Skip to main content
Download PDF
Download ePub

Fully automated epicardial adipose tissue volume quantification with deep learning and relationship with CAC score and micro/macrovascular complications in people living with type 2 diabetes: the multicenter EPIDIAB study

Cardiovascular Diabetology volume 23, Article number: 328 (2024) Cite this article

Abstract

Background

The aim of this study (EPIDIAB) was to assess the relationship between epicardial adipose tissue (EAT) and the micro and macrovascular complications (MVC) of type 2 diabetes (T2D).

Methods

EPIDIAB is a post hoc analysis from the AngioSafe T2D study, which is a multicentric study aimed at determining the safety of antihyperglycemic drugs on retina and including patients with T2D screened for diabetic retinopathy (DR) (n = 7200) and deeply phenotyped for MVC. Patients included who had undergone cardiac CT for CAC (Coronary Artery Calcium) scoring after inclusion (n = 1253) were tested with a validated deep learning segmentation pipeline for EAT volume quantification.

Results

Median age of the study population was 61 [54;67], with a majority of men (57%) a median duration of the disease 11 years [5;18] and a mean HbA1c of7.8 ± 1.4%. EAT was significantly associated with all traditional CV risk factors. EAT volume significantly increased with chronic kidney disease (CKD vs no CKD: 87.8 [63.5;118.6] vs 82.7 mL [58.8;110.8], p = 0.008), coronary artery disease (CAD vs no CAD: 112.2 [82.7;133.3] vs 83.8 mL [59.4;112.1], p = 0.0004, peripheral arterial disease (PAD vs no PAD: 107 [76.2;141] vs 84.6 mL[59.2; 114], p = 0.0005 and elevated CAC score (> 100 vs  < 100 AU: 96.8 mL [69.1;130] vs 77.9 mL [53.8;107.7], p < 0.0001). By contrast, EAT volume was neither associated with DR, nor with peripheral neuropathy. We further evidenced a subgroup of patients with high EAT volume and a null CAC score. Interestingly, this group were more likely to be composed of young women with a high BMI, a lower duration of T2D, a lower prevalence of microvascular complications, and a higher inflammatory profile.

Conclusions

Fully-automated EAT volume quantification could provide useful information about the risk of both renal and macrovascular complications in T2D patients.

Article Highlights

Why did we undertake this study? What is the specific question(s) we wanted to answer?

This study addresses the unmet need to assess epicardial fat volume quantification in high-risk people living with type 2 diabetes using a fully-automated deep learning AI tool.

What did we find?

Fully automated epicardial fat volume quantification with cardiac CT performed for CAC scoring is possible and reliable in T2D.

Epicardial fat volume was associated with all cardiovascular risk factors, CKD and macrovascular complications but not with diabetic retinopathy or peripheral neuropathy.

We identified a subgroup of T2D patients with a null CAC score and high EAT volume which was characterized by a higher systemic proinflammatory profile.

What are the implications of our findings?

This study provides new insights for non-invasive deep phenotyping of patients living with type 2 diabetes with epicardial fat volume quantification using cardiac CT performed for CAC scoring, that could be used in clinical practice.

These findings set the stage for personalized medicine and prospective randomized trials testing new antihyperglycemic drugs that target inflammation.

Introduction

While it is now recognized that regional adiposity such as visceral adiposity is a stronger cardio-metabolic risk factor than overall obesity [1], the ability of the subcutaneous adipose tissue to expand with positive energy balance constitutes the main determinant of ectopic fat deposition in peripheral organs such as the heart [2, 3]. Epicardial adipose tissue (EAT) is a perivascular depot located between myocardium and the visceral pericardium [4]. EAT has unique properties that distinguish it from other depots of visceral fat, due to its unobstructed proximity to the myocardium and to the coronary arteries, allowing bidirectional paracrine or vasocrine crosstalk between adipocytes, cardiomyocytes and cells of the vascular wall [5]. An imbalance between cardioprotective and harmful adipokines secreted by epicardial fat has been linked to the development of coronary atherosclerosis in humans [6, 7].

Given its high metabolism, thermogenic capacity, unique transcriptome, secretory profile, and simply measurability, epicardial fat has drawn increasing attention in the recent years to help clinicians in evaluating individual cardiovascular risk and implementing precision medicine [8,9,10]. The quantity and the inflammation of this perivascular fat depot has been independently associated with major adverse cardiovascular events, incident myocardial infarction, all-cause mortality and cardiac mortality in high risk individuals [11,12,13]. We and others have evidenced that EAT is associated with the progression of coronary artery calcification and blunted coronary microvascular response, especially in young subjects and subjects with low coronary artery calcium (CAC) score measured by cardiac computed tomography (CCT), suggesting that EAT may promote early atherosclerosis development [14, 15]. Type 2 diabetes (T2D) has been consistently associated with an increase in EAT [11, 16]. Besides, a loss of transcriptomic brown-like fat features in EAT has been evidenced in people living with T2D [17], promoting a switch to a pro-atherogenic and a pro-arrhythmogenic profile [4, 18]. EAT is increasing being recognized as a determinant of vascular complications of metabolic diseases such as obesity or T2D [19]. However, manual quantification of EAT in clinical practice is time-consuming and requires important individual imaging data post processing treatment [20]. In this context, we developed a deep-learning-based quantification of epicardial adipose tissue volume using non-contrast CCT performed for CAC scoring [21], in a large sample of people living with T2D (n = 1253) and included in the Angiosafe T2D cohort. The aim of this study was to evaluate the relationship between EAT volume and the micro and macrovascular complications (MVC) of T2D.

