA). Clinically, elevated LDL-C is often accompanied by CCTA abnormalities such as non-calcified plaques, high-risk plaques, and changes in pericoronary adipose tissue (PCAT) attenuation [1, 2, 3]. In some patients, LDL-C elevation coincides with reduced PCAT attenuation, which may represent different stages of the same pathological process. However, the patient-level relationship between LDL-C and PCAT phenotype has not been systematically quantified, and existing evidence has rarely moved beyond qualitative observations or single-parameter imaging summaries.

Evidence-based research has established low-density lipoprotein (LDL) as a causal factor in atherosclerotic cardiovascular disease (ASCVD) [3, 4] and has demonstrated a dose-dependent, log-linear relationship between cumulative LDL-C exposure and cardiovascular event risk, with longer exposure associated with greater risk [5, 6]. Recent clinical evidence further suggests that statin therapy is associated with a reduction in PCAT lesion attenuation, whereas no significant changes are observed in the absence of statin treatment [7, 8]. After adjustment for cardiovascular risk factors, changes in LDL-C are independently associated with the percentage change in PCAT lesions. In a cohort of 180 patients with chest pain and intermediate risk of coronary artery disease [9], $\ge$1 year of statin therapy reduced the mean CT attenuation of perivascular adipose tissue (PVAT) surrounding non-calcified plaques from –68 $\pm$ 9 Hounsfield unit (HU) to –72 $\pm$ 8 HU (p 0.001), whereas no significant change was observed in PVAT attenuation surrounding calcified plaques. These findings suggest that the lipid status surrounding the coronary arteries may be reflected in the local PCAT phenotype in an imageable manner; however, existing evidence largely remains focused on the single imaging parameter of mean PCAT attenuation.

Whether LDL-C is associated with more complex PCAT microstructural signatures (e.g., spatial heterogeneity and texture patterns) on CCTA remains unclear.

Although previous studies have established PCAT as an imaging biomarker of coronary inflammation [10, 11], capable of improving coronary heart disease risk stratification and independently associated with adverse cardiovascular events [12, 13, 14], a single HU value alone cannot fully capture the complex alterations in adipose tissue microstructure and metabolic inflammation. Radiomics, through high-precision segmentation and computational algorithms [15, 16], can automatically extract multidimensional features from CCTA images, including first-order statistics, texture, and higher-order filtered features, thereby capturing tissue characteristics at multiple spatial scales [17]. Applying radiomics to PCAT enables in-depth characterization of its microstructural patterns, thereby overcoming the limitations of reliance on a single HU value [18].

Despite these observations, key gaps remain: the patient-level association between LDL-C and PCAT has rarely been quantified beyond mean attenuation, and informative PCAT radiomic signatures related to elevated LDL-C have not been systematically evaluated. To address these gaps, we (1) quantified multidimensional PCAT radiomic features on routine CCTA; (2) identified key features associated with elevated LDL-C using an internal training/hold-out evaluation strategy; and (3) applied Shapley additive explanations (SHAP)-based interpretation to improve model interpretability.

Based on these considerations, the aim of this study was to extract and select PCAT radiomic features from CCTA images of patients with varying lipid levels and to systematically assess their correlation with elevated LDL-C levels, identify key features most closely associated with LDL-C levels, and explore their potential value in elucidating the mechanisms of LDL-C–related coronary inflammation, thereby providing evidence to support the precise clinical evaluation and management of LDL-C–related cardiovascular risk.

To our knowledge, this study is among the first to evaluate the relationship between elevated LDL-C levels and PCAT phenotypes using CCTA-based radiomics. Individuals with higher lipid exposure demonstrated greater spatial heterogeneity of PCAT, characterized by larger high-attenuation (high-gray-level) regions. These observations are consistent with histologic evidence indicating that inflammation, fibrosis, and neovascularization contribute to adipose tissue inhomogeneity [1, 23, 24]. From a clinical perspective, these imaging changes are consistent with the cumulative metabolic burden of long-term LDL-C elevation, thereby providing imaging-based evidence supporting the link between cholesterol levels and localized coronary inflammation.

