Machine Learning Computed Tomography Radiomics of Abdominal Adipose Tissue to Optimize Cardiovascular Risk Assessment

Machine Learning Computed Tomography Radiomics of Abdominal Adipose Tissue to Optimize Cardiovascular Risk Assessment | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Machine Learning Computed Tomography Radiomics of Abdominal Adipose Tissue to Optimize Cardiovascular Risk Assessment Jennifer Mancio, Alice Lopes, Inês Sousa, Fabio Nunes, Sonia Xara, and 8 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4654020/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Background Subcutaneous (SAF) and visceral (VAF) abdominal fat have specific properties which the global body fat and total abdominal fat (TAF) size metrics do not capture. Beyond size, radiomics allows deep tissue phenotyping and may capture fat dysfunction. We aimed to characterize the computed tomography (CT) radiomics of SAF and VAF and assess their incremental value above fat size to detect coronary calcification. Methods SAF, VAF and TAF area, signal distribution and texture were extracted from non-contrast CT of 1001 subjects (57% male, 57 ± 10 years) with no established cardiovascular disease who underwent CT for coronary calcium score (CCS) with additional abdominal slice (L4/5-S1). XGBoost machine learning models (ML) were used to identify the best features that discriminate SAF from VAF and to train/test ML to detect any coronary calcification (CCS > 0). Results SAF and VAF appearance in non-contrast CT differs: SAF displays brighter and finer texture than VAF. Compared with CCS = 0, SAF of CCS > 0 has higher signal and homogeneous texture, while VAF of CCS > 0 has lower signal and heterogeneous texture. SAF signal/texture improved SAF area performance to detect CCS > 0. A ML including SAF and VAF area performed better than TAF area to discriminate CCS > 0 from CCS = 0, however, a combined ML of the best SAF and VAF features detected CCS > 0 as the best TAF features. Conclusion In non-contrast CT, SAF and VAF appearance differs and SAF radiomics improves the detection of CCS > 0 when added to fat area; TAF radiomics (but not TAF area) spares the need for separate SAF and VAF segmentations. Health sciences/Diseases/Cardiovascular diseases Health sciences/Risk factors Subcutaneous abdominal fat visceral abdominal fat abdominal obesity radiomics radiomic analysis computed tomography cardiovascular risk coronary calcification Figures Figure 1 Figure 2 Figure 3 INTRODUCTION Accumulation of abdominal fat is a well-established adverse metabolic and cardiovascular (CV) risk factor (1–4) . Two metabolic and immunologically distinct compartments of the total abdominal fat (TAF) – the subcutaneous and visceral abdominal fat (SAF and VAF), have been recognized as independent predictors of cardiometabolic diseases (1, 5–9) . In clinical practice, the assessment of obesity involves non-imaging biomarkers, such as body mass index (BMI) and waist circumference (WC) (10–12) , and imaging biomarkers, such as the fat depot amount measured by magnetic resonance imaging and computed tomography (CT) (5, 13, 14) . The susceptibility to obesity related metabolic complications is not automatically mediated by overall body fat mass (15) and most investigation in the field of ectopic fat has centred on the importance of VAF quantity evaluation. It is known that VAF depots are associated with a proinflammatory and a proatherogenic risk profile, but VAF only contributes to 15% of the total circulation of systemic fatty acids. The majority of fatty acids originate from extra peritoneal adipose tissue, probably from SAF (16) . More recently, the radiodensity of SAF and VAF by CT were associated with cardiometabolic risk (17) , subclinical atherosclerosis (18) , hypertension, insulin resistance and metabolic syndrome independent of fat volume (19) . Beyond CT attenuation, radiomic analysis enables in-depth characterization of the signal distribution and spatial relationships between voxels depicting underlying microstructural tissue differences undetectable by its volume and attenuation (20) . These radiomic features have been investigated as biomarkers of the tissue biology in histopathology, genomics, transcriptomics and proteomics studies (21, 22) . In a previous study of a large cohort of subjects with no prior CV diseases who underwent non-contrast CT, we showed that with aging SAF amount (assessed by its area) decreases while and VAF depot enlarges, and coronary artery calcification worsens suggesting that losing SAF ability to expand and store fat away from the internal organs is a crucial determinant that anticipate the shift to an adverse metabolic and pro-atherosclerotic environment. Detecting the turning point at which SAF becomes dysfunctional before fat starts to accumulate in the VAF compartment is essential to modify the individual patient risk. In the same cohort, we subsequently performed a comprehensive radiomic phenotyping of SAF and VAF by extracting thousands of features to reveal which is the non-contrast CT differential appearance of SAF and VAF, the adverse profile of SAF and VAF that is associated with and add value to fat area to detect coronary calcification. SUBJECTS AND METHODS Study design, terminology and data analysis followed the recommendations of the Quantitative Imaging Biomarker Alliance (QIBA), and methods are reported as per the aspects of the Radiomics Quality Score (RQS) according to Lambin et al . (23) ( Supplementary Table S1 ). Study Population Prospective registry of patients who underwent a coronary CT angiography (CCTA) in a tertiary centre from January 2008 to July 2016. In this analysis, only subjects with no prior CV diseases (i.e., known coronary artery disease, prior stroke, prior valvular surgery or atrial fibrillation), and with coronary calcium score (CCS) and abdominal fat images were included. All patients were assessed by a nutritionist for anthropometric assessment on the day of the CT scan. CV risk factors and medication history were recorded based on the information available on the electronic patient record. All participants provided written informed consent, and the study was approved by the Institutional Ethics Committee. Computed Tomography and Coronary Calcium Score CT scanning was performed in a 64-slice-scanner (Somatom Sensation Cardiac 64, Siemens, Forchheim, Germany) following the institutional protocol for CCTA which at our institution included non-contrast CT images (for CCS and abdominal fat assessment) followed by iodine contrast injection for CCTA. CCS imaging was performed using a prospectively ECG triggered scanning protocol, subsequently, followed by an abdominal single slice acquisition (as detailed elsewhere (24) ). CCS was reported using the Agatston method. Abdominal Fat Radiomic Features Figure 1 illustrates the abdominal fat regions from where the radiomic analysis was performed following Pyradiomics. Radiomic features were extracted from tissue with mean attenuation values compatible with fat ranging from − 150 to -50 HU located in the TAF, SAF and VAF compartments. TAF corresponds to all adipose tissue located in the acquired L4/5-S1 slice, VAF to the mask created by a manual segmentation of the abdominal muscular wall, and SAF compartment corresponds to the region located between the TAF and VAF masks (i.e., obtained by subtracting the VAF from TAF masks). Signal dependent fat features include: the First order (Statistics) and Texture features (25) . First order features are histogram-based features and describe fat attenuation (gray-level) values distribution. Textural features describe the spatial relationship between neighbouring voxels gray-level values and measure adipose tissue heterogeneity and coarseness based on five matrices: gray level co-occurrence matrix (GLCM), gray level dependence matrix (GLDM), gray level run-length matrix (GLRLM), gray level size zone matrix (GLSZM) and neighbouring gray tone differences matrix (NGTDM). Size/shape features (including, the classical TAF, SAF and VAF areas) are signal intensity-independent features and describe the 2D- and 3D-geometric properties of the fat mask (26) . First order and textural features can be extracted from original images and after further application of image filters (Gradient, Laplacian of Gaussian, Wavelet High-High, Wavelet Low-Low and Wavelet High-Low); shape/size features are extracted from original images only. In total, we extracted 665 first order, textural and shape fat radiomic features (Supplementary Tables S2 and S3). Data Analysis We used STATA software (version 13.1, StataCorp LP, Texas, US) for statistical analysis, and R environment and Scikit Learn of Python (version 3.7) for radiomic analysis; radiomic features were normalized using Z-score prior to analysis (27) . Patients’ characteristics were summarized as number (percentages) or mean (standard deviation) for categorical and continuous variables, respectively (unless specified otherwise). Between-group comparisons were performed using chi-square test for categorical variables and T-test or Mann-Whitney’s U test for continuous variables. Unique radiomic profile of SAF and VAF Orthogonal Projection to Latent Structures-Discriminant Analysis (OPLS-DA) was used to identify the most different CT radiomic features between SAF and VAF (28) . The main CT radiomic features were selected based on the regression coefficients (i.e., absolute value of regression coefficient greater than 0.1). SAF and VAF radiomic signature of CCS > 0 An XGBoost ensemble machine learning (ML) model in an internal 10-cross-validation method was employed to identify the SAF, VAF and TAF CT features that discriminate CCS > 0 from CCS = 0 patients. The most important radiomic features (i.e., higher mean decrease in Gini impurity) were selected to build a SAF, VAF and TAF radiomic signature of CCS > 0. Finally, we combined the best performing SAF, VAF and TAF radiomic signature features into a multivariate XGBoost ML model to separate CCS of zero and CCS > 0 patients; model performance was assessed using receiver operating characteristics (ROC) curve parameters, and bootstrapping (100 samples) was used to calculate the confidence interval for the area under the curve (AUC). Improvement in classification of fat texture above the respective fat area alone was assessed using an Integrated Discrimination Improvement (IDI) analysis. RESULTS Study Population In this analysis, we included 1001 subjects, 