Study population

The present study, named EPIDIAB study is a post-hoc analysis from AngioSafe T2D study, which is a bicentric 3-year longitudinal study (NCT02671864) aimed at determining the safety of antihyperglycemic drugs in the retina and angiogenesis and described elsewhere [22]. The overall study plans to include 7200 people living with T2D at the Centre Universitaire du Diabète et de ses Complications, Lariboisière Hospital, APHP, Paris, or at the Endocrinology, Metabolic diseases and Nutrition department, Pole ENDO, APHM, Marseille, France and screened for diabetic retinopathy (DR) by retinal fundus photographs. The International classification of Diabetic Retinopathy [23] was used to perform DR staging. All consecutive T2D patients (≥ 18 years old) attending these two centers from June 2016 to March 2024 and who had undergone a non-contrast cardiac CT (= 1253) performed for CAC (Coronary Artery Calcium) scoring were selected for the analysis. Exclusion criteria were pregnancy, Maturity-Onset Diabetes of the Youth (MODY), type 1 or pancreatic diabetes, dense cataract preventing DR grading, or patients with panretinal photocoagulation of a duration of 10 years or longer. All patients gave their written informed content before inclusion. The local ethics committee (Comité de protection des Personnes ILE DE FRANCE V) gave his agreement for this ancillary study (15.00261.015070-MS05), all the trial procedures were in compliance with the Declaration of Helsinki.

Data collection

Data were extracted from patients’ medical records and collected in a secure health database.

Collected data included clinical characteristics (age, sex, waist circumference, body mass index, duration of diabetes), cardiovascular risk factors (hypertension, dyslipidemia, current tobacco consumption and pack-year smoking history, obesity defined as BMI ≥ 30 kg/m2). HbA1c (high performance liquid chromatography), total, low-density lipoprotein, and high-density lipoprotein cholesterol; triglycerides; aspartate aminotransferase (ASAT), alanine aminotransferase (ALT), gamma glutamyl transferase (GGT), urine albumin-to-creatinine ratio (UACR), and estimated glomerular filtration rate (eGFR) using the Chronic Kidney Disease-Epidemiology Collaboration Equation), uric acid, high sensitive CRP, fibrinogen, brain natriuretic peptide (BNP) were assessed the day of inclusion.

Chronic kidney disease was defined as eGFR < 60 mL/min/1.73 m2 and/or UACR ≥ 3 mg/mmoL). The kidney failure risk equation (KFRE) was applied to calculate the patient risk progression to kidney failure requiring dialysis or transplant at 5 years using calculator https://www.kidneyfailurerisk.co.uk/. Neuropathy was defined as any sign or symptom of polyneuropathy i.e. sensory signs (disruption of sensitivity to light touch e.g., cotton-wool and pinprick, monofilament, or to vibration with tuning fork) and/or symptoms (pain, cramps, numbness, paresthesia), and/or absence of at least one tendon reflex motor, and/or deficit or trophic disorders. The presence of macrovascular complications was defined according to the presence of a previous history of ischemic heart disease (including a history of myocardial infarction and/or coronary artery revascularization or heart failure), cerebrovascular disease (including history of stroke or transient ischemic attack (TIA)) and/or peripheral artery disease (amputation owing to ischemic disease and/or lower limb artery revascularization).

CT imaging and deep learning model automated segmentation of epicardial fat volume

CAC scores and EAT volumes were calculated using ECG-gated cardiac CT without contrast injection. CT scans were performed on several scanners (FRONTIER, REVO EVO, APEX; General Electric Healthcare, Buc, France). Coronary artery calcium score was quantified by the Agatston method on noncontrast cardiac CT scans using commercially available software (Aquarius Workstation® V.4.4.11–13, TeraRecon Inc, Foster City, CA, USA), in those patients with an indication for CAC score assessment.

For each patient, images were extracted from the picture archiving and communicating system and imported in the DICOM format on the validated post-processing software 3D Slicer (3D Slicer v4.11.20210226) [24].

To quantify the epicardial fat from a CCT scan, a deep learning segmentation pipeline, was first developed from low-dose computed tomography (LDCT) and validated in a cohort of 353 consecutive patients with COVID-19 and who underwent LDCT for lesions extension[21]. Then, in the EPIDIAB study, we trained and optimized the AI algorithm on cardiac CT already performed for CAC scoring in high-risk people living with type 2 diabetes and referred for diabetes care in the Assistance Publique Hôpitaux de Marseille (APHM) Endocrinology department. HRNet, the main backbone convolutional network, initially trained with 95 manually annotated patients (47,214 CCT slices), was fine-tuned with 65 additional T2D patients, randomly extracted from the Angiosafe T2D database (the target distribution), which gave a total of 160 training examples. To control model performance on the target distribution, i.e. real-world patients with T2D, a visual quality control was performed in 150 unseen CCT from the Angiosafe T2D cohort referred for CAC scoring. Finally, the deep learning pipeline was used to quantify the EAT volume of 300 unseen patients from Angiosafe Marseille and of 1000 other patients from Angiosafe Paris.

Briefly, manual segmentation had been performed slice by slice on the entire intrapericardial soft tissue volume by delineating the external border of the pericardium using thresholding, painting, and erasing methods. The superior and inferior limits of the pericardium were first identified as the top of the left atrium, and the lower limit corresponded to the last slice in which the left ventricle was identified. Pericardial fat was excluded. The EAT tissue was then identified inside the intrapericardial volume by using the standard fat attenuation range as a threshold, from − 190 Hounsfield units (HU) to − 30 HU [25]. A post-processing with small connected region (< 50 voxels) removal, morphological closing and median filter with a kernel size of 5, was finally applied on each slice. The obtained segmentation masks were all validated by one experienced chest radiologist (A.J., 25 years of experience).

Algorithm development was run on a Biprocessor Intel Xeon Silver 4216 2.1 GHz, RAM = 96Go, 2 GPU Nvidia Quadro RTX5000, 16Go.