We identified several radiomic features associated with elevated LDL-C levels. Notably, second-order GLSZM metrics such as LargeAreaHighGrayLevelEmphasis and ZoneVariance were strongly related to LDL-C status, reflecting clustered regions of high gray-level voxels and increased textual heterogeneity within PCAT. These findings align with histopathological evidence: Mazurek et al. [23] demonstrated abundant inflammatory mediators, fibrosis, and microvascular proliferation within epicardial and pericoronary adipose tissue, providing biological support for the imaging manifestations of high-density clustering and increased heterogeneity. Mechanistically, LDL-C may influence PCAT through an “inside-out” signaling pathway, in which endothelial dysfunction activates adipocyte paracrine signaling, thereby promoting immune cell recruitment, lipid droplet remodeling, and stromal reorganization [11, 25, 26]. Animal studies have shown that high-fat diets exacerbate PCAT inflammation and remodeling [27], while human pathological studies corroborate these changes, showing reduced lipid droplet content, fibrosis, and inflammatory cell infiltration in hyperlipidemia and atherosclerosis [28, 29]. Taken together, these findings suggest that chronic LDL-C elevation may promote clustering and heterogeneity in PCAT, reflecting underlying inflammatory and structural remodeling processes, and strengthening the biological plausibility of our results. Nevertheless, these mechanistic interpretations remain hypothesis-generating and warrant independent biological validation in future studies.

Large-scale epidemiological studies have shown that lifelong elevation of LDL-C drives atherosclerosis, with cardiovascular risk increasing proportionally to both concentration and exposure duration [5, 6]. Further evidence indicates that statin therapy is associated with reduced PCAT attenuation, whereas untreated patients do not exhibit such changes [7, 8, 9]. Even after adjustment for potential confounders, changes in LDL-C remain independently correlated with percentage changes in PCAT attenuation. In the present study, first-order radiomic features reflected previously reported attenuation findings, whereas second-order features captured voxel clustering and heterogeneity, thereby providing structural insights beyond mean attenuation alone. These findings are consistent with mechanistic studies by Antonopoulos et al. [1], which demonstrated that local coronary inflammation alters PCAT through paracrine signaling, leading to impaired adipocyte maturation, lipid droplet depletion, and increased intracellular water [30, 31]. Collectively, these processes manifest as increased CT attenuation and greater textural complexity on imaging.

In prior work, Antonopoulos et al. [1] established PCAT attenuation–based metrics (FAI) as imaging surrogates of coronary inflammation and provided biological rationale linking vascular inflammation to perivascular fat remodeling. In parallel, Lin et al. [18] demonstrated that PCAT radiomics can capture spatial patterns beyond voxel intensity alone and may provide incremental discrimination compared with attenuation-only measures in specific clinical settings. Against this background, our study addresses a complementary question by focusing on the patient-level association between LDL-C status and multidimensional PCAT radiomic phenotypes (texture/heterogeneity). Accordingly, the radiomic signatures reported here should be interpreted as exploratory imaging biomarkers that may complement—rather than replace—established attenuation/FAI measures, pending further validation.

Within the framework of residual cardiovascular risk, low-grade inflammation often explains adverse outcomes in statin-treated patients more effectively than residual cholesterol burden [32, 33]. A large percutaneous coronary intervention (PCI) registry showed that high-sensitivity C-reactive protein (hs-CRP) levels $\ge$2 mg/L, rather than high LDL-C, were independently associated with major adverse cardiovascular events (MACE) [32]. PCAT is recognized as a key imaging window into localized inflammation. Prior studies, including Oikonomou et al. [10] and Tzolos et al. [13] have supported PCAT attenuation–based metrics (e.g., FAI) as noninvasive markers of coronary inflammation with potential value for risk stratification and associations with adverse cardiovascular outcomes. In this context, we observed that elevated LDL-C levels were associated with increased clustering and heterogeneity within PCAT. These exploratory radiomic signatures may reflect microstructural patterns related to lipid–inflammation interactions and warrant further evaluation alongside established PCAT attenuation/FAI measures in future studies. This perspective not only provides new imaging -based evidence of the LDL-C–PCAT association but also enhances current understanding of lipid–inflammation interactions.

