Warning: fopen(/home/virtual/enm-kes/journal/upload/ip_log/ip_log_2025-07.txt): failed to open stream: Permission denied in /home/virtual/lib/view_data.php on line 100 Warning: fwrite() expects parameter 1 to be resource, boolean given in /home/virtual/lib/view_data.php on line 101 Deep Learning-Based Adrenal Gland Volumetry for the Prediction of Diabetes
Skip Navigation
Skip to contents

Endocrinol Metab : Endocrinology and Metabolism

clarivate
OPEN ACCESS
SEARCH
Search

Articles

Page Path
HOME > Endocrinol Metab > Ahead-of print > Article
Original Article
Deep Learning-Based Adrenal Gland Volumetry for the Prediction of Diabetes
Eu Jeong Ku1,2*orcid, Soon Ho Yoon3*orcid, Seung Shin Park2,4orcid, Ji Won Yoon1,2orcid, Jung Hee Kim2,4orcid

DOI: https://doi.org/10.3803/EnM.2025.2336
Published online: June 18, 2025

1Department of Internal Medicine, Seoul National University Hospital Healthcare System Gangnam Center, Seoul, Korea

2Department of Internal Medicine, Seoul National University College of Medicine, Seoul, Korea

3Department of Radiology, Seoul National University Hospital, Seoul, Korea

4Department of Internal Medicine, Seoul National University Hospital, Seoul, Korea

Corresponding authors: Jung Hee Kim Department of Internal Medicine, Seoul National University Hospital, Seoul National University College of Medicine, 101 Daehak-ro, Jongno-gu, Seoul 03080, Korea Tel: +82-2072-4839, Fax: +82-762-9662, E-mail: jhee1@snu.ac.kr
Ji Won Yoon Department of Internal Medicine, Seoul National University Hospital Healthcare System Gangnam Center, Seoul National University College of Medicine, 152 Teheran-ro, Gangnam-gu, Seoul 06236, Korea Tel: +82-2-2112-5500, E-mail: jwyoonmd@gmail.com
*These authors contributed equally to this work.
• Received: February 6, 2025   • Revised: March 25, 2025   • Accepted: April 7, 2025