An example of automated machine segmentation of epicardial fat volume on anterior superior and inferior view is presented in Fig. 1.

Fig. 1
figure 1

Schematic representation of the scientific approach to the development of the deep-learning model Automated Segmentation of Epicardial adipose tissue within the Pericardium Schematic representation of the scientific approach to the development of the deep-learning model from LDCCT, to CCT used for CAC scoring. The development of the deep learning network model was performed through internal and external subgroups of the Angiosafe DT2 cohort, to validate a reliable application for the automated epicardial adipose tissue (EAT) volume quantification LDCCT low dose chest computed tomography CCT cardiac computed tomography

Full size image

Statistical analysis

Anthropometric and biological parameters were expressed as mean ± SD or median [25th;75th percentile] according to the normality of the distribution and categorical variables as numbers (n) and percentages. The normal distribution of quantitative data was assessed using the Shapiro–Wilk normality test. Significant differences between groups “EAT volume > 100 mL and CAC > 100 AU” and “EAT > 100 mL and CAC = 0 AU” were determined using the Student’s t-test or Mann–Whitney test where appropriate. The Chi-2 test was used to compare categorical variables between groups. Pearson’s and/or Spearman’s correlations were performed to identify the parameters associated with EAT. Multivariate analysis included all the parameters with at least 50% of data available and correlated with EAT volume in univariate analysis. In sub-group analyses, an EAT volume cut-off of 100 mL was chosen to identify patients with high EAT volume, as previously proposed [26]. Statistical analyses were performed with Prism 9 (Graphpad, MA, USA). A two-sided value of less than 0.05 was considered statistically significant. No data replacement procedure was used for missing data.

Results

EAT algorithm performance

The mean Dice coefficients for the automatic segmentation of all the pericardial and EAT volumes were 0.948 ± 0.024 and 0.848 ± 0.068, respectively. Mean acquisition time per patient was less than 2 min.

Study population

Among the 1300 patients living with T2D included in the Angiosafe DT2 cohort, who had had CCT for CAC scoring and who met study inclusion criteria with retinopathy grading, the validated algorithm was used. An EAT volume was obtained for 1253 patients. Their clinical characteristics are depicted in Table 1 with n representing the number of available data.

Table 1 Patients characteristics
Full size table

The median age of the study population was 61 years old [54; 67], with a majority of men (57%), and a high number of people with cardiovascular risk factors: 42% lived with obesity; 63% had hypertension; 52% had dyslipidemia, 42% had ever smoked (14% of active smokers). The median duration of the disease was 11 years [5;18], with a mean HbA1c 7.8 ± 1.4%. A majority of patients were treated with oral antihyperglycemic agents (86%), 24% were under GLP-1 receptor agonist treatment, 2% under SGLT-2 inhibitors and 34% were requiring insulin treatment.

Regarding microvascular complications, 27% had clinical symptoms of peripheral neuropathy, 35% had CKD, and 32% had DR: 201 with mild (16%), 85 with moderate (7%), 37 with severe non proliferative (3%) and 76 with proliferative DR or history of laser panphotocoagulation (6%).

Macrovascular complications were present in 11% of patients with 4% of coronary artery disease (CAD), 2% of cerebrovascular disease, and 5% of peripheral arterial disease (PAD) and were in secondary cardiovascular disease prevention. The median of CAC score was relatively low at 20 AU [0;198], with 430 patients with a CAC score at 0 AU (34%).

Quantitative parameters associated with epicardial fat volume

EAT volume was positively associated with age, BMI, waist circumference, fasting plasma triglycerides, ALAT, GGT, uric acid, BNP, high sensitive CRP, fibrinogen, CAC score, serum creatinine, albuminuria, proteinuria and KFRE. It was negatively correlated with HDL, LDL cholesterol and GFR (CKD epi) (Supplementary Table 1).

EAT volume and CV risk factors in T2D

EAT volume was associated with all CV risk factors and was significantly higher in men than women (median 87.9 mL [62.4; 121.7] versus 78.7 mL [54.3; 107.4], respectively p < 0.0001), in people living with obesity than without (97 mL [70.9; 126.8] 75.9 mL [53.7; 106], respectively, p < 0.0001, hypertension (HTA) (“with HTA” 87.8 mL [62.6; 118.6] vs “without HTA” 76.6 mL [53.2; 105.1], smoking (“active smoking + past smoking < 3 years” 95.4 mL [68.6; 128.8], versus “non smokers and past smokers > 3 years” 77.6 mL [53.6; 105.4], p < 0.0001), and higher in patients with dyslipidemia than without 88.0 mL [63.0; 116.7], versus 80.6 [55.8; 111.3], respectively p = 0.0011 (Fig. 2A).

Fig. 2
figure 2

A Association deep-learning model Automated Segmentation of EAT volume with cardiovascular risk factors. W: women M: men ***p < 0.0001. B EAT volume relationship with CAC score represented by tertiles and diabetic retinopathy status assessed by retinophotography. DR diabetic retinopathy; NPDR non proliferative diabetic retinopathy; PDR proliferative diabetic retinopathy

Full size image

EAT volume and T2D complications

EAT volume was significantly higher in patients with CKD (yes 87.8 [63.5; 118.6]; no 82.7 mL [58.8; 110.8], p = 0.008), with CAD (yes 112.2 [82.7; 133.3] vs no 83.8 [59.4; 112.1], p = 0.0004), with PAD (yes 107 [76.2; 141] vs no 84.6 mL [59.2; 114], p = 0.0005) and with high CAC score (CAC score > 100 AU: 96.8 mL [69.1; 130] vs CAC score < 100 AU: 77.9 mL [53.8; 107.7], p < 0.0001) (Fig. 2).