57% male, 57 ± 10 years old, 59% had arterial hypertension and 52% had dyslipidaemia, 15% were diabetic. Mean BMI was 28 ± 5 kg/m 2 with a mean SAF of 223 ± 107cm 2 and mean VAF of 128 ± 69cm 2 . Nearly half (47%) had no calcification in their coronary arteries (CCS of zero) and 7% had CCS > 300. Table 1 summarizes the characteristics of study population. Radiomics Phenotyping of Subcutaneous and Visceral Fat Identification of the most relevant CT radiomic features to discriminate SAF from VAF depots revealed that SAF appearance in non-contrast CT images as compared with VAF was characterized by: Greater proportion of larger size zones with higher gray-level values (Wavelet_LL_GLDM_Large Dependence High Gray Level Emphasis, Wavelet_LL_GLSZM_Large Area High Gray Level Emphasis, Wavelet_LL_GLRLM_Long Run High Gray Level Emphasis, and Wavelet_LL_GLSZM_High Gray Level Zone Emphasis received higher values in SAF compared with VAF). Heterogeneous (non-uniform) texture (Wavelet_HH_GLDM_Gray Level Non-Uniformity _Gray_Level_Non_Uniformity received higher values in SAF than in VAF), and Fine texture (Wavelet_LL_GLCM_Autocorrelation received higher values in SAF than in VAF) Overall, SAF radiomic CT phenotype was characterized in the image by a greater proportion of larger zones with higher gray level values heterogeneous and finer texture compared with VAF ( Fig. 2 and Supplementary Table S4). Fat Radiomic Signature of Coronary Classification Compared with CCS of zero, CCS > 0 subjects had lower amount of SAF area which was characterized by higher values of fat attenuation (90th percentile signal intensity was higher in CCS > 0 indicating (signal distribution shifted to the right) with a finer and uniform texture compared with SAF of CCS = 0 subjects. The most important features of VAF that discriminate CCS > 0 from CCS = 0 were shape features indicating accumulation of more VAF in CCS > 0 subjects (larger 2D-area) along larger maximum and minimum abdominal dimensions (Original_Shape_Minor Axis Length and Original_Shape_Minor Axis Length were higher in CCS > 0) in an spherical-like shape abdomen (Original_Shape_Elongation and Original_Shape_Area Volume Surface Ratio were lower in CCS > 0); opposite to SAF, VAF of CCS > 0 patients displayed lower signal intensity with a higher proportion of lower gray level values areas and a non-uniform and heterogenous texture (Table 2) . Incremental Value of Fat Texture above Fat Area to Detect Coronary Calcification A ML model including SAF area alone detected the presence of coronary calcification correctly in 58% of the patients with an AUC of 0.62 (95% CI: 0.58–0.66); by adding the SAF texture signature features to SAF area, there was a significant improvement in the classification of patients providing an AUC of 0.70 (95% CI: 0.66–0.74) and an accuracy score of 65% with an IDI of 0.02 (95% CI: 0.01–0.03; p 0 (IDI: 0.001; 95% CI: 0.002–0.003; p = 0.128) ( Fig. 3 -B ) . A model including the best SAF and VAF features performed better than a combined model of SAF and VAF indexed area (p < 0.0001). There was no significant difference in the performance of the models including the best SAF texture features and VAF texture features (AUC with SAF texture 0.70 (95% CI: 0.68–0.74) vs AUC with VAF texture 0.69 (95% CI: 0.65–0.72), p = 0.08), although there was a trend for a better performance with SAF texture features. A ML model including both the SAF and VAF indexed areas yielded an AUC of 0.68 (95% CI: 0.65–0.71) which is significantly higher than the performance of indexed TAF area alone (i.e., TAF area with no discrimination of fat compartments) (AUC 0.53; 95% CI: 0.47–0.59) ( Fig. 3 -C ) . However, a model including the best performing SAF and VAF texture features that discriminate patients with coronary calcification provided a similar accuracy than a model including the best texture features extracted from any abdominal fat; AUC with best SAF and best VAF features was 0.71 (95% CI: 0.66–0.76) vs AUC with best TAF features was 0.69 (95% CI: 0.66–0.73) ( Fig. 3 -D ) . DISCUSSION In this study, we combined radiomics methodology with ML to characterize the abdominal fat compartments and stratify CV risk as assessed by CCS. We were able to detect abdominal fat gray-level and texture differences between SAF and VAF depots supporting the concept of their intrinsically distinct biological properties, and between SAF and VAF of CCS > 0 and CCS = 0 subjects. We demonstrated incremental value of abdominal fat texture above its area to detect coronary calcification, and, differently than TAF area, automatic radiomic feature extraction from any abdominal fat tissue may eliminate the need for SAF and VAF individual segmentation thereby simplifying the radiomics workflow for individualized risk assessment. Globally, obesity and abdominal adiposity have been associated with a wide range of metabolic complications, particularly CV disease (29) . Although the accumulation of VAF is one of the main contributors to quantifying cardiometabolic risk above BMI, the heterogeneity of body composition makes it difficult to assess fat distribution in clinical practice. Radiomics, first described in 2012 (30) , enables high-throughput extraction of quantitative features from medical images and describes tissue heterogeneity captured in visually unrecognizable voxel grey-level intensities (30–32) . Initially applied to the oncology field, radiomics allow tumour phenotyping and identification of imaging features related to poor prognosis. In the CV field, studies in magnetic resonance imaging showed accurate distinction between subacute and chronic myocardial infarction, detecting of myocardial fibrosis without gadolinium and cardiomyopathy differential diagnosis (33, 34) . CT radiomics detected high-risk plaques better than the classical features and coronary inflammation by 3D phenotyping of perivascular fat attenuation. The CT attenuation of adipose tissue may indicate some tissue characteristics, including (18) : ( 1 ) more negative HU values (i.e.; low gray-level values) are associated with more lipid-dense fat tissue and poorly vascularized adipose tissue resulting from the radiographic properties of the blood, while ( 2 ) less negative HU values (i.e.; high gray-level values) can result from tissue oedema/inflammation (i.e. increased water content), lower lipid content and fibrotic adipose tissue due to excessive collagen deposition, and ( 3 ) heterogenous texture can be explained by cellular composition, microvasculature, metabolic characteristics and extracellular matrix composition (35) , which wavelet decompositions can capture discontinuities. In this study we found greater proportion of larger zones with higher gray-level values (i.e. less negative HU) and heterogeneous texture characterizing the SAF depot, which suggests increased inflammation, fibrosis, changes in vascularization and oxidative stress (36–38) . Anatomically, the SAF compartment is subdivided into superficial SAF and deep SAF separated by a fascial plane (fascia superficialis). The two sub compartments present different risk profiles whose understanding depends on the precise quantification of each of them. Thus, our observation might represent the deep SAF, which contains larger, less organized and more vascularized lobules (39) , related to the onset of adipose tissue dysfunction, impaired glucose metabolism, hyperinsulinemia and insulin resistance (40) . The correct distinction between deep SAF and superficial SAF becomes a challenge because, the difference in volume between the two compartments decreases as fat accumulation increases (making it harder to differentiate in individuals with a high amount of fat); moreover, SAF accumulation profile changes along the lumbar levels. In some regions (L1-L2), the superficial SAF and deep SAF amounts are equal, with prominent superficial fascia visible through MRI; however, the caudal progression to L5 makes it difficult to distinguish, with multiple fascial lines (41) . CT-based texture captured regional differences between abdominal fat compartments. Juan Shi et al (42) showed that, in a Chinese population, heterogeneous texture features extracted from VAF were significantly associated with metabolic syndrome and related disorders. In adipose tissue, the extracellular matrix plays an important role in tissue expansion and angiogenesis. Adipose tissue expansion depends on extracellular matrix remodeling through cycles of collagen deposition. When adipose tissue expansion becomes dysfunctional, excessive and unregulated accumulation of collagen and other extracellular matrix components results in fibrosis, which limits the expansion capacity of adipocytes (43) . Anthropometric indices of obesity are easily implemented, but newer imaging-based methods offer greater sensitivity and specificity for measuring specific deposits. Overall, in subjects with coronary calcification, our imaging technique captured pathogenic features in the two compartments of abdominal fat even before the average BMI value reached the obesity threshold. The area of adipose tissue quantified from a slice image has strong correlations with the total volume of abdominal adipose tissue (44) and is associated with underlying metabolic disorders related to obesity. By contrast, TAF area performs very poorly and is unable to detect coronary calcification. This demands individual segmentation of VAF and SAF subtraction from TAF. In this work, we demonstrated additional classification value of fat texture above fat area to detecting coronary calcification. Moreover, a ML model of the best SAF combined with best VAF radiomic features that discriminate CCS > 0 from CCS = 0 subjects performed similarly as the model including the best TAF radiomic features (i.e. features extracted from any fat tissue located in the obtained slice). Because TAF can be detected in CT without the need for contrast in a single slice at L4 and L5-S1 with a minimal estimated radiation exposure of 0.06 mSv, automatic feature extraction from TAF using a threshold method only, and without the need for segmentation of SAF and VAF, can serve to find new imaging biomarkers of obesity and build improved and easy to implement individual risk prediction models. Limitations Biological interpretability of radiomic features remains challenging; in this study, we did not correlate imaging findings with fat samples or analytical parameters. Our study design is cross-sectional with CT performed in a single