Copyright © 2025 Korean Endocrine Society

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

  • 264 Views
  • 20 Download
  • Background
    The long-term association between adrenal gland volume (AGV) and type 2 diabetes (T2D) remains unclear. We aimed to determine the association between deep learning-based AGV and current glycemic status and incident T2D.
  • Methods
    In this observational study, adults who underwent abdominopelvic computed tomography (CT) for health checkups (2011–2012), but had no adrenal nodules, were included. AGV was measured from CT images using a three-dimensional nnU-Net deep learning algorithm. We assessed the association between AGV and T2D using a cross-sectional and longitudinal design.
  • Results
    We used 500 CT scans (median age, 52.3 years; 253 men) for model development and a Multi-Atlas Labeling Beyond the Cranial Vault dataset for external testing. A clinical cohort included a total of 9708 adults (median age, 52.0 years; 5,769 men). The deep learning model demonstrated a dice coefficient of 0.71±0.11 for adrenal segmentation and a mean volume difference of 0.6±0.9 mL in the external dataset. Participants with T2D at baseline had a larger AGV than those without (7.3 cm3 vs. 6.7 cm3 and 6.3 cm3 vs. 5.5 cm3 for men and women, respectively, all P<0.05). The optimal AGV cutoff values for predicting T2D were 7.2 cm3 in men and 5.5 cm3 in women. Over a median 7.0-year follow-up, T2D developed in 938 participants. Cumulative T2D risk was accentuated with high AGV compared with low AGV (adjusted hazard ratio, 1.27; 95% confidence interval, 1.11 to 1.46).
  • Conclusion
    AGV, measured using deep learning algorithms, is associated with current glycemic status and can significantly predict the development of T2D.
Type 2 diabetes (T2D), which is a chronic disease characterized by insulin resistance and relative insulin deficiency, is associated with various endocrine and metabolic abnormalities [1]. Considering the rapid increase in the incidence of T2D domestically and globally, which poses a significant public health issue, exploring new predictive markers of this disease and understanding its underlying mechanisms are of great importance [2,3].
In patients with T2D, hyperinsulinemia induced by insulin resistance and dysregulation of the hypothalamus-pituitary-adrenal (HPA) axis is simultaneously observed [4-7]. Hyperinsulinemia, projected by insulin resistance, has been suggested to promote overactivity of the HPA axis [8-12]. Hyperglycemia can lead to an increase in proinflammatory cytokine levels, which can dysregulate the HPA axis [13,14]. The adrenal glands are crucial endocrine organs that produce several hormones regulating multiple physiological processes, including metabolism, stress, and immune responses [9]. Hypercortisolism is characterized by central obesity and induces metabolic syndrome, which is accompanied by worsened insulin resistance, elevated blood pressure, and dyslipidemia [15].
Morphometric changes in the adrenal glands, such as changes in size, shape, or cellular structure, can reflect alterations in their functional capacity to produce hormones [16]. However, quantification of adrenal gland volume (AGV) has been challenging owing to the glands’ small size and irregular shape. Advancements in imaging technology and deep learning algorithms have enabled automatic segmentation of the adrenal glands [17-19]. These developments have paved the way for investigating the potential use of AGV as a biomarker of metabolic disorders.
To the best of our knowledge, there has been no solid evidence on the association between adrenal volume and glycemic status. In particular, few long-term follow-up studies investigating whether adrenal volume in a large cohort of healthy participants can predict future diabetes have been published. Therefore, this study aimed to compare the differences in AGV, measured using automated artificial intelligence (AI)-based segmentation technology, among a cohort of healthy participants, after adjusting for various confounding variables according to glycemic status. Furthermore, we aimed to determine whether AGV can be a potential marker for future T2D.
Study participants
This cross-sectional study enrolled healthy participants who visited the Seoul National University Hospital Healthcare System Gangnam Center between January 2011 and September 2012. A subset of these participants was subsequently observed longitudinally, with a median follow-up duration of 7.0 years (interquartile range, 4.0 to 8.0) (Supplemental Fig. S1) [20]. The inclusion criterion was as follows: undergoing abdominopelvic computed tomography (CT) during examination. The exclusion criteria were as follows: (1) adrenal incidentaloma (n=453); (2) lesions either known to have or with high suspicion of malignancy (n=5); (3) clinical manifestations of adrenal disease, such as pheochromocytoma, Cushing’s syndrome, or primary aldosteronism (n=10); and (4) prior history of adrenalectomy due to any cause (n=1).
This retrospective study was approved by the Seoul National University Hospital Institutional Review Board (SNUH IRB No. 2107-191-1237) and was conducted in compliance with the Declaration of Helsinki guidelines. The requirement for written informed consent was waived due to the retrospective nature of the study.
Clinical and laboratory assessments
Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured after the participants rested for at least 15 minutes. Height, body weight, and waist circumference (WC) were measured while the participants wore light examination gowns, determined using standardized devices and expressed to one decimal place in centimeters (cm) and kilograms (kg), respectively. Body mass index (BMI) was calculated by dividing weight by height squared (kg/m2). Blood samples were collected after at least 12 hours of fasting, and glycated hemoglobin (HbA1c), fasting glucose, and lipid profiles, including total cholesterol (TC), triglycerides (TGs), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C), were collected. Based on the submitted questionnaires in the medical records, we reviewed pre-existing conditions, such as T2D, hypertension, and dyslipidemia, and cross-referenced these with the medication records. The classification of T2D, prediabetes, and normal glucose tolerance (NGT) group was determined according to the American Diabetes Association and Korean Diabetes Association guidelines [1,21]. Hypertension was defined as cases in which the participant indicated hypertension as an underlying condition on the questionnaire or was taking antihypertensive agents, or when the SBP or DBP exceeded 140 or 90 mm Hg, respectively. Dyslipidemia was defined as prior diagnosis, the administration of lipid-lowering agents, or the presence of any lipid profile abnormalities, including TC ≥240 mg/dL, HDL-C <40 mg/dL for men and <50 mg/dL for women, TG >200 mg/dL, or LDL-C ≥160 mg/dL. The triglyceride-glucose (TyG) index was calculated using the following formula: Ln [fasting TG (mg/dL)×fasting plasma glucose (mg/dL)/2].