By contrast, EAT volume was neither associated with DR status, nor with peripheral neuropathy (Fig. 2).

In multivariate analysis, EAT volume remained significantly associated with age, sex, waist circumference (but not BMI), ALAT, BNP, and fibrinogen (supplementary Table 1).

Characteristics of T2D patients with null CAC score and a relatively high EAT volume > 100 mL

Remarkably, we evidenced 48 T2D patients a high volume of EAT (median of the cohort 85 mL [60; 115] (Table 1) and a null CAC score. Interestingly, this group of T2D patients were more likely to be young women with a high BMI, a lower duration of the disease, a lower rate of microvascular complications (CKD and DR) and with a higher “low-grade” inflammatory profile, (ie. higher high-sensitive CRP and fibrinogen (Table 2).

Table 2 Comparison of clinical and biological characteristics of T2D patients with null CAC score and a relatively high EAT volume > 100 mL and patients with EAT > 100 mL and CAC > 100
Full size table

Discussion

This study provides evidence in a large cohort of patients living with T2D and screened for diabetic retinopathy, that deep-learning fully-automated segmentation of epicardial fat volume is achievable, robust and reliable on cardiac CT performed for CAC scoring. Mean acquisition time was less than 2 min, with expert manual quantification high agreement, and a respectable Dice similarity coefficient compared to other studies [27, 28]. Second, we showed that epicardial adipose tissue volume increased with all CV risk factors, and with T2D micro and macrovascular complications (MCV) such as CAD, and PAD, but also CKD. However, EAT volume was not significantly associated with diabetic retinopathy and did not increase with the severity of DR. Finally, we report for the first time a group of patients living with T2D and with a null CAC score, who have significant epicardial fat accumulation associated with a low-grade inflammatory profile, but less microvascular complications. These findings could potentially indicate that these patients might be at risk of CV events independently of their subclinical coronary atherosclerosis, and could take benefit from treatments that target inflammation, to improve cardiovascular prognosis. In the ORFAN study including more than 40 000 individuals (18.2% living with diabetes), EAT inflammation quantification using artificial intelligence risk model has been recently shown to exceed traditional risk factor-based risk calculators to predict major adverse cardiac events (MACE) and cardiac mortality in patients without obstructive CAD, suggesting that this parameter could be a critical point to take into account in clinical practice to better phenotype patients living with T2D [13].

Epicardial adipose tissue is a unique ectopic fat depot located in direct contact with myocardium and coronary arteries, with a distinctive transcriptome, proteome and immune cells such as innate lymphoid cells (ILCs), essential effectors of innate immunity via the rapid production of both proinflammatory and regulatory cytokines that infiltrate epicardial adipocytes and could modulate its inflammatory/beiging phenotype [8, 29] compared to other adipose tissues. We recently evidenced a positive correlation between the T helper cell subtype Th2 immune pathway and browning genes in human EAT versus thoracic subcutaneous adipose tissue [29]. Gene expression phenotyping confirmed specific upregulation of Th2 pathway and browning genes (IL-33 and uncoupling protein 1 [UCP-1]) in EAT, whereas ILC1was the most prevalent type of ILCs in all adipose tissues [29]. Accumulation of adipose ILC1s can trigger an increase local inflammation and systemic insulin resistance and is strongly associated with glucose homeostasis in humans, such as glycated haemoglobin (HbA1c), fasting plasma glucose (FPG), homoeostasis model assessment for insulin resistance (HOMA-IR), adipose tissue insulin resistance index (Adipo-IR) and serum free fatty acids (FFAs) [30]. An imbalance between proinflammatory/beiging phenotype has been observed in patients with T2D and multivessel disease with a decrease of PGC1α, and UCP1 mRNA expression in EAT of patients with CAD compared to patients non diabetics with CAD or with patients with no CAD disease [17].

Recent research in the field of cardiometabolic diseases shows that scientists and clinicians should take a step back from BMI-centric, to organ-specific adiposity and that EAT could become a therapeutic target for drugs inducing significant weight-loss, decreasing inflammation and providing cardiovascular or renal benefits [4, 19]. A recent real-world study using multi-organ quantitative multiparametric MRI from Diamond et al., demonstrated a high burden of combined steatosis and fibro-inflammation, within the liver, pancreas and kidneys associated with visceral adiposity and poor vascular health in patients with T2D [31]. Unfortunately, EAT was not evaluated in this study. But this highlights the renewing interest in multi-parametric non-invasive imaging and the importance of taking into account organ specific adiposity in evaluating the risk of lipotoxic organ dysfunction. In our study, we quantified both CAC score and EAT and showed that some patients with a null CAC score could have increased EAT, implicating that EAT is not only a surrogate of atherosclerosis. Furthermore, these patients had a systemic low grade inflammation profile with a two-fold higher hs CRP than patients with a CAC score and EAT > 100. Some researchers make the hypothesis that key features of T2D such as insulin resistance, decreased insulin secretion and glycosuria are primarily mechanisms that could protect against overnutrition by preventing the accumulation and overloading of tissues with cell nutrients [32]. If these mechanisms are exceeded, damaging innate immune activation occurs, resulting in the mounting of the proinflammatory cytokine network [32]. Recent large outcome studies such as CANTOS trial showed that newly developed IL-1β- blocking antibodies can reduce cardiovascular complications in patients with metabolic syndrome and high CRP levels [33]. Besides, statins have been shown to reduce EAT volume and EAT Hounsfield units [HUs] attenuation on computed tomography which is considered as a marker/proxy of pericoronary inflammation, as evidenced by Raggi et al., in a cohort of 420 postmenopausal women, randomized to either 80 mg of atorvastatine or 40 mg pravastatin daily, and rescanned after one year [34]. Remarkably, this EAT HU decrease was independent of lipid lowering and change in LDL cholesterol, suggesting a possible pleiotropic effect of this lipid-lowering drug [34].