centre; ML models were trained and tested in the entire sample size to optimize data learning and minimize overfitting; external validation was not tested. The segmentation of the compartments was performed manually, which is human dependent and can be a source of variability affecting radiomics reproducibility. The radiomic features were extracted from two-dimensional images and the abdominal distribution of adiposity showed interindividual local variability, which we were not able to investigate. Epicardial adipose tissue which represents an important cardiometabolic risk factor was not analysed. Main conclusions In non-contrast CT, SAF and VAF appearance differs and abdominal fat tissue radiomic profile is associated with CV risk. Radiomic analysis of TAF can derive new imaging biomarkers of obesity, which can improve and facilitate individual risk assessment. Abbreviations AUC Area under the curve BMI Body mass index CCS Coronary calcium score CCTA Coronary computed tomography angiography CT Computed tomography CV Cardiovascular ML Machine learning ROC Receiving operating characteristics SAF Subcutaneous abdominal fat VAF Visceral abdominal fat TAF Total abdominal fat WC Waist circumference Declarations Competing Interests The authors declare no competing financial interests Conflict of interest The authors declare no conflicts of interest. Author Contributions AL contributed with data collection and database cleaning, results analysis, and manuscript writing. IS segmented the abdominal to define the region of interest. FN collected CT imaging raw data. SX performed the anthropometric assessment. MC and WF performed the computed tomography and contributed with patient recruitment. NF, VGR, NB and RFC contributed with manuscript revision. AB contributed with statistical data analysis and results interpretation. JP performed the radiomic feature extraction and revised the manuscript. JM designed the study, contributed with data analysis, results interpretation, wrote and revised the manuscript. DATA AVAILABILITY The data that support the findings of this study are available from [third party name] but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Faculty of Medicine of the University of Porto. References Després J-P, Lemieux I, Bergeron J, Pibarot P, Mathieu P, Larose E, et al. Abdominal obesity and the metabolic syndrome: contribution to global cardiometabolic risk. Arteriosclerosis, thrombosis, and vascular biology. 2008;28(6):1039-49. Kwon H, Kim D, Kim JS. Body fat distribution and the risk of incident metabolic syndrome: a longitudinal cohort study. Scientific reports. 2017;7(1):10955. Hu HH, Chen J, Shen W. Segmentation and quantification of adipose tissue by magnetic resonance imaging. Magnetic Resonance Materials in Physics, Biology and Medicine. 2016;29:259-76. Kumari R, Kumar S, Kant R. An update on metabolic syndrome: Metabolic risk markers and adipokines in the development of metabolic syndrome. Diabetes & Metabolic Syndrome: Clinical Research & Reviews. 2019;13(4):2409-17. Ibrahim MM. Subcutaneous and visceral adipose tissue: structural and functional differences. Obesity reviews. 2010;11(1):11-8. Singh S, Dulai PS, Zarrinpar A, Ramamoorthy S, Sandborn WJ. Obesity in IBD: epidemiology, pathogenesis, disease course and treatment outcomes. Nature reviews Gastroenterology & hepatology. 2017;14(2):110-21. Pickhardt PJ, Graffy PM, Perez AA, Lubner MG, Elton DC, Summers RM. Opportunistic screening at abdominal CT: use of automated body composition biomarkers for added cardiometabolic value. RadioGraphics. 2021;41(2):524-42. Ritchie S, Connell J. The link between abdominal obesity, metabolic syndrome and cardiovascular disease. Nutrition, Metabolism and cardiovascular diseases. 2007;17(4):319-26. Elffers TW, de Mutsert R, Lamb HJ, de Roos A, Willems van Dijk K, Rosendaal FR, et al. Body fat distribution, in particular visceral fat, is associated with cardiometabolic risk factors in obese women. PloS one. 2017;12(9):e0185403. Graffy PM, Pickhardt PJ. Quantification of hepatic and visceral fat by CT and MR imaging: relevance to the obesity epidemic, metabolic syndrome and NAFLD. The British journal of radiology. 2016;89(1062):20151024. Update AS. Heart disease and stroke statistics—2020 update: a report from the American Heart Association. Circulation. 2020;141(9):e139-e596. Perreault L. Obesity in adults: prevalence, screening, and evaluation. Waltham, MA, USA: UpToDate Inc. 2018. Mei Z, Grummer-Strawn LM, Pietrobelli A, Goulding A, Goran MI, Dietz WH. Validity of body mass index compared with other body-composition screening indexes for the assessment of body fatness in children and adolescents. The American journal of clinical nutrition. 2002;75(6):978-85. Ortega FB, Sui X, Lavie CJ, Blair SN, editors. Body mass index, the most widely used but also widely criticized index: would a criterion standard measure of total body fat be a better predictor of cardiovascular disease mortality? Mayo Clinic Proceedings; 2016: Elsevier. Zhang X, Ha S, Lau HC-H, Yu J, editors. Excess body weight: Novel insights into its roles in obesity comorbidities. Seminars in Cancer Biology; 2023: Elsevier. Piché M-E, Poirier P, Lemieux I, Després J-P. Overview of epidemiology and contribution of obesity and body fat distribution to cardiovascular disease: an update. Progress in cardiovascular diseases. 2018;61(2):103-13. Shah RV, Allison M, Lima J, Abbasi S, Eisman A, Lai C, et al. Abdominal fat radiodensity, quantity and cardiometabolic risk: The Multi-Ethnic Study of Atherosclerosis. Nutrition, Metabolism and Cardiovascular Diseases. 2016;26(2):114-22. Alvey NJ, Pedley A, Rosenquist KJ, Massaro JM, O'Donnell CJ, Hoffmann U, et al. Association of fat density with subclinical atherosclerosis. Journal of the American Heart Association. 2014;3(4):e000788. Rosenquist KJ, Pedley A, Massaro JM, Therkelsen KE, Murabito JM, Hoffmann U, et al. Visceral and subcutaneous fat quality and cardiometabolic risk. JACC: Cardiovascular Imaging. 2013;6(7):762-71. Wang Z, Zhang J, Zhang A, Sun Y, Su M, You H, et al. The role of epicardial and pericoronary adipose tissue radiomics in identifying patients with non-ST-segment elevation myocardial infarction from unstable angina. Heliyon. 2023;9(5). Lambin P, Leijenaar RT, Deist TM, Peerlings J, De Jong EE, Van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nature reviews Clinical oncology. 2017;14(12):749-62. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278(2):563-77. Lambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nature reviews Clinical oncology. 2017;14(12):749-62. Mancio J, Fonseca P, Figueiredo B, Ferreira W, Carvalho M, Ferreira N, et al. Association of body mass index and visceral fat with aortic valve calcification and mortality after transcatheter aortic valve replacement: the obesity paradox in severe aortic stenosis. Diabetology & Metabolic Syndrome. 2017;9:1-10. van Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer research. 2017;77(21):e104-e7. Raisi-Estabragh Z, Izquierdo C, Campello VM, Martin-Isla C, Jaggi A, Harvey NC, et al. Cardiac magnetic resonance radiomics: basic principles and clinical perspectives. Eur Heart J Cardiovasc Imaging. 2020;21(4):349-56. Haga A, Takahashi W, Aoki S, Nawa K, Yamashita H, Abe O, et al. Standardization of imaging features for radiomics analysis. The journal of medical investigation : JMI. 2019;66(1.2):35-7. J. Trygg SW. Orthogonal projections to latent structures (O-PLS). Journal of Chemometrics. 2002;16:119–28. Haffner SM. Relationship of metabolic risk factors and development of cardiovascular disease and diabetes. Obesity. 2006;14(S6):121S-7S. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, Van Stiphout RG, Granton P, et al. Radiomics: extracting more information from medical images using advanced feature analysis. European journal of cancer. 2012;48(4):441-6. Aerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nature communications. 2014;5(1):4006. Kumar V, Gu Y, Basu S, Berglund A, Eschrich SA, Schabath MB, et al. Radiomics: the process and the challenges. Magnetic resonance imaging. 2012;30(9):1234-48. Baessler B, Mannil M, Oebel S, Maintz D, Alkadhi H, Manka R. Subacute and Chronic Left Ventricular Myocardial Scar: Accuracy of Texture Analysis on Nonenhanced Cine MR Images. Radiology. 2018;286(1):103-12. Baessler B, Mannil M, Maintz D, Alkadhi H, Manka R. Texture analysis and machine learning of non-contrast T1-weighted MR images in patients with hypertrophic cardiomyopathy-Preliminary results. European journal of radiology. 2018;102:61-7. Lee M-J, Wu Y, Fried SK. Adipose tissue heterogeneity: implication of depot differences in adipose tissue for obesity complications. Molecular aspects of medicine. 2013;34(1):1-11. Kim S-H, Chung J-h, Song S-W, Jung WS, Lee Y-A, Kim H-N. Relationship between deep subcutaneous abdominal adipose tissue and metabolic syndrome: a case control study. Diabetology & metabolic syndrome. 2016;8:1-9. Oikonomou EK, Antoniades C. The role of adipose tissue in cardiovascular health and disease. Nature Reviews Cardiology. 2019;16(2):83-99. Crewe C, An YA, Scherer PE. The ominous triad of adipose tissue dysfunction: inflammation, fibrosis, and impaired angiogenesis. The Journal of clinical investigation. 2017;127(1):74-82. Marinou K, Hodson L, Vasan SK, Fielding BA, Banerjee R, Brismar K, et al. Structural and functional properties of deep abdominal subcutaneous adipose tissue explain its association with insulin resistance and cardiovascular risk in men. Diabetes care. 2014;37(3):821-9. Hoffstedt J, Arner E, Wahrenberg H, Andersson D, Qvisth V, Löfgren P, et al. Regional impact of adipose tissue morphology on the metabolic profile in morbid obesity. Diabetologia. 2010;53:2496-503. Bhanu PK, Arvind CS, Yeow LY, Chen WX, Lim WS, Tan CH. CAFT: a deep learning-based comprehensive abdominal fat analysis tool for large cohort studies. Magnetic Resonance Materials in Physics, Biology and Medicine. 2022:1-16. Shi J, Bao G, Hong J, Wang S, Chen Y, Zhao S, et al. Deciphering CT texture features of human visceral fat to evaluate metabolic disorders and surgery-induced weight loss effects. EBioMedicine. 2021;69. Morais JBS, da Silva Dias TM, Cardoso BEP, de Paiva Sousa M, de Sousa TGV, de Araújo DSC, et al. Adipose tissue dysfunction: impact on metabolic changes? Hormone and Metabolic Research. 2022;54(12):785-94. Hsu L-Y, Ali Z, Bagheri H, Huda F, Redd BA, Jones EC. Comparison of CT and Dixon MR Abdominal Adipose Tissue Quantification Using a Unified Computer-Assisted Software Framework. Tomography. 