Body composition and adrenal gland volume measurement
All CT images (CT protocols in Supplemental Table S1) were obtained during the portal venous phase of contrast-enhanced abdominopelvic scans [20]. These CT images were processed using a commercially available body composition analysis software (DeepCatch version 1.0.0.0, MEDICALIP Co. Ltd., Seoul, Korea). This software incorporates a previously developed three-dimensional (3D) U-Net, which yielded segmentation accuracies (dice similarity scores) of 0.92 to 0.99 for muscle, abdominal visceral fat, and subcutaneous fat and 0.94 to 0.98 for internal organ masks [22]. Using these segmentation masks, the software calculated the skeletal muscle area (SMA), visceral fat area (VFA), and subcutaneous fat area (SFA) at the level of the third lumbar vertebral body.
The architecture of the deep learning model for adrenal gland segmentation is shown in Supplemental Fig. S2 [20]. To train the adrenal segmentation model, reference masks for the adrenal glands were generated using a dataset of 500 CT scans. The dataset included subjects without adrenal nodules and was collected from two tertiary referral hospitals, Seoul National University Hospital and Seoul National University Bundang Hospital, between 2000 and 2020. The subjects had an average age of 52.3 years, and 253 of them were men. A technician initially performed semiautomatic segmentation of the adrenal glands from CT images, and the masks were subsequently finalized by an experienced body radiologist (18 years in body CT interpretation) to serve as a reference. Subsequently, the 500 CT scans were randomly divided into three datasets: 400 cases for training, 50 for internal validation, and 50 for internal testing.
The 3D nnU-Net was used for adrenal gland segmentation from the CT images in this study [23]. The 3D nnU-Net model is advantageous in its ability to automatically adapt its pre-processing steps and network hyper-parameters in accordance with the characteristics of the dataset. In our implementation, CT image normalization was executed based on the global CT values found within the foreground voxels across all training cases. This process involved clipping the image values at the 0.5 and 99.5 percentiles to mitigate the impact of extreme values, followed by normalization using the global mean and standard deviation. The input patch size was configured to be 160×192×56. This size was chosen to balance the need for sufficient contextual information and computational efficiency.
To enhance the robustness of our model and prevent overfitting, we incorporated a variety of data augmentation techniques during the training phase, including rotations, scaling, gamma correction, smoothing, and sharp kernel transformations. The network was optimized using the Stochastic Gradient Descent algorithm, augmented with a Nesterov momentum of 0.99. The loss function was a combination of dice loss and cross-entropy loss, selected to optimize the segmentation performance effectively. The initial weight was set to 0.01, and a polynomial learning rate scheduler was employed to adjust the learning rate over time. After 1,000 epochs of training, the model instantiation that demonstrated the highest dice similarity coefficient on the tuning dataset was selected as the final model.
For assessing segmentation accuracy, the dice similarity coefficient, sensitivity, and positive predictive value were employed as metrics [24]. The dice similarity coefficient quantifies the segmentation overlap between the model-generated masks and the reference masks, serving as an indicator of similarity. Sensitivity captures the proportion of voxels identified in the reference mask that are also accurately detected in the model’s mask. Meanwhile, positive predictive value determines the fraction of voxels in the model’s mask that are correctly identified, in comparison to the reference mask. The input data for the network were internal organ masks, and the output was adrenal gland masks. For external testing, adrenal gland masks from a Multi-Atlas Labeling Beyond the Cranial Vault dataset were used [25].
Statistical analyses
Variables with a normal distribution are expressed as means±standard deviations. Those with a skewed distribution are described as medians with interquartile ranges, and categorical variables are expressed as frequencies and percentages. Based on the results of the normality test, continuous variables were compared between the groups using either the independent t test or the Mann–Whitney U test. For categorical variables, the chi-squared or Fisher’s exact test was used, depending on the data. The participants were stratified by sex, and analysis of covariance adjusted for age and BMI was performed to examine the differences in AGV based on glycemic status. Correlation analysis was performed to assess the association between AGV and various metabolic parameters, including age, BMI, WC, SMA, VFA, SFA, TyG index, and HbA1c level. Pearson’s correlation coefficient (r) was calculated. Subsequently, a partial correlation analysis was performed, controlling for both age and BMI, to further elucidate the association between AGV and the aforementioned metabolic variables. Receiver operating characteristic (ROC) curve analysis was performed to determine the optimal AGV cutoff value for predicting diabetes from the cross-sectional data at baseline. The cutoff values showing the highest Youden index were separately identified for men and women. Univariate logistic regression analyses were performed to investigate the association between the baseline T2D and various factors. The association between AGV exceeding the cutoff value and T2D at baseline was further analyzed using multivariate logistic methods after adjusting for confounding factors as follows: model 1, age and sex; model 2, model 1 plus BMI, SMA, VFA, and SFA; and model 3, model 2 plus hypertension, dyslipidemia, and TyG index.