This study extends existing evidence by linking elevated LDL-C levels with PCAT not only through established attenuation patterns but also through novel radiomic signatures of clustering and heterogeneity. These textural features enhance the interpretability of imaging along the lipid–inflammation axis. Importantly, they may serve as external imaging correlates of LDL-C elevation, potentially providing early warning signs of coronary inflammation beyond traditional lipid markers. However, given the single-center retrospective design and the absence of independent external validation, these radiomic signatures should be considered exploratory imaging biomarkers, and their performance and generalizability should be interpreted cautiously. With further validation, these features may evolve into adjunctive tools for therapeutic monitoring and risk stratification. Prospective multicenter studies with independent external validation and outcome follow-up are required to establish generalizability and clinical utility.

This single-center, retrospective study with a modest sample size (n = 150) may limit representativeness and generalizability. The machine-learning model was evaluated only internally without independent external validation; therefore, model optimism and potential overfitting cannot be fully excluded and the findings should be interpreted as exploratory. Residual confounding cannot be fully ruled out because multivariable adjustment was not performed, including potential effects of lipid-lowering therapy. In addition, although images were acquired on the same scanner using a standardized protocol, we did not perform formal radiomic reproducibility testing (e.g., ICC/test–retest) and did not assess reproducibility or harmonization across scanners/protocols. Finally, the analysis was cross-sectional and lacked outcome follow-up, precluding assessment of prognostic value and causal inference; the absence of histologic or molecular validation also limits definitive mechanistic interpretation. Future work should include multicenter external validation with longitudinal follow-up and, where feasible, biological validation to establish robustness, generalizability, and clinical utility.

Elevated LDL-C levels were significantly associated with increased spatial heterogeneity and high gray-level clustering in PCAT, providing imaging-based evidence of the LDL-C–PCAT association. These findings suggest that PCAT radiomic features may serve as external imaging phenotypes of lipid burden, offering more nuanced insight into the relationship between LDL-C and the local coronary microenvironment. Collectively, these radiomic features hold potential clinical value by supporting more precise assessment and management of coronary risk in patients with lipid dysregulation.

AUC, area under the curve; CAD, coronary artery disease; CCTA, coronary computed tomography angiography; FAI, fat attenuation index; GBM, gradient boosting machine; GLSZM, gray-level size zone matrix; HDL-C, high-density lipoprotein cholesterol; HU, hounsfield unit; LAD, left anterior descending artery; LCX, left circumflex artery; LDL, low-density lipoprotein; LDL-C, low-density lipoprotein cholesterol; PCAT, pericoronary adipose tissue; RCA, right coronary artery; ROI, region of interest; SHAP, shapley additive explanations; SVM, support vector machine.

The datasets used in this study are available from the corresponding author upon reasonable request.

FZ and XL designed and conducted the study. LY and XL performed PCAT segmentation and radiomics feature extraction using the CoronaryDoc platform and PyRadiomics. FZ and YZ performed the statistical analyses and developed the machine-learning workflow, including model training/validation and performance evaluation. FZ conducted model interpretation using SHAP analysis and drafted the manuscript. YY, ST, and FL screened patients, maintained the clinical database, coordinated patient identification and clinical data collection, and verified key clinical variables. XZ and XD designed the statistical analysis plan and optimized the modeling. NT, SZ, YX, and CZ assisted with image quality assessment, data curation, and data verification. LS and GS discussed and interpreted the data and revised the manuscript. All authors contributed to drafting and critical revision of the manuscript for important intellectual content, approved the final version, and agree to be accountable for all aspects of the work.

The study was carried out in accordance with the guidelines of the Declaration of Helsinki and approved by the Ethics Committee of Kunming Yan’an Hospital (Kunming, China; Approval No. 2023-083-01). Informed consent was waived due to the retrospective study design.

Not applicable.

The study was supported by Yunnan Provincial Department of Science and Technology-Kunming Medical University-Joint Project (NO. 202301AY070001-178), Yunnan Provincial Department of Science and Technology (NO. 202301AT070482), Kunming Health Science and Technology Talents Project (NO. 2023-SW(Leading talents)-03), Key Laboratory of Cardiovascular Disease of Yunnan Province (NO. 2018DG008), Kunming Health Research Project (NO. 20230901013).

The authors declare no conflicts of interest.

Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.31083/RCM47037.