2023;9(3):1041-51. Additional Declarations There is NO conflict of interest to disclose Supplementary Files Supplementaryfiles.pdf Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: revise 12 Nov, 2025 Review # 2 received at journal 26 Oct, 2025 Reviewer # 2 agreed at journal 14 Oct, 2025 Review # 1 received at journal 14 Jul, 2025 Reviewer # 1 agreed at journal 01 Jul, 2025 Reviewers invited by journal 06 Aug, 2024 Submission checks completed at journal 01 Jul, 2024 Editor assigned by journal 28 Jun, 2024 First submitted to journal 28 Jun, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4654020","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":336853653,"identity":"6aa35987-08c2-48ea-9a1a-d51d0436436c","order_by":0,"name":"Jennifer Mancio","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA20lEQVRIiWNgGAWjYBACxgYgkQBiMTMwPgBSPHykaGE2AGlhI8VGNgkwSUgZc3vvsQcPc+zy+NuZj1V+zbGTYWNgfvjoBj6H9ZxLN0jcllwscZgt7bbstmSgw9iMjXPwaZmRYyaRuI05cQMzj9ltyW3MQC08bNJ4tcx/A9JSD9ZSLLmtnggtM3hAWg6DtTB+3HaYCC09YIcdT5xxmC1ZmnHbcR42ZgJ+MWw/Yyb5c1t1Yn//4YMfgQx7fvbmh4/xamlA4jDzgEk8ykFAHsWVPwioHgWjYBSMgpEJADW5QY8pFz8cAAAAAElFTkSuQmCC","orcid":"","institution":"Faculty of Medicine, University of Porto; Guys and St Thomas NHS Trust Foundation, London","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Jennifer","middleName":"","lastName":"Mancio","suffix":""},{"id":336853654,"identity":"5d2ebd06-8706-42de-a615-fa19abe13699","order_by":1,"name":"Alice Lopes","email":"","orcid":"","institution":"","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Alice","middleName":"","lastName":"Lopes","suffix":""},{"id":336853655,"identity":"4610ccfa-ad02-48e0-9d70-61d945d19541","order_by":2,"name":"Inês Sousa","email":"","orcid":"","institution":"","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Inês","middleName":"","lastName":"Sousa","suffix":""},{"id":336853656,"identity":"68721a56-769e-49c8-8286-699fc29494df","order_by":3,"name":"Fabio Nunes","email":"","orcid":"","institution":"","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Fabio","middleName":"","lastName":"Nunes","suffix":""},{"id":336853657,"identity":"9423cbd8-ab7f-4053-80a0-6fb109837af2","order_by":4,"name":"Sonia Xara","email":"","orcid":"","institution":"","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Sonia","middleName":"","lastName":"Xara","suffix":""},{"id":336853658,"identity":"0e4a5fb7-6bdf-4281-b79a-0a016c56c67f","order_by":5,"name":"Mónica Carvalho","email":"","orcid":"","institution":"Centro Hospitalar de Vila Nova de Gaia e Espinho","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Mónica","middleName":"","lastName":"Carvalho","suffix":""},{"id":336853659,"identity":"a8da9576-b232-47fe-803d-d4daafb80e56","order_by":6,"name":"Wilson Ferreira","email":"","orcid":"","institution":"Centro Hospitalar de Vila Nova de Gaia e Espinho","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Wilson","middleName":"","lastName":"Ferreira","suffix":""},{"id":336853660,"identity":"77ccc286-ba65-4778-82a2-c2e0ad504777","order_by":7,"name":"Nuno Ferreira","email":"","orcid":"","institution":"Centro Hospitalar de Vila Nova de Gaia/Espinho","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Nuno","middleName":"","lastName":"Ferreira","suffix":""},{"id":336853661,"identity":"bbf6fa68-f2b1-4a30-8f66-310485d25e81","order_by":8,"name":"Antonio Barros","email":"","orcid":"https://orcid.org/0000-0002-9103-5852","institution":"","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Antonio","middleName":"","lastName":"Barros","suffix":""},{"id":336853662,"identity":"cd1dd7c2-456f-4c77-883d-e772059d47a6","order_by":9,"name":"Ricardo Fontes-Carvalho","email":"","orcid":"https://orcid.org/0000-0003-2306-8393","institution":"Faculdade de Medicina Universidade do Porto","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Ricardo","middleName":"","lastName":"Fontes-Carvalho","suffix":""},{"id":336853663,"identity":"9263d74b-90ab-456a-b19c-916049eb7b15","order_by":10,"name":"Vasco Gama Ribeiro","email":"","orcid":"","institution":"Centro Hospitalar de Vila Nova de Gaia","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Vasco","middleName":"Gama","lastName":"Ribeiro","suffix":""},{"id":336853664,"identity":"2512862e-0a1f-4002-9a0d-912d792509bd","order_by":11,"name":"Nuno Bettencourt","email":"","orcid":"","institution":"","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Nuno","middleName":"","lastName":"Bettencourt","suffix":""},{"id":336853665,"identity":"cbadd516-01d2-4fc6-91e6-cc8e95285440","order_by":12,"name":"Joao Pedrosa","email":"","orcid":"","institution":"","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Joao","middleName":"","lastName":"Pedrosa","suffix":""}],"badges":[],"createdAt":"2024-06-28 10:21:46","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4654020/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4654020/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":63786676,"identity":"294289e4-8d01-45d7-a1f3-8183c7c10604","added_by":"auto","created_at":"2024-09-02 10:43:22","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":266517,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAbdominal adipose tissue compartments and radiomic feature extraction\u003c/strong\u003e. (A) Total abdominal fat (TAF) can be detected by computed tomography (CT) without the need for contrast in a single slice at L4 and L5-S1 with a minimal estimated radiation exposure; using a threshold method for attenuation values ranging from -150 to -50 Hounsfield units, it is possible to identify fat tissue within the volume of interest (VOI); thus, TAF corresponds to any fat located inside the abdominal slice (B) To segment the visceral abdominal fat (VAF), a mask was created by manual segmentation of the abdominal muscle layer. (C) the Subcutaneous abdominal fat (SAF) compartment was defined by the subtraction of VAF from the TAF masks (“C=A-B”). (D) From the resulting fat masks, high-dimensional data were calculated to describe the area and other complementary 2D- and 3D-size/shape features, the distribution of signal intensity (or gray-level values) inside the VOI (First order features) and the heterogeneity/non-uniformity and coarseness of the fat texture based on the relationship between neighbouring voxel gray-level intensity values by five texture matrices (GLCM, GLDM, GLRLM, GLSZM and NGTDM).\u003c/p\u003e\n\u003cp\u003eGLCM: Gray level co-occurrence matrix; GLDM: Gray level dependence matrix; GLRLM: Gray level run length matrix; GLSZM: Gray level size zone matrix; NGTDM: Neighbouring gray tone difference matrix. SAF: subcutaneous abdominal fat; VAF: visceral abdominal fat.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-4654020/v1/a137426829e46be0e4f1b92a.png"},{"id":63786674,"identity":"16a8bd2f-9ccb-471e-a78f-c06ed4d6d8c1","added_by":"auto","created_at":"2024-09-02 10:43:22","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":333410,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDifferential radiomic profile of subcutaneous and visceral abdominal fat depots on non-contrast computed tomography. \u003c/strong\u003e(A) Orthogonal projection to latent structures-discriminant analysis scatter plots showing separation of SAF and VAF based on its non-contrast CT radiomic features. (B) Loading weights of the most important radiomic features contributing to differentiate the abdominal fat compartments of SAF from VAF. (C) Illustration of the main VAF and SAF phenotypic (“appearance”) differences showing higher gray-level intensity values and heterogeneous, however, finer texture of SAF as compared with VAF on non-contrast CT radiomics.\u003c/p\u003e\n\u003cp\u003eGLCM: Gray level co-occurrence matrix; GLDM: Gray level dependence matrix; GLRLM: Gray level run length matrix; GLSZM: Gray level size zone matrix; LGE: late-gadolinium enhancement; MCC: maximal correlation coefficient; NGTDM: neighbouring gray tone difference matrix; SAF: subcutaneous abdominal fat; VAF: visceral abdominal fat.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-4654020/v1/00834df7cb72cfa7533b9d0d.png"},{"id":63787311,"identity":"5ec6eb49-b174-4579-a600-14ffe2c9a912","added_by":"auto","created_at":"2024-09-02 10:51:22","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":305329,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIncremental classification value of signal dependent fat radiomic features above classical fat area size to detect the presence of coronary calcification on non-contrast computed tomography\u003c/strong\u003e. (A) Receiver operating characteristics (ROC) curves for the multivariate model combining the best SAF signal/texture features added to SAF area to discriminate CCS\u0026gt;0 from CCS=0 patients. (B) ROC curve for the multivariate model combining the best VAF signal/texture features added to VAF area to discriminate CCS\u0026gt;0 from CCS=0 patients. (C) ROC curves for the combined model of SAF and VAF areas and for TAF area alone to discriminate CCS\u0026gt;0 from CCS=0 patients. (D) ROC curves for the combined model of the best SAF and VAF signal/texture features and for the best TAF signal/texture features to discriminate CCS\u0026gt;0 from CCS=0 patients.\u003c/p\u003e\n\u003cp\u003eCCS: coronary calcium score; SAF: subcutaneous abdominal fat; VAF: visceral abdominal fat.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-4654020/v1/2df765eacfb526c9b9327ef9.png"},{"id":63788184,"identity":"37665d03-3860-4872-866e-a70aaa11f924","added_by":"auto","created_at":"2024-09-02 10:59:22","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1562005,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4654020/v1/7492244e-9044-42a4-a228-0b8dfc176195.pdf"},{"id":63787312,"identity":"6be0c315-9ef1-4a16-88a7-9b46cd9fc8a7","added_by":"auto","created_at":"2024-09-02 10:51:22","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":279640,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cbr\u003e\u003c/p\u003e","description":"","filename":"Supplementaryfiles.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4654020/v1/6fd90ead6d67c49bf8c1ffcf.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e conflict of interest to disclose","formattedTitle":"Machine Learning Computed Tomography Radiomics of Abdominal Adipose Tissue to Optimize Cardiovascular Risk Assessment","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eAccumulation of abdominal fat is a well-established adverse metabolic and cardiovascular (CV) risk factor\u003csup\u003e(1\u0026ndash;4)\u003c/sup\u003e. Two metabolic and immunologically distinct compartments of the total abdominal fat (TAF) \u0026ndash; the subcutaneous and visceral abdominal fat (SAF and VAF), have been recognized as independent predictors of cardiometabolic diseases\u003csup\u003e(1, 5\u0026ndash;9)\u003c/sup\u003e. In clinical practice, the assessment of obesity involves non-imaging biomarkers, such as body mass index (BMI) and waist circumference (WC)\u003csup\u003e(10\u0026ndash;12)\u003c/sup\u003e, and imaging biomarkers, such as the fat depot amount measured by magnetic resonance imaging and computed tomography (CT)\u003csup\u003e(5, 13, 14)\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe susceptibility to obesity related metabolic complications is not automatically mediated by overall body fat mass\u003csup\u003e(15)\u003c/sup\u003e and most investigation in the field of ectopic fat has centred on the importance of VAF quantity evaluation. It is known that VAF depots are associated with a proinflammatory and a proatherogenic risk profile, but VAF only contributes to 15% of the total circulation of systemic fatty acids. The majority of fatty acids originate from extra peritoneal adipose tissue, probably from SAF\u003csup\u003e(16)\u003c/sup\u003e. More recently, the radiodensity of SAF and VAF by CT were associated with cardiometabolic risk\u003csup\u003e(17)\u003c/sup\u003e, subclinical atherosclerosis\u003csup\u003e(18)\u003c/sup\u003e, hypertension, insulin resistance and metabolic syndrome independent of fat volume\u003csup\u003e(19)\u003c/sup\u003e. Beyond CT attenuation, radiomic analysis enables in-depth characterization of the signal distribution and spatial relationships between voxels depicting underlying microstructural tissue differences undetectable by its volume and attenuation\u003csup\u003e(20)\u003c/sup\u003e. These radiomic features have been investigated as biomarkers of the tissue biology in histopathology, genomics, transcriptomics and proteomics studies\u003csup\u003e(21, 22)\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIn a previous study of a large cohort of subjects with no prior CV diseases who underwent non-contrast CT, we showed that with aging SAF amount (assessed by its area) decreases while and VAF depot enlarges, and coronary artery calcification worsens suggesting that losing SAF ability to expand and store fat away from the internal organs is a crucial determinant that anticipate the shift to an adverse metabolic and pro-atherosclerotic environment. Detecting the turning point at which SAF becomes dysfunctional before fat starts to accumulate in the VAF compartment is essential to modify the individual patient risk. In the same cohort, we subsequently performed a comprehensive radiomic phenotyping of SAF and VAF by extracting thousands of features to reveal which is the non-contrast CT differential appearance of SAF and VAF, the adverse profile of SAF and VAF that is associated with and add value to fat area to detect coronary calcification.\u003c/p\u003e"},{"header":"SUBJECTS AND METHODS","content":"\u003cp\u003eStudy design, terminology and data analysis followed the recommendations of the Quantitative Imaging Biomarker Alliance (QIBA), and methods are reported as per the aspects of the Radiomics Quality Score (RQS) according to Lambin \u003cem\u003eet al\u003c/em\u003e.\u003csup\u003e(23)\u003c/sup\u003e (\u003cb\u003eSupplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e\u003c/b\u003e).\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Population\u003c/h2\u003e \u003cp\u003eProspective registry of patients who underwent a coronary CT angiography (CCTA) in a tertiary centre from January 2008 to July 2016. In this analysis, only subjects with no prior CV diseases (i.e., known coronary artery disease, prior stroke, prior valvular surgery or atrial fibrillation), and with coronary calcium score (CCS) and abdominal fat images were included. All patients were assessed by a nutritionist for anthropometric assessment on the day of the CT scan. CV risk factors and medication history were recorded based on the information available on the electronic patient record. All participants provided written informed consent, and the study was approved by the Institutional Ethics Committee.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eComputed Tomography and Coronary Calcium Score\u003c/h2\u003e \u003cp\u003e CT scanning was performed in a 64-slice-scanner (Somatom Sensation Cardiac 64, Siemens, Forchheim, Germany) following the institutional protocol for CCTA which at our institution included non-contrast CT images (for CCS and abdominal fat assessment) followed by iodine contrast injection for CCTA. CCS imaging was performed using a prospectively ECG triggered scanning protocol, subsequently, followed by an abdominal single slice acquisition (as detailed elsewhere\u003csup\u003e(24)\u003c/sup\u003e). CCS was reported using the Agatston method.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eAbdominal Fat Radiomic Features\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the abdominal fat regions from where the radiomic analysis was performed following Pyradiomics. Radiomic features were extracted from tissue with mean attenuation values compatible with fat ranging from \u0026minus;\u0026thinsp;150 to -50 HU located in the TAF, SAF and VAF compartments. TAF corresponds to all adipose tissue located in the acquired L4/5-S1 slice, VAF to the mask created by a manual segmentation of the abdominal muscular wall, and SAF compartment corresponds to the region located between the TAF and VAF masks (i.e., obtained by subtracting the VAF from TAF masks). Signal dependent fat features include: the First order (Statistics) and Texture features\u003csup\u003e(25)\u003c/sup\u003e. First order features are histogram-based features and describe fat attenuation (gray-level) values distribution. Textural features describe the spatial relationship between neighbouring voxels gray-level values and measure adipose tissue heterogeneity and coarseness based on five matrices: gray level co-occurrence matrix (GLCM), gray level dependence matrix (GLDM), gray level run-length matrix (GLRLM), gray level size zone matrix (GLSZM) and neighbouring gray tone differences matrix (NGTDM). Size/shape features (including, the classical TAF, SAF and VAF areas) are signal intensity-independent features and describe the 2D- and 3D-geometric properties of the fat mask\u003csup\u003e(26)\u003c/sup\u003e. First order and textural features can be extracted from original images and after further application of image filters (Gradient, Laplacian of Gaussian, Wavelet High-High, Wavelet Low-Low and Wavelet High-Low); shape/size features are extracted from original images only. In total, we extracted 665 first order, textural and shape fat radiomic features \u003cb\u003e(Supplementary Tables S2 and S3).\u003c/b\u003e\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eData Analysis\u003c/h2\u003e \u003cp\u003eWe used STATA software (version 13.1, StataCorp LP, Texas, US) for statistical analysis, and R environment and Scikit Learn of Python (version 3.7) for radiomic analysis; radiomic features were normalized using Z-score prior to analysis\u003csup\u003e(27)\u003c/sup\u003e. Patients\u0026rsquo; characteristics were summarized as number (percentages) or mean (standard deviation) for categorical and continuous variables, respectively (unless specified otherwise). Between-group comparisons were performed using chi-square test for categorical variables and T-test or Mann-Whitney\u0026rsquo;s U test for continuous variables.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eUnique radiomic profile of SAF and VAF\u003c/h2\u003e \u003cp\u003eOrthogonal Projection to Latent Structures-Discriminant Analysis (OPLS-DA) was used to identify the most different CT radiomic features between SAF and VAF\u003csup\u003e(28)\u003c/sup\u003e. The main CT radiomic features were selected based on the regression coefficients (i.e., absolute value of regression coefficient greater than 0.1).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eSAF and VAF radiomic signature of CCS\u0026thinsp;\u0026gt;\u0026thinsp;0\u003c/h2\u003e \u003cp\u003eAn XGBoost ensemble machine learning (ML) model in an internal 10-cross-validation method was employed to identify the SAF, VAF and TAF CT features that discriminate CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 from CCS\u0026thinsp;=\u0026thinsp;0 patients. The most important radiomic features (i.e., higher mean decrease in Gini impurity) were selected to build a SAF, VAF and TAF radiomic signature of CCS\u0026thinsp;\u0026gt;\u0026thinsp;0. Finally, we combined the best performing SAF, VAF and TAF radiomic signature features into a multivariate XGBoost ML model to separate CCS of zero and CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 patients; model performance was assessed using receiver operating characteristics (ROC) curve parameters, and bootstrapping (100 samples) was used to calculate the confidence interval for the area under the curve (AUC). Improvement in classification of fat texture above the respective fat area alone was assessed using an Integrated Discrimination Improvement (IDI) analysis.\u003c/p\u003e \u003c/div\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003eStudy Population\u003c/h2\u003e\n \u003cp\u003eIn this analysis, we included 1001 subjects, 57% male, 57\u0026thinsp;\u0026plusmn;\u0026thinsp;10 years old, 59% had arterial hypertension and 52% had dyslipidaemia, 15% were diabetic. Mean BMI was 28\u0026thinsp;\u0026plusmn;\u0026thinsp;5 kg/m\u003csup\u003e2\u003c/sup\u003e with a mean SAF of 223\u0026thinsp;\u0026plusmn;\u0026thinsp;107cm\u003csup\u003e2\u003c/sup\u003e and mean VAF of 128\u0026thinsp;\u0026plusmn;\u0026thinsp;69cm\u003csup\u003e2\u003c/sup\u003e. Nearly half (47%) had no calcification in their coronary arteries (CCS of zero) and 7% had CCS\u0026thinsp;\u0026gt;\u0026thinsp;300. \u003cstrong\u003eTable\u0026nbsp;1\u003c/strong\u003e summarizes the characteristics of study population.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003eRadiomics Phenotyping of Subcutaneous and Visceral Fat\u003c/h2\u003e\n \u003cp\u003eIdentification of the most relevant CT radiomic features to discriminate SAF from VAF depots revealed that SAF appearance in non-contrast CT images as compared with VAF was characterized by:\u003c/p\u003e\n \u003col style=\"list-style-type: lower-roman;\"\u003e\n \u003cli\u003eGreater proportion of larger size zones with higher gray-level values (Wavelet_LL_GLDM_Large Dependence High Gray Level Emphasis, Wavelet_LL_GLSZM_Large Area High Gray Level Emphasis, Wavelet_LL_GLRLM_Long Run High Gray Level Emphasis, and Wavelet_LL_GLSZM_High Gray Level Zone Emphasis received higher values in SAF compared with VAF).\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eHeterogeneous (non-uniform) texture (Wavelet_HH_GLDM_Gray Level Non-Uniformity _Gray_Level_Non_Uniformity received higher values in SAF than in VAF), and\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eFine texture (Wavelet_LL_GLCM_Autocorrelation received higher values in SAF than in VAF)\u003c/li\u003e\n \u003c/ol\u003e\n \u003cp\u003eOverall, SAF radiomic CT phenotype was characterized in the image by a greater proportion of larger zones with higher gray level values heterogeneous and finer texture compared with VAF \u003cstrong\u003e(\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e \u003cstrong\u003eand Supplementary Table S4).\u003c/strong\u003e\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003eFat Radiomic Signature of Coronary Classification\u003c/h2\u003e\n \u003cp\u003eCompared with CCS of zero, CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 subjects had lower amount of SAF area which was characterized by higher values of fat attenuation (90th percentile signal intensity was higher in CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 indicating (signal distribution shifted to the right) with a finer and uniform texture compared with SAF of CCS\u0026thinsp;=\u0026thinsp;0 subjects. The most important features of VAF that discriminate CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 from CCS\u0026thinsp;=\u0026thinsp;0 were shape features indicating accumulation of more VAF in CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 subjects (larger 2D-area) along larger maximum and minimum abdominal dimensions (Original_Shape_Minor Axis Length and Original_Shape_Minor Axis Length were higher in CCS\u0026thinsp;\u0026gt;\u0026thinsp;0) in an spherical-like shape abdomen (Original_Shape_Elongation and Original_Shape_Area Volume Surface Ratio were lower in CCS\u0026thinsp;\u0026gt;\u0026thinsp;0); opposite to SAF, VAF of CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 patients displayed lower signal intensity with a higher proportion of lower gray level values areas and a non-uniform and heterogenous texture \u003cstrong\u003e(Table\u0026nbsp;2)\u003c/strong\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003eIncremental Value of Fat Texture above Fat Area to Detect Coronary Calcification\u003c/h2\u003e\n \u003cp\u003eA ML model including SAF area alone detected the presence of coronary calcification correctly in 58% of the patients with an AUC of 0.62 (95% CI: 0.58\u0026ndash;0.66); by adding the SAF texture signature features to SAF area, there was a significant improvement in the classification of patients providing an AUC of 0.70 (95% CI: 0.66\u0026ndash;0.74) and an accuracy score of 65% with an IDI of 0.02 (95% CI: 0.01\u0026ndash;0.03; p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001) \u003cstrong\u003e(\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e-A\u003cstrong\u003e)\u003c/strong\u003e. VAF texture added to VAF area did not improve the detection of CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 (IDI: 0.001; 95% CI: 0.002\u0026ndash;0.003; p\u0026thinsp;=\u0026thinsp;0.128) \u003cstrong\u003e(\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e-B\u003cstrong\u003e)\u003c/strong\u003e. A model including the best SAF and VAF features performed better than a combined model of SAF and VAF indexed area (p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001). There was no significant difference in the performance of the models including the best SAF texture features and VAF texture features (AUC with SAF texture 0.70 (95% CI: 0.68\u0026ndash;0.74) vs AUC with VAF texture 0.69 (95% CI: 0.65\u0026ndash;0.72), p\u0026thinsp;=\u0026thinsp;0.08), although there was a trend for a better performance with SAF texture features.\u003c/p\u003e\n \u003cp\u003eA ML model including both the SAF and VAF indexed areas yielded an AUC of 0.68 (95% CI: 0.65\u0026ndash;0.71) which is significantly higher than the performance of indexed TAF area alone (i.e., TAF area with no discrimination of fat compartments) (AUC 0.53; 95% CI: 0.47\u0026ndash;0.59) \u003cstrong\u003e(\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e-C\u003cstrong\u003e)\u003c/strong\u003e. However, a model including the best performing SAF and VAF texture features that discriminate patients with coronary calcification provided a similar accuracy than a model including the best texture features extracted from any abdominal fat; AUC with best SAF and best VAF features was 0.71 (95% CI: 0.66\u0026ndash;0.76) vs AUC with best TAF features was 0.69 (95% CI: 0.66\u0026ndash;0.73) \u003cstrong\u003e(\u003c/strong\u003eFig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e-D\u003cstrong\u003e)\u003c/strong\u003e.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eIn this study, we combined radiomics methodology with ML to characterize the abdominal fat compartments and stratify CV risk as assessed by CCS. We were able to detect abdominal fat gray-level and texture differences between SAF and VAF depots supporting the concept of their intrinsically distinct biological properties, and between SAF and VAF of CCS \u0026gt; 0 and CCS = 0 subjects. We demonstrated incremental value of abdominal fat texture above its area to detect coronary calcification, and, differently than TAF area, automatic radiomic feature extraction from any abdominal fat tissue may eliminate the need for SAF and VAF individual segmentation thereby simplifying the radiomics workflow for individualized risk assessment.\u003c/p\u003e \u003cp\u003eGlobally, obesity and abdominal adiposity have been associated with a wide range of metabolic complications, particularly CV disease\u003csup\u003e(29)\u003c/sup\u003e. Although the accumulation of VAF is one of the main contributors to quantifying cardiometabolic risk above BMI, the heterogeneity of body composition makes it difficult to assess fat distribution in clinical practice. Radiomics, first described in 2012\u003csup\u003e(30)\u003c/sup\u003e, enables high-throughput extraction of quantitative features from medical images and describes tissue heterogeneity captured in visually unrecognizable voxel grey-level intensities\u003csup\u003e(30–32)\u003c/sup\u003e. Initially applied to the oncology field, radiomics allow tumour phenotyping and identification of imaging features related to poor prognosis. In the CV field, studies in magnetic resonance imaging showed accurate distinction between subacute and chronic myocardial infarction, detecting of myocardial fibrosis without gadolinium and cardiomyopathy differential diagnosis\u003csup\u003e(33, 34)\u003c/sup\u003e. CT radiomics detected high-risk plaques better than the classical features and coronary inflammation by 3D phenotyping of perivascular fat attenuation.\u003c/p\u003e \u003cp\u003eThe CT attenuation of adipose tissue may indicate some tissue characteristics, including\u003csup\u003e(18)\u003c/sup\u003e: (\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e) more negative HU values (i.e.; low gray-level values) are associated with more lipid-dense fat tissue and poorly vascularized adipose tissue resulting from the radiographic properties of the blood, while (\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e) less negative HU values (i.e.; high gray-level values) can result from tissue oedema/inflammation (i.e. increased water content), lower lipid content and fibrotic adipose tissue due to excessive collagen deposition, and (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e) heterogenous texture can be explained by cellular composition, microvasculature, metabolic characteristics and extracellular matrix composition\u003csup\u003e(35)\u003c/sup\u003e, which wavelet decompositions can capture discontinuities.\u003c/p\u003e \u003cp\u003eIn this study we found greater proportion of larger zones with higher gray-level values (i.e. less negative HU) and heterogeneous texture characterizing the SAF depot, which suggests increased inflammation, fibrosis, changes in vascularization and oxidative stress\u003csup\u003e(36–38)\u003c/sup\u003e. Anatomically, the SAF compartment is subdivided into superficial SAF and deep SAF separated by a fascial plane (fascia superficialis). The two sub compartments present different risk profiles whose understanding depends on the precise quantification of each of them. Thus, our observation might represent the deep SAF, which contains larger, less organized and more vascularized lobules\u003csup\u003e(39)\u003c/sup\u003e, related to the onset of adipose tissue dysfunction, impaired glucose metabolism, hyperinsulinemia and insulin resistance\u003csup\u003e(40)\u003c/sup\u003e. The correct distinction between deep SAF and superficial SAF becomes a challenge because, the difference in volume between the two compartments decreases as fat accumulation increases (making it harder to differentiate in individuals with a high amount of fat); moreover, SAF accumulation profile changes along the lumbar levels. In some regions (L1-L2), the superficial SAF and deep SAF amounts are equal, with prominent superficial fascia visible through MRI; however, the caudal progression to L5 makes it difficult to distinguish, with multiple fascial lines\u003csup\u003e(41)\u003c/sup\u003e. CT-based texture captured regional differences between abdominal fat compartments. Juan Shi \u003cem\u003eet al\u003c/em\u003e\u003csup\u003e(42)\u003c/sup\u003e showed that, in a Chinese population, heterogeneous texture features extracted from VAF were significantly associated with metabolic syndrome and related disorders. In adipose tissue, the extracellular matrix plays an important role in tissue expansion and angiogenesis. Adipose tissue expansion depends on extracellular matrix remodeling through cycles of collagen deposition. When adipose tissue expansion becomes dysfunctional, excessive and unregulated accumulation of collagen and other extracellular matrix components results in fibrosis, which limits the expansion capacity of adipocytes\u003csup\u003e(43)\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eAnthropometric indices of obesity are easily implemented, but newer imaging-based methods offer greater sensitivity and specificity for measuring specific deposits. Overall, in subjects with coronary calcification, our imaging technique captured pathogenic features in the two compartments of abdominal fat even before the average BMI value reached the obesity threshold. The area of adipose tissue quantified from a slice image has strong correlations with the total volume of abdominal adipose tissue\u003csup\u003e(44)\u003c/sup\u003e and is associated with underlying metabolic disorders related to obesity. By contrast, TAF area performs very poorly and is unable to detect coronary calcification. This demands individual segmentation of VAF and SAF subtraction from TAF. In this work, we demonstrated additional classification value of fat texture above fat area to detecting coronary calcification. Moreover, a ML model of the best SAF combined with best VAF radiomic features that discriminate CCS \u0026gt; 0 from CCS = 0 subjects performed similarly as the model including the best TAF radiomic features (i.e. features extracted from any fat tissue located in the obtained slice). Because TAF can be detected in CT without the need for contrast in a single slice at L4 and L5-S1 with a minimal estimated radiation exposure of 0.06 mSv, automatic feature extraction from TAF using a threshold method only, and without the need for segmentation of SAF and VAF, can serve to find new imaging biomarkers of obesity and build improved and easy to implement individual risk prediction models.\u003c/p\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eLimitations\u003c/h2\u003e \u003cp\u003eBiological interpretability of radiomic features remains challenging; in this study, we did not correlate imaging findings with fat samples or analytical parameters. Our study design is cross-sectional with CT performed in a single centre; ML models were trained and tested in the entire sample size to optimize data learning and minimize overfitting; external validation was not tested. The segmentation of the compartments was performed manually, which is human dependent and can be a source of variability affecting radiomics reproducibility. The radiomic features were extracted from two-dimensional images and the abdominal distribution of adiposity showed interindividual local variability, which we were not able to investigate. Epicardial adipose tissue which represents an important cardiometabolic risk factor was not analysed.\u003c/p\u003e \u003c/div\u003e "},{"header":"Main conclusions","content":"\u003cp\u003eIn non-contrast CT, SAF and VAF appearance differs and abdominal fat tissue radiomic profile is associated with CV risk. Radiomic analysis of TAF can derive new imaging biomarkers of obesity, which can improve and facilitate individual risk assessment.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eArea under the curve\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eBMI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eBody mass index\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eCCS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eCoronary calcium score\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eCCTA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eCoronary computed tomography angiography\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eCT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eComputed tomography\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eCV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eCardiovascular\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eML\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eMachine learning\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eROC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eReceiving operating characteristics\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eSAF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eSubcutaneous abdominal fat\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eVAF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eVisceral abdominal fat\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eTAF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eTotal abdominal fat\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"18.80199667221298%\" valign=\"top\"\u003e\n \u003cp\u003eWC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"81.19800332778702%\" valign=\"top\"\u003e\n \u003cp\u003eWaist circumference\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Declarations","content":"\u003ch2\u003eCompeting Interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare no competing financial interests\u003c/p\u003e\n\u003ch2\u003eConflict of interest\u003c/h2\u003e\n\u003cp\u003eThe authors declare no conflicts of interest.\u003c/p\u003e\n\u003ch2\u003eAuthor Contributions\u003c/h2\u003e\n\u003cp\u003eAL contributed with data collection and database cleaning, results analysis, and manuscript writing. IS segmented the abdominal to define the region of interest. FN collected CT imaging raw data. SX performed the anthropometric assessment. MC and WF performed the computed tomography and contributed with patient recruitment. NF, VGR, NB and RFC contributed with manuscript revision. AB contributed with statistical data analysis and results interpretation. JP performed the radiomic feature extraction and revised the manuscript. JM designed the study, contributed with data analysis, results interpretation, wrote and revised the manuscript.\u003c/p\u003e\n\u003ch2\u003eDATA AVAILABILITY\u003c/h2\u003e\n\u003cp\u003eThe data that support the findings of this study are available from [third party name] but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Faculty of Medicine of the University of Porto.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDespr\u0026eacute;s J-P, Lemieux I, Bergeron J, Pibarot P, Mathieu P, Larose E, et al. Abdominal obesity and the metabolic syndrome: contribution to global cardiometabolic risk. Arteriosclerosis, thrombosis, and vascular biology. 2008;28(6):1039-49.\u003c/li\u003e\n\u003cli\u003eKwon H, Kim D, Kim JS. Body fat distribution and the risk of incident metabolic syndrome: a longitudinal cohort study. Scientific reports. 2017;7(1):10955.\u003c/li\u003e\n\u003cli\u003eHu HH, Chen J, Shen W. Segmentation and quantification of adipose tissue by magnetic resonance imaging. Magnetic Resonance Materials in Physics, Biology and Medicine. 2016;29:259-76.\u003c/li\u003e\n\u003cli\u003eKumari R, Kumar S, Kant R. An update on metabolic syndrome: Metabolic risk markers and adipokines in the development of metabolic syndrome. Diabetes \u0026amp; Metabolic Syndrome: Clinical Research \u0026amp; Reviews. 2019;13(4):2409-17.\u003c/li\u003e\n\u003cli\u003eIbrahim MM. Subcutaneous and visceral adipose tissue: structural and functional differences. Obesity reviews. 2010;11(1):11-8.\u003c/li\u003e\n\u003cli\u003eSingh S, Dulai PS, Zarrinpar A, Ramamoorthy S, Sandborn WJ. Obesity in IBD: epidemiology, pathogenesis, disease course and treatment outcomes. Nature reviews Gastroenterology \u0026amp; hepatology. 2017;14(2):110-21.\u003c/li\u003e\n\u003cli\u003ePickhardt PJ, Graffy PM, Perez AA, Lubner MG, Elton DC, Summers RM. Opportunistic screening at abdominal CT: use of automated body composition biomarkers for added cardiometabolic value. RadioGraphics. 2021;41(2):524-42.\u003c/li\u003e\n\u003cli\u003eRitchie S, Connell J. The link between abdominal obesity, metabolic syndrome and cardiovascular disease. Nutrition, Metabolism and cardiovascular diseases. 2007;17(4):319-26.\u003c/li\u003e\n\u003cli\u003eElffers TW, de Mutsert R, Lamb HJ, de Roos A, Willems van Dijk K, Rosendaal FR, et al. Body fat distribution, in particular visceral fat, is associated with cardiometabolic risk factors in obese women. PloS one. 2017;12(9):e0185403.\u003c/li\u003e\n\u003cli\u003eGraffy PM, Pickhardt PJ. Quantification of hepatic and visceral fat by CT and MR imaging: relevance to the obesity epidemic, metabolic syndrome and NAFLD. The British journal of radiology. 2016;89(1062):20151024.\u003c/li\u003e\n\u003cli\u003eUpdate AS. Heart disease and stroke statistics\u0026mdash;2020 update: a report from the American Heart Association. Circulation. 2020;141(9):e139-e596.\u003c/li\u003e\n\u003cli\u003ePerreault L. Obesity in adults: prevalence, screening, and evaluation. Waltham, MA, USA: UpToDate Inc. 2018.\u003c/li\u003e\n\u003cli\u003eMei Z, Grummer-Strawn LM, Pietrobelli A, Goulding A, Goran MI, Dietz WH. Validity of body mass index compared with other body-composition screening indexes for the assessment of body fatness in children and adolescents. The American journal of clinical nutrition. 2002;75(6):978-85.\u003c/li\u003e\n\u003cli\u003eOrtega FB, Sui X, Lavie CJ, Blair SN, editors. Body mass index, the most widely used but also widely criticized index: would a criterion standard measure of total body fat be a better predictor of cardiovascular disease mortality? Mayo Clinic Proceedings; 2016: Elsevier.\u003c/li\u003e\n\u003cli\u003eZhang X, Ha S, Lau HC-H, Yu J, editors. Excess body weight: Novel insights into its roles in obesity comorbidities. Seminars in Cancer Biology; 2023: Elsevier.\u003c/li\u003e\n\u003cli\u003ePich\u0026eacute; M-E, Poirier P, Lemieux I, Despr\u0026eacute;s J-P. Overview of epidemiology and contribution of obesity and body fat distribution to cardiovascular disease: an update. Progress in cardiovascular diseases. 2018;61(2):103-13.\u003c/li\u003e\n\u003cli\u003eShah RV, Allison M, Lima J, Abbasi S, Eisman A, Lai C, et al. Abdominal fat radiodensity, quantity and cardiometabolic risk: The Multi-Ethnic Study of Atherosclerosis. Nutrition, Metabolism and Cardiovascular Diseases. 2016;26(2):114-22.\u003c/li\u003e\n\u003cli\u003eAlvey NJ, Pedley A, Rosenquist KJ, Massaro JM, O\u0026apos;Donnell CJ, Hoffmann U, et al. Association of fat density with subclinical atherosclerosis. Journal of the American Heart Association. 2014;3(4):e000788.\u003c/li\u003e\n\u003cli\u003eRosenquist KJ, Pedley A, Massaro JM, Therkelsen KE, Murabito JM, Hoffmann U, et al. Visceral and subcutaneous fat quality and cardiometabolic risk. JACC: Cardiovascular Imaging. 2013;6(7):762-71.\u003c/li\u003e\n\u003cli\u003eWang Z, Zhang J, Zhang A, Sun Y, Su M, You H, et al. The role of epicardial and pericoronary adipose tissue radiomics in identifying patients with non-ST-segment elevation myocardial infarction from unstable angina. Heliyon. 2023;9(5).\u003c/li\u003e\n\u003cli\u003eLambin P, Leijenaar RT, Deist TM, Peerlings J, De Jong EE, Van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nature reviews Clinical oncology. 2017;14(12):749-62.\u003c/li\u003e\n\u003cli\u003eGillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology. 2016;278(2):563-77.\u003c/li\u003e\n\u003cli\u003eLambin P, Leijenaar RTH, Deist TM, Peerlings J, de Jong EEC, van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nature reviews Clinical oncology. 2017;14(12):749-62.\u003c/li\u003e\n\u003cli\u003eMancio J, Fonseca P, Figueiredo B, Ferreira W, Carvalho M, Ferreira N, et al. Association of body mass index and visceral fat with aortic valve calcification and mortality after transcatheter aortic valve replacement: the obesity paradox in severe aortic stenosis. Diabetology \u0026amp; Metabolic Syndrome. 2017;9:1-10.\u003c/li\u003e\n\u003cli\u003evan Griethuysen JJM, Fedorov A, Parmar C, Hosny A, Aucoin N, Narayan V, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer research. 2017;77(21):e104-e7.\u003c/li\u003e\n\u003cli\u003eRaisi-Estabragh Z, Izquierdo C, Campello VM, Martin-Isla C, Jaggi A, Harvey NC, et al. Cardiac magnetic resonance radiomics: basic principles and clinical perspectives. Eur Heart J Cardiovasc Imaging. 2020;21(4):349-56.\u003c/li\u003e\n\u003cli\u003eHaga A, Takahashi W, Aoki S, Nawa K, Yamashita H, Abe O, et al. Standardization of imaging features for radiomics analysis. The journal of medical investigation : JMI. 2019;66(1.2):35-7.\u003c/li\u003e\n\u003cli\u003eJ. Trygg SW. Orthogonal projections to latent structures (O-PLS). Journal of Chemometrics. 2002;16:119\u0026ndash;28.\u003c/li\u003e\n\u003cli\u003eHaffner SM. Relationship of metabolic risk factors and development of cardiovascular disease and diabetes. Obesity. 2006;14(S6):121S-7S.\u003c/li\u003e\n\u003cli\u003eLambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, Van Stiphout RG, Granton P, et al. Radiomics: extracting more information from medical images using advanced feature analysis. European journal of cancer. 2012;48(4):441-6.\u003c/li\u003e\n\u003cli\u003eAerts HJ, Velazquez ER, Leijenaar RT, Parmar C, Grossmann P, Carvalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nature communications. 2014;5(1):4006.\u003c/li\u003e\n\u003cli\u003eKumar V, Gu Y, Basu S, Berglund A, Eschrich SA, Schabath MB, et al. Radiomics: the process and the challenges. Magnetic resonance imaging. 2012;30(9):1234-48.\u003c/li\u003e\n\u003cli\u003eBaessler B, Mannil M, Oebel S, Maintz D, Alkadhi H, Manka R. Subacute and Chronic Left Ventricular Myocardial Scar: Accuracy of Texture Analysis on Nonenhanced Cine MR Images. Radiology. 2018;286(1):103-12.\u003c/li\u003e\n\u003cli\u003eBaessler B, Mannil M, Maintz D, Alkadhi H, Manka R. Texture analysis and machine learning of non-contrast T1-weighted MR images in patients with hypertrophic cardiomyopathy-Preliminary results. European journal of radiology. 2018;102:61-7.\u003c/li\u003e\n\u003cli\u003eLee M-J, Wu Y, Fried SK. Adipose tissue heterogeneity: implication of depot differences in adipose tissue for obesity complications. Molecular aspects of medicine. 2013;34(1):1-11.\u003c/li\u003e\n\u003cli\u003eKim S-H, Chung J-h, Song S-W, Jung WS, Lee Y-A, Kim H-N. Relationship between deep subcutaneous abdominal adipose tissue and metabolic syndrome: a case control study. Diabetology \u0026amp; metabolic syndrome. 2016;8:1-9.\u003c/li\u003e\n\u003cli\u003eOikonomou EK, Antoniades C. The role of adipose tissue in cardiovascular health and disease. Nature Reviews Cardiology. 2019;16(2):83-99.\u003c/li\u003e\n\u003cli\u003eCrewe C, An YA, Scherer PE. The ominous triad of adipose tissue dysfunction: inflammation, fibrosis, and impaired angiogenesis. The Journal of clinical investigation. 2017;127(1):74-82.\u003c/li\u003e\n\u003cli\u003eMarinou K, Hodson L, Vasan SK, Fielding BA, Banerjee R, Brismar K, et al. Structural and functional properties of deep abdominal subcutaneous adipose tissue explain its association with insulin resistance and cardiovascular risk in men. Diabetes care. 2014;37(3):821-9.\u003c/li\u003e\n\u003cli\u003eHoffstedt J, Arner E, Wahrenberg H, Andersson D, Qvisth V, L\u0026ouml;fgren P, et al. Regional impact of adipose tissue morphology on the metabolic profile in morbid obesity. Diabetologia. 2010;53:2496-503.\u003c/li\u003e\n\u003cli\u003eBhanu PK, Arvind CS, Yeow LY, Chen WX, Lim WS, Tan CH. CAFT: a deep learning-based comprehensive abdominal fat analysis tool for large cohort studies. Magnetic Resonance Materials in Physics, Biology and Medicine. 2022:1-16.\u003c/li\u003e\n\u003cli\u003eShi J, Bao G, Hong J, Wang S, Chen Y, Zhao S, et al. Deciphering CT texture features of human visceral fat to evaluate metabolic disorders and surgery-induced weight loss effects. EBioMedicine. 2021;69.\u003c/li\u003e\n\u003cli\u003eMorais JBS, da Silva Dias TM, Cardoso BEP, de Paiva Sousa M, de Sousa TGV, de Ara\u0026uacute;jo DSC, et al. Adipose tissue dysfunction: impact on metabolic changes? Hormone and Metabolic Research. 2022;54(12):785-94.\u003c/li\u003e\n\u003cli\u003eHsu L-Y, Ali Z, Bagheri H, Huda F, Redd BA, Jones EC. Comparison of CT and Dixon MR Abdominal Adipose Tissue Quantification Using a Unified Computer-Assisted Software Framework. Tomography. 2023;9(3):1041-51.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-obesity","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"ijo","sideBox":"Learn more about [International Journal of Obesity](http://www.nature.com/ijo/)","snPcode":"41366","submissionUrl":"https://mts-ijo.nature.com/cgi-bin/main.plex","title":"International Journal of Obesity","twitterHandle":"@intjobesity","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Subcutaneous abdominal fat, visceral abdominal fat, abdominal obesity, radiomics, radiomic analysis, computed tomography, cardiovascular risk, coronary calcification","lastPublishedDoi":"10.21203/rs.3.rs-4654020/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4654020/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eSubcutaneous (SAF) and visceral (VAF) abdominal fat have specific properties which the global body fat and total abdominal fat (TAF) size metrics do not capture. Beyond size, radiomics allows deep tissue phenotyping and may capture fat dysfunction. We aimed to characterize the computed tomography (CT) radiomics of SAF and VAF and assess their incremental value above fat size to detect coronary calcification.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eSAF, VAF and TAF area, signal distribution and texture were extracted from non-contrast CT of 1001 subjects (57% male, 57\u0026thinsp;\u0026plusmn;\u0026thinsp;10 years) with no established cardiovascular disease who underwent CT for coronary calcium score (CCS) with additional abdominal slice (L4/5-S1). XGBoost machine learning models (ML) were used to identify the best features that discriminate SAF from VAF and to train/test ML to detect any coronary calcification (CCS\u0026thinsp;\u0026gt;\u0026thinsp;0).\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eSAF and VAF appearance in non-contrast CT differs: SAF displays brighter and finer texture than VAF. Compared with CCS\u0026thinsp;=\u0026thinsp;0, SAF of CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 has higher signal and homogeneous texture, while VAF of CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 has lower signal and heterogeneous texture. SAF signal/texture improved SAF area performance to detect CCS\u0026thinsp;\u0026gt;\u0026thinsp;0. A ML including SAF and VAF area performed better than TAF area to discriminate CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 from CCS\u0026thinsp;=\u0026thinsp;0, however, a combined ML of the best SAF and VAF features detected CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 as the best TAF features.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eIn non-contrast CT, SAF and VAF appearance differs and SAF radiomics improves the detection of CCS\u0026thinsp;\u0026gt;\u0026thinsp;0 when added to fat area; TAF radiomics (but not TAF area) spares the need for separate SAF and VAF segmentations.\u003c/p\u003e","manuscriptTitle":"Machine Learning Computed Tomography Radiomics of Abdominal Adipose Tissue to Optimize Cardiovascular Risk Assessment","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-09-02 10:43:17","doi":"10.21203/rs.3.rs-4654020/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"revise","date":"2025-11-12T10:52:04+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"This content is not available.","date":"2025-10-26T10:22:07+00:00","index":2,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2025-10-15T03:26:10+00:00","index":2,"fulltext":"This content is not available."},{"type":"editorInvitedReview","content":"This content is not available.","date":"2025-07-14T18:47:05+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2025-07-01T15:04:04+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewersInvited","content":"","date":"2024-08-06T18:50:57+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-07-01T13:12:55+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-06-28T10:19:43+00:00","index":"","fulltext":""},{"type":"submitted","content":"International Journal of Obesity","date":"2024-06-28T10:19:43+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"international-journal-of-obesity","isNatureJournal":false,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"ijo","sideBox":"Learn more about [International Journal of Obesity](http://www.nature.com/ijo/)","snPcode":"41366","submissionUrl":"https://mts-ijo.nature.com/cgi-bin/main.plex","title":"International Journal of Obesity","twitterHandle":"@intjobesity","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"5ff9fe81-cdb4-487d-806a-1f57a0fa7716","owner":[],"postedDate":"September 2nd, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":35681203,"name":"Health sciences/Diseases/Cardiovascular diseases"},{"id":35681204,"name":"Health sciences/Risk factors"}],"tags":[],"updatedAt":"2026-07-04T10:51:03+00:00","versionOfRecord":[],"versionCreatedAt":"2024-09-02 10:43:17","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4654020","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4654020","identity":"rs-4654020","version":["v1"]},"buildId":"7rjqhiLT3MXkJMwkYKINL","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}