Using the Cox proportional hazards model, we examined the risk of T2D by AGV during the follow-up period for participants without T2D at baseline. The adjusted variables were consistent with those used in the logistic regression analyses, with baseline HbA1c level as a covariate in model 3. Hazard ratios (HRs) are presented with 95% confidence intervals (CIs). Kaplan–Meier estimates were used to evaluate the effects of future T2D development among participants with AGV measurements above the cutoff value. Statistical significance was determined at two-tailed P<0.05. Statistical analyses were performed using R version 4.3.0 (R Foundation for Statistical Computing, Vienna, Austria) and IBM SPSS Statistics for Windows version 26 (IBM Corp., Armonk, NY, USA).
Deep learning performance for adrenal gland segmentation and volume measurement
The 3D nnU-Net averagely demonstrated an average dice similarity coefficient of 0.86±0.03, a sensitivity of 0.84±0.05, and a positive predictive value of 0.88±0.06 in the internal test set (Supplemental Table S2) [20]. When applied to the external test set, the model’s performance showed an average dice similarity coefficient of 0.71±0.11, a sensitivity of 0.65±0.15, and a positive predictive value of 0.80±0.10.
Regarding volume measurements, the volume difference per adrenal gland between nnU-Net generated and reference masks were 0.2±0.4 and 0.6±0.9 mL in the internal and external test sets, respectively. The network-driven adrenal volume was strongly correlated with the reference adrenal volume, achieving R-squared values of 0.89 and 0.78 for the internal and external test sets, respectively.
Baseline characteristics of the clinical cohort
Table 1 shows the baseline characteristics of men and women according to the presence of T2D. Among the 9,708 participants, 59.4% (5,769) were men, and the median age was 52.0 years (Supplemental Table S3) [20]. At baseline, 657 men (11.4%) and 210 women (5.3%) had T2D. Men and women with T2D were significantly older and had a higher prevalence of hypertension and dyslipidemia compared with those without T2D. BMI and VFA were also higher in men and women with T2D compared with those without T2D. In both men and women, participants with T2D had significantly larger volumes of the total, right, and left adrenal glands compared with those without T2D (7.3 cm3 vs. 6.7 cm3, 3.4 cm3 vs. 3.1 cm3, and 3.9 cm3 vs. 3.6 cm3 in men, respectively; 6.3 cm3 vs. 5.5 cm3, 2.8 cm3 vs. 2.5 cm3, and 3.4 cm3 vs. 3.0 cm3 in women, respectively; all P<0.001).
Adrenal gland volume according to the glycemic status at baseline
Fig. 1 presents the adrenal volumes categorized by glycemic status into three groups—NGT, prediabetes, and T2D at baseline—with values adjusted for age and BMI. In men, there was a statistically significant increase in total, right, and left AGV in the order of NGT, prediabetes, and T2D (6.6, 6.9, and 7.3 cm3 for total AGV, all P<0.05). Moreover, in women, there was a trend of increasing total, right, and left AGV in the order of NGT, prediabetes, and T2D (5.4, 5.6, and 6.3 cm3 for total AGV, all P<0.05). These patterns were consistently observed in both sexes, as confirmed in Supplemental Table S4 [20].
Correlation among AGV, body composition, TyG index, and glycated hemoglobin level
Correlation analysis demonstrated a notable positive correlation between AGV and WC (r=0.576), VFA (r=0.407), TyG index (r=0.310), and log-transformed HbA1c level (r=0.151) (all P<0.001) (Fig. 2). Although statistically significant, the correlation between AGV and age was negligible, indicating a slight tendency for AGV to decrease with age (r=–0.033, P=0.001). As age and BMI were also associated with AGV, we performed a partial correlation analysis, controlling for both age and BMI. AGV remained to have a positive association with WC, VFA, TyG index, and log-transformed HbA1c level (all P<0.001) (data not shown).
Adrenal volume as a risk factor for T2D at baseline
In the ROC curve analysis, the areas under the curve for T2D were 0.60 for men and 0.68 for women (Supplemental Fig. S3) [20]. We identified the optimal cutoff values of total AGV for T2D as 7.20 cm3 for men and 5.50 cm3 for women. Total AGVs higher than the cutoff values were significantly associated with higher odds ratios (ORs) for T2D at baseline (OR, 1.919; 95% CI, 1.667 to 2.212) (Table 2). Even after adjusting for potential confounding factors, including age, sex, body composition, comorbidities, and TyG index, the statistical significance of the OR remained (OR, 1.698; 95% CI, 1.442 to 2.002).
Adrenal volume as a marker for future development of T2D in participants undergoing follow-up
Among the 8,841 participants without T2D at baseline, 6,166 were followed up. There was no significant difference in body composition or AGV between those who were followed up and those who were not (Supplemental Table S5) [20]. During the median follow-up duration of 7.0 years (range, 4.0 to 8.0), 17.9% of men (646/3,606) and 11.4% of women (292/2,560) developed T2D. In the Cox proportional hazards regression model for new-onset T2D, AGV, age, sex, BMI, SMA, VFA, hypertension, HbA1c level, and TyG index were significant predictors (Table 3). In the multivariate Cox proportional hazards model based on whether the AGV exceeded the cutoff value for both men and women, participants with a high AGV were at a significantly higher risk of new-onset T2D after adjusting for confounding factors (adjusted HR, 1.268; 95% CI, 1.105 to 1.455; P<0.001). Kaplan–Meier curve analysis also demonstrated a significant association between high baseline AGV and incident T2D (log-rank test, P<0.001) (Fig. 3).
In the present study, we underscored the notable association between AGV and glycemic status in a large-scale cohort of participants without adrenal nodules using AI-based automated segmentation technology on acquired abdominopelvic CT images. After adjusting for multiple confounding factors, a significantly higher AGV was observed in participants with T2D compared with those without. The progressive increase in AGV from NGT to prediabetes and then to T2D further reinforces this association. Moreover, high AGV in participants without T2D at baseline significantly preceded new-onset T2D.