The paracrine effects of EAT, particularly on vascular tone and vascular inflammation, are significantly altered in metabolic diseases such as T2D. With weight gain, EAT loses its ability to affect insulin-induced vasodilatation and to antagonize sympathetic tone, predominately due to impaired adiponectin secretion [35]. The aortic perivascular adipose tissue EAT, under diabetic conditions, shifts towards a pro-inflammatory (increment in CRP, chemokine (C–C motif) ligand 2, CD36), pro-oxidant (increased aldose reductase and reduced anti-oxidant deference enzymes), and vasoconstriction state [36]. Targeting EAT inflammatory phenotype using new radiotranscriptomics analysis such as the fat attenuation index could represent in the next future, a way to personalize antidiabetic medications and decrease the CV and renal risk [19].

In our study, we found that EAT volume was not associated with diabetic retinopathy (DR). Cosson et al., reported a lower EAT volume in patients living with diabetes complicated with diabetic retinopathy compared to patients without DR and who had a CCT with EAT volume and CAC score quantification using the automated software package AW VolumeShare7 [37, 38]. One explanation regarding these discrepancy results could be that this study included also type 1 diabetic patients, and patients with another type of diabetes, who have less ectopic fat deposition and prevalence of obesity than patients living with T2D. One strength of our study is that we evaluated all the stages of DR, and we observed that patients with severe non proliferative or with proliferative DR, EAT volume was not different from patients with less severe stages of DR, suggesting that ectopic fat in the heart is not associated with retinal microvascular damage. But further longitudinal studies are warranted to determine whether epicardial fat could participate in the appearance or in the progression of DR and longitudinal follow-up of Angiosafe-DT2 at 3 years will probably help to answer this still open question.

Our study has several limitations. First, its observational design prevented us from being able to draw definite conclusions about causal relationships between EAT volume and T2D micro or macrovascular complications. Second, we included patients screened for DR and who had a cardiac CT for CAC scoring, only 4% had CAD disease. Therefore, our results may not be representative of all diabetic patients, such as type 1 diabetic patients, or for patients in secondary CV prevention. Third we explored global but not regional EAT, and EAT volume but not its density, which could have provided more information on EAT inflammatory phenotype [8, 39,40,41,42]. However, Christensen et al. has already evidenced in 1030 patients with T2D that a high level of EAT measured with echocardiography was associated with the composite endpoint of incident cardiovascular disease and mortality, even after adjusting for CV risk factors and particularly in men [43]. Finally, radiation exposure was one of the limitation of EAT volume quantification with CT compared to cardiac transthoracic echography and/or MRI. However, no additional radiation was done for the EPIDIAB study, as the EAT volume quantification was performed on cardiac CT already performed for CAC scoring as part of the clinical care of patients.

Conclusion

We showed in a well phenotyped cohort of patients living with T2D that fully automated segmentation of epicardial fat volume is achievable, robust and reliable on cardiac CT performed for CAC scoring with a short acquisition time. This non-invasive multi-parametric imaging could be used in clinical practice to quantify ectopic fat in the heart and contribute to personalized medicine. Besides, EAT volume was significantly associated with all CV risk factors, and increased with CAC score and diabetes complications such as CAD, PAD, and CKD. EAT volume was not associated with DR and peripheral neuropathy. We identified a group of patients with a null CAC score and a high EAT volume with a high systemic inflammatory profile which could benefit from treatments targeting immune system or decreasing inflammation.

Data availability

The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.

References

  1. Rosito GA, Massaro JM, Hoffmann U, Ruberg FL, Mahabadi AA, Vasan RS, et al. Pericardial Fat, Visceral Abdominal Fat, Cardiovascular Disease Risk Factors, and Vascular Calcification in a Community-Based Sample: The Framingham Heart Study. Circulation. 2008;117(5):605–13.

    Article  PubMed  Google Scholar 

  2. Tchernof A, Despres JP. Pathophysiology of Human Visceral Obesity: An Update. Physiol Rev. 2013;93(1):359–404.

    Article  CAS  PubMed  Google Scholar 

  3. Pellegrinelli V, Carobbio S, Vidal-Puig A. Adipose tissue plasticity: how fat depots respond differently to pathophysiological cues. Diabetologia. 2016;59(6):1075–88.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  4. Iacobellis G. Epicardial adipose tissue in contemporary cardiology. Nat Rev Cardiol. 2022;19(9):593–606.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  5. Gaborit B, Sengenes C, Ancel P, Jacquier A, Dutour A. Role of Epicardial Adipose Tissue in Health and Disease: A Matter of Fat? Compr Physiol. 2017;7(3):1051–82.

    Article  PubMed  Google Scholar 

  6. Karastergiou K, Evans I, Ogston N, Miheisi N, Nair D, Kaski JC, et al. Epicardial adipokines in obesity and coronary artery disease induce atherogenic changes in monocytes and endothelial cells. Arterioscler Thromb Vasc Biol. 2010;30(7):1340–6.

    Article  CAS  PubMed  Google Scholar 

  7. Hirata Y, Tabata M, Kurobe H, Motoki T, Akaike M, Nishio C, et al. Coronary atherosclerosis is associated with macrophage polarization in epicardial adipose tissue. J Am Coll Cardiol. 2011;58(3):248–55.

    Article  CAS  PubMed  Google Scholar 

  8. Gaborit B, Venteclef N, Ancel P, Pelloux V, Gariboldi V, Leprince P, et al. Human epicardial adipose tissue has a specific transcriptomic signature depending on its anatomical peri-atrial, peri-ventricular, or peri-coronary location. Cardiovasc Res. 2015;108(1):62–73.

    Article  CAS  PubMed  Google Scholar 

  9. Oikonomou EK, Williams MC, Kotanidis CP, Desai MY, Marwan M, Antonopoulos AS, et al. A novel machine learning-derived radiotranscriptomic signature of perivascular fat improves cardiac risk prediction using coronary CT angiography. Eur Heart J. 2019;40(43):3529–43.

    Article  PubMed  PubMed Central  Google Scholar 

  10. Kotanidis CP, Antoniades C. Perivascular Fat Imaging by Computed Tomography: A Virtual Guide. Br J Pharmacol. 2021. https://doi.org/10.1111/bph.15634.

    Article  PubMed  Google Scholar 

  11. Oikonomou EK, Marwan M, Desai MY, Mancio J, Alashi A, Hutt Centeno E, et al. Non-invasive detection of coronary inflammation using computed tomography and prediction of residual cardiovascular risk (the CRISP CT study): a post-hoc analysis of prospective outcome data. Lancet. 2018;392(10151):929–39.

    Article  PubMed  PubMed Central  Google Scholar 

  12. Mahabadi AA, Berg MH, Lehmann N, Kälsch H, Bauer M, Kara K, et al. Association of epicardial fat with cardiovascular risk factors and incident myocardial infarction in the general population: the Heinz Nixdorf Recall Study. J Am Coll Cardiol. 2013;61(13):1388–95.

    Article  PubMed  Google Scholar 

  13. Chan K, Wahome E, Tsiachristas A, Antonopoulos AS, Patel P, Lyasheva M, et al. Inflammatory risk and cardiovascular events in patients without obstructive coronary artery disease: the ORFAN multicentre, longitudinal cohort study. Lancet. 2024;S0140–6736(24):00596–8.

    Google Scholar 

  14. Gaborit B, Kober F, Jacquier A, Moro PJ, Flavian A, Quilici J, et al. Epicardial fat volume is associated with coronary microvascular response in healthy subjects: a pilot study. Obesity (Silver Spring). 2012;20(6):1200–5.

    Article  CAS  PubMed  Google Scholar 

  15. Mahabadi AA, Lehmann N, Kälsch H, Robens T, Bauer M, Dykun I, et al. Association of epicardial adipose tissue with progression of coronary artery calcification is more pronounced in the early phase of atherosclerosis: results from the Heinz Nixdorf recall study. JACC Cardiovasc Imaging. 2014;7(9):909–16.

    Article  PubMed  Google Scholar 

  16. Li Y, Liu B, Li Y, Jing X, Deng S, Yan Y, et al. Epicardial fat tissue in patients with diabetes mellitus: a systematic review and meta-analysis. Cardiovasc Diabetol. 2019;18(1):3.

    Article  PubMed  PubMed Central  Google Scholar 

  17. Moreno-Santos I, Pérez-Belmonte LM, Macías-González M, Mataró MJ, Castellano D, López-Garrido M, et al. Type 2 diabetes is associated with decreased PGC1α expression in epicardial adipose tissue of patients with coronary artery disease. J Transl Med. 2016;14(1):243.

    Article  PubMed  PubMed Central  Google Scholar 

  18. Ojha S, Fainberg HP, Wilson V, Pelella G, Castellanos M, May ST, et al. Gene pathway development in human epicardial adipose tissue during early life. JCI Insight. 2016;1(13).

    Article  PubMed  PubMed Central  Google Scholar 

  19. Antoniades C, Tousoulis D, Vavlukis M, Fleming I, Duncker DJ, Eringa E, et al. Perivascular adipose tissue as a source of therapeutic targets and clinical biomarkers. Eur Heart J. 2023;44(38):3827–44.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  20. Gaborit B, Kober F, Jacquier A, Moro PJ, Cuisset T, Boullu S, et al. Assessment of epicardial fat volume and myocardial triglyceride content in severely obese subjects: relationship to metabolic profile, cardiac function and visceral fat. Int J Obes (Lond). 2012;36(3):422–30.

    Article  CAS  PubMed  Google Scholar 

  21. Bartoli A, Fournel J, Ait-Yahia L, Cadour F, Tradi F, Ghattas B, et al. Automatic Deep-Learning Segmentation of Epicardial Adipose Tissue from Low-Dose Chest CT and Prognosis Impact on COVID-19. Cells. 2022;11(6):1034.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  22. Gaborit B, Julla JB, Besbes S, Proust M, Vincentelli C, Alos B, et al. Glucagon-like Peptide 1 Receptor Agonists, Diabetic Retinopathy and Angiogenesis: The AngioSafe Type 2 Diabetes Study. J Clin Endocrinol Metab. 2020. https://doi.org/10.1210/clinem/dgz069.

    Article  PubMed  Google Scholar 

  23. Wilkinson CP, Ferris FL, Klein RE, Lee PP, Agardh CD, Davis M, et al. Proposed international clinical diabetic retinopathy and diabetic macular edema disease severity scales. Ophthalmology. 2003;110(9):1677–82.

    Article  CAS  PubMed  Google Scholar 

  24. Fedorov A, Beichel R, Kalpathy-Cramer J, Finet J, Fillion-Robin JC, Pujol S, et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging. 2012;30(9):1323–41.

    Article  PubMed  PubMed Central  Google Scholar 

  25. Kim WH, Kim CG, Kim DW. Optimal CT Number Range for Adipose Tissue When Determining Lean Body Mass in Whole-Body F-18 FDG PET/CT Studies. Nucl Med Mol Imaging. 2012;46(4):294–9.

    Article  PubMed  PubMed Central  Google Scholar 

  26. Oka T, Yamamoto H, Ohashi N, Kitagawa T, Kunita E, Utsunomiya H, et al. Association between epicardial adipose tissue volume and characteristics of non-calcified plaques assessed by coronary computed tomographic angiography. Int J Cardiol. 2012;161(1):45–9.

    Article  PubMed  Google Scholar 

  27. Commandeur F, Goeller M, Razipour A, Cadet S, Hell MM, Kwiecinski J, et al. Fully Automated CT Quantification of Epicardial Adipose Tissue by Deep Learning: A Multicenter Study. Radiol Artif Intell. 2019;1(6).

    Article  PubMed  PubMed Central  Google Scholar 

  28. Commandeur F, Goeller M, Betancur J, Cadet S, Doris M, Chen X, et al. Deep Learning for Quantification of Epicardial and Thoracic Adipose Tissue From Non-Contrast CT. IEEE Trans Med Imaging. 2018;37(8):1835–46.

    Article  PubMed  PubMed Central  Google Scholar 

  29. Doukbi E, Ancel P, Dutour A, Soghomonian A, Ahmed S, Castejon V, et al. Human epicardial fat has a beige profile and contains higher type 2 innate lymphoid cells than subcutaneous fat. Obesity (Silver Spring). 2024. https://doi.org/10.1002/oby.24023.

    Article  PubMed  Google Scholar 

  30. Liu F, Wang H, Feng W, Ye X, Sun X, Jiang C, et al. Type 1 innate lymphoid cells are associated with type 2 diabetes. Diabetes Metab. 2019;45(4):341–6.

    Article  CAS  PubMed  Google Scholar 

  31. Diamond C, Pansini M, Hamid A, Eichert N, Pandya P, Ali SN, et al. Quantitative imaging reveals steatosis and fibro-inflammation in multiple organs in people with type 2 diabetes: a real-world study. Diabetes. 2024. https://doi.org/10.2337/db23-0926.

    Article  PubMed  PubMed Central  Google Scholar 

  32. Donath MY, Dinarello CA, Mandrup-Poulsen T. Targeting innate immune mediators in type 1 and type 2 diabetes. Nat Rev Immunol. 2019;19(12):734–46.

    Article  CAS  PubMed  Google Scholar 

  33. Ridker PM, Everett BM, Thuren T, MacFadyen JG, Chang WH, Ballantyne C, et al. Antiinflammatory Therapy with Canakinumab for Atherosclerotic Disease. N Engl J Med. 2017;377(12):1119–31.

    Article  CAS  PubMed  Google Scholar 

  34. Raggi P, Gadiyaram V, Zhang C, Chen Z, Lopaschuk G, Stillman AE. Statins Reduce Epicardial Adipose Tissue Attenuation Independent of Lipid Lowering: A Potential Pleiotropic Effect. J Am Heart Assoc. 2019;8(12).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  35. Greenstein AS, Khavandi K, Withers SB, Sonoyama K, Clancy O, Jeziorska M, et al. Local inflammation and hypoxia abolish the protective anticontractile properties of perivascular fat in obese patients. Circulation. 2009;119(12):1661–70.

    Article  CAS  PubMed  Google Scholar 

  36. Azul L, Leandro A, Boroumand P, Klip A, Seiça R, Sena CM. Increased inflammation, oxidative stress and a reduction in antioxidant defense enzymes in perivascular adipose tissue contribute to vascular dysfunction in type 2 diabetes. Free Radic Biol Med. 2020;146:264–74.

    Article  CAS  PubMed  Google Scholar 

  37. Cosson E, Nguyen MT, Rezgani I, Tatulashvili S, Sal M, Berkane N, et al. Epicardial adipose tissue volume and coronary calcification among people living with diabetes: a cross-sectional study. Cardiovasc Diabetol. 2021;20(1):35.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  38. Cosson E, Nguyen MT, Rezgani I, Berkane N, Pinto S, Bihan H, et al. Epicardial adipose tissue volume and myocardial ischemia in asymptomatic people living with diabetes: a cross-sectional study. Cardiovasc Diabetol. 2021;20(1):224.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  39. Wang Z, Huang Z, Wang X. The Correlation between Region-specific Epicardial Adipose Tissue and Myocardial Ischemia Defined by CT-FFR in Type 2 Diabetes Mellitus Patients. Curr Med Imaging. 2024. https://doi.org/10.2174/0115734056291770240105101731.

    Article  PubMed  Google Scholar 

  40. Sardu C, D’Onofrio N, Torella M, Portoghese M, Loreni F, Mureddu S, et al. Pericoronary fat inflammation and Major Adverse Cardiac Events (MACE) in prediabetic patients with acute myocardial infarction: effects of metformin. Cardiovasc Diabetol. 2019;18(1):126.

    Article  PubMed  PubMed Central  Google Scholar 

  41. Wang TD, Lee WJ, Shih FY, Huang CH, Chang YC, Chen WJ, et al. Relations of epicardial adipose tissue measured by multidetector computed tomography to components of the metabolic syndrome are region-specific and independent of anthropometric indexes and intraabdominal visceral fat. J Clin Endocrinol Metab. 2009;94(2):662–9.

    Article  CAS  PubMed  Google Scholar 

  42. Gaborit B, Dutour A. Looking beyond ectopic fat amount: A SMART method to quantify epicardial adipose tissue density. Eur J Prev Cardiol. 2017;24(6):657–9.

    Article  PubMed  Google Scholar 

  43. Christensen RH, von Scholten BJ, Hansen CS, Jensen MT, Vilsbøll T, Rossing P, et al. Epicardial adipose tissue predicts incident cardiovascular disease and mortality in patients with type 2 diabetes. Cardiovasc Diabetol. 2019;18(1):114.

    Article  PubMed  PubMed Central  Google Scholar 

Download references

Acknowledgements

The authors thank the Unité de Recherche Clinique of Lariboisière/Fernand Widal (Pr Eric Vicaut), the Délégation à la Recherche Clinique de Paris– Ile de France and the Direction de la Recherche Santé de Marseille for their administrative support. The authors also thank Nassima Haddadi, Djamila Bellili, Nathalie Lesavre, Chirine Djebaili, for their technical support, nurses and patients who acccepted to be involved in this study.

Funding

This study was funded by ANR-DGOS (Project AngioSafe T2D).

Author information

Authors and Affiliations

  1. Aix Marseille Univ, INSERM, INRAE, C2VN, Marseille, France

    Bénédicte Gaborit, Patricia Ancel, Astrid Soghomonian, Camille Deprade, Adèle Lasbleiz, Florian Mourre, Patrice Darmon & Anne Dutour

  2. Department of Endocrinology, Metabolic Diseases and Nutrition, Pôle ENDO, Chemin des Bourrely, APHM, Hôpital Nord, 13915 Marseille Cedex 20, Marseille, France

    Bénédicte Gaborit, Astrid Soghomonian, Camille Deprade, Adèle Lasbleiz, Florian Mourre, Patrice Darmon & Anne Dutour

  3. IMMEDIAB Laboratory, Institut Necker Enfants Malades (INEM), CNRS UMR 8253, INSERM U1151, Université Paris Cité, 75015, Paris, France

    Jean Baptiste Julla, Nicolas Venteclef, Louis Potier & Jean François Gautier

  4. Diabetology and Endocrinology Department, Féderation de Diabétologie, Université Paris Cité, Lariboisière Hospital, APHP, 75015, Paris, France

    Jean Baptiste Julla & Jean François Gautier

  5. Aix Marseille Univ, CNRS, CRMBM, Marseille, France

    Joris Fournel, Adèle Lasbleiz, Axel Bartoli & Alexis Jacquier

  6. Medical Evaluation Department, Assistance-Publique Hôpitaux de Marseille, CIC-CPCET, 13005, Marseille, France

    Marie Houssays

  7. Aix Marseille School of Economics, Aix Marseille University, CNRS, Marseille, France

    Badih Ghattas

  8. Ophtalmology Department, Assistance-Publique Hôpitaux de Marseille, Aix-Marseille Univ, 13005, Marseille, France

    Maud Righini

  9. Centre Monticelli Paradis, 433 Bis Rue Paradis, 13008, Marseille, France

    Pierre Gascon & Frédéric Matonti

  10. National Center for Scientific Research (CNRS), Timone Neuroscience Institute (INT), Aix Marseille Univ, 13008, Marseille, France

    Frédéric Matonti

  11. Diabetology and Endocrinology Department, Fédération de Diabétologie, Bichat Hospital, Paris, France

    Louis Potier

  12. Support Unit for Clinical Research and Economic Evaluation, Assistance Publique-Hôpitaux de Marseille, 13385, Marseille, France

    Noémie Resseguier

  13. Aix-Marseille Univ, EA 3279 CEReSS-Health Service Research and Quality of Life Center, Marseille, France

    Noémie Resseguier

  14. Department of Radiology, Hôpital de la TIMONE, AP-HM, Marseille, France

    Axel Bartoli & Alexis Jacquier

Authors
  1. Bénédicte Gaborit
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Jean Baptiste Julla
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Joris Fournel
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Patricia Ancel
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Astrid Soghomonian
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Camille Deprade
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Adèle Lasbleiz
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Marie Houssays
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Badih Ghattas
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Pierre Gascon
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Maud Righini
    View author publications

    You can also search for this author in PubMed Google Scholar

  12. Frédéric Matonti
    View author publications

    You can also search for this author in PubMed Google Scholar

  13. Nicolas Venteclef
    View author publications

    You can also search for this author in PubMed Google Scholar

  14. Louis Potier
    View author publications

    You can also search for this author in PubMed Google Scholar

  15. Jean François Gautier
    View author publications

    You can also search for this author in PubMed Google Scholar

  16. Noémie Resseguier
    View author publications

    You can also search for this author in PubMed Google Scholar

  17. Axel Bartoli
    View author publications

    You can also search for this author in PubMed Google Scholar

  18. Florian Mourre
    View author publications

    You can also search for this author in PubMed Google Scholar

  19. Patrice Darmon
    View author publications

    You can also search for this author in PubMed Google Scholar

  20. Alexis Jacquier
    View author publications

    You can also search for this author in PubMed Google Scholar

  21. Anne Dutour
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

BG, JBJ, AD, AJ, JFG contributed to the conception; BG, PA, PD, NR participated in the design of the work; BG, JBJ, AS, CD, AL, MH, PG, MR, FM, LP, JFG, AD, PD, FM participated in the inclusion of patients; JBJ, PA, MH, NV, LP, JFG, NR contributed to the acquisition, analysis or interpretation of data; JF, BGh, AB and AJ created the AI software used in the work; BG and PA drafted the work and prepared figures; JBJ, NV, JFG,LP,NR, AJ, FM substantively revised it AND all the authors have approved the submitted version.

Corresponding author

Correspondence to Bénédicte Gaborit.

Ethics declarations

Competing interests

The authors declare no competing interests.

Data access and responsibility

BG is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Gaborit, B., Julla, J.B., Fournel, J. et al. Fully automated epicardial adipose tissue volume quantification with deep learning and relationship with CAC score and micro/macrovascular complications in people living with type 2 diabetes: the multicenter EPIDIAB study. Cardiovasc Diabetol 23, 328 (2024). https://doi.org/10.1186/s12933-024-02411-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12933-024-02411-y

Keywords

Cardiovascular Diabetology

ISSN: 1475-2840