AGV was measured in a large-scale cohort of 9,708 participants using deep learning-based auto-segmentation techniques. The most recent deep learning studies on the adrenal glands have focused on detecting abnormal adrenal glands [18,19]. To date, the results of a study on normal AGV without adrenal nodules have been reported by Chen et al. [26]. Chen et al. [26] measured AGV in a relatively large number of 1,043 participants, and assessed differences by age and sex, but not by associations with metabolic diseases such as T2D. In terms of deep learning model, our model exhibited average dice similarity coefficients of 0.86 for the internal test set and 0.71 for the external test set. Although the segmentation performance was observed to decrease on the external set, it is important to highlight the challenging nature of segmenting adrenal glands, given their small and thin structure. Moreover, human labeling variability of the reference mask adds to the complexity of this task. Our model shows competitive performance with pre-existing models in the field, which have reported dice similarity coefficients ranging from 0.70 to 0.80 in their respective internal test sets [18,19]. Notably, the difference in the adrenal volume between our model’s output and that of the reference was <1 mL, even for the external dataset. This underlines the potential accuracy and reliability of the model for clinical applications.
Previous studies with small sample sizes also reported higher AGV in participants with T2D than in those without T2D [27,28]. The present study confirmed the association between AGV without adrenal disease and T2D in a large number of patients. Although several putative mechanisms have been suggested, the underlying mechanisms remain to be elucidated. AGV has been considered a chronic and stable marker of HPA axis activity [29-31]. Systemic inflammation, marked by elevated levels of inflammatory cytokines, such as tumor necrosis factor-α, interleukin-1β, and C-reactive protein, is closely associated with the development and progression of T2D [13,14]. The HPA axis is the most significant physiological process regulating stress response and hormonal balance [32]. This axis is activated during inflammation, leading to increased cortisol secretion, which initially acts as an anti-inflammatory response. However, persistent inflammation and stress conditions, such as T2D, can trigger the dysregulation of the HPA axis [8-12]. In addition, autonomic dysfunction in patients with T2D may increase the activity of this axis [33]. Certainly, participants with T2D exhibit higher basal levels of adrenocorticotropic hormone (ACTH) compared with those without T2D [6,34]. Elevated ACTH levels can boost cortisol secretion and promote adrenal tissue growth. Moreover, hyperinsulinemia in patients with T2D may stimulate adrenal growth and steroid production [35].
In contrast, glucocorticoids secreted from the adrenal gland itself may affect glucose metabolism [36]. Consistent with this, increased circulating cortisol levels have been found in participants with T2D [37]. As a novel finding distinct from those of previous studies, our study demonstrated that AGV had already increased prior to the development of T2D through longitudinal analysis over a median follow-up of 7.0 years.
Several factors may affect the AGV. We found that men had a significantly higher AGV than women, which is consistent with previous findings [27,28]. Men consistently exhibit larger adrenal volumes in clinical imaging studies, which may reflect differences in lean body mass, visceral adiposity, or metabolic demand rather than stem cell dynamics alone [38,39]. In our sex-specific analysis, AGV remained significantly associated with T2D in both men and women, supporting its potential role as a sex-independent imaging biomarker of metabolic risk. Interestingly, prior reports have shown that AGV correlates more strongly with visceral fat in women, implying that adrenal enlargement may reflect relatively greater metabolic stress in women [40]. These findings highlight the need for future studies to consider sex-specific interpretation when evaluating AGV in clinical and epidemiological contexts. Furthermore, BMI is a potential confounder of the crosstalk between AGV and T2D [8,28]. However, even after controlling for BMI to adjust for body size, AGV remained significantly correlated with WC, VFA, TyG index, and HbA1c level, while maintaining a negative correlation with SFA. This is consistent with the results of previous studies showing an association between AGV and insulin resistance and central obesity [8,41]. However, the association between age and AGV remains controversial. Previous studies involving relatively small sample sizes (<200 participants) have shown that AGV increases as age increases [27,41]. Recently, Chen et al. [26] reported that men showed an increasing then decreasing trend in AGV that peaked at 38 to 47 years of age, using a semiautomated segmentation model in a cohort of approximately 1,000 participants [27]. In our study, age and AGV showed a weak negative correlation.
This study has some limitations. First, as this was an observational study, causality could not be established. Therefore, it remains unclear whether AGV is a cause or a consequence of T2D. Second, the measurement of AGV might depend on the imaging tools and analysis methods. Third, this study was conducted among Koreans who underwent health screening. Therefore, the generalizability of our results should be considered with caution. AGV should be considered along with ethnic variations, and it may be necessary to adjust the cutoff values used to predict diabetes in each ethnic group accordingly [27,28]. Simultaneously, differences in imaging tools and analysis methods should also be taken into consideration. Fourth, participants with adrenal nodules were excluded. However, hormone tests to assess adrenal diseases were not available because of the retrospective nature of the data of participants who underwent health screening. Therefore, patients with idiopathic hyperaldosteronism without adrenal nodules were included in this study. In addition, plasma ACTH and cortisol levels were not available, which limits our ability to directly assess the association between AGV and HPA axis or cortisol production. Furthermore, despite multiple adjustments, central obesity may remain a potential confounder in the observed association between AGV and T2D. Lastly, although the study population consisted of generally healthy adults, the inclusion of a small number of individuals with glucocorticoid exposure cannot be entirely excluded.
In conclusion, we have demonstrated that an increase in AGV reflects the current glycemic status and serves as a significant marker of new-onset T2D. Moreover, a deeper exploration of the underlying mechanisms driving the association between AGV and metabolic parameters can unravel novel avenues for early intervention and management of T2D.

Supplemental Table S1.

CT Protocols in Data for Model Development and Clinical Cohort
enm-2025-2336-Supplemental-Table-S1.pdf

Supplemental Table S2.

Performance of the Deep Learning Model for Adrenal Gland Segmentation
enm-2025-2336-Supplemental-Table-S2.pdf

Supplemental Table S3.

Baseline Characteristics of the Study Population (n = 9,708)
enm-2025-2336-Supplemental-Table-S3.pdf

Supplemental Table S4.

Adrenal Gland Volume according to the Glycemic Status at Baseline
enm-2025-2336-Supplemental-Table-S4.pdf

Supplemental Table S5.

Baseline Characteristics of Subjects without T2D at Baseline, Stratified by Follow-up Status
enm-2025-2336-Supplemental-Table-S5.pdf

Supplemental Fig. S1.

Study flowchart. CT, computed tomography; SNUH, Seoul National University Hospital; DM, diabetes mellitus.
enm-2025-2336-Supplemental-Fig-S1.pdf

Supplemental Fig. S2.

Architecture of the deep learning model for adrenal gland segmentation. 3D, three-dimensional; LReLU, leaky rectified linear unit.
enm-2025-2336-Supplemental-Fig-S2.pdf

Supplemental Fig. S3.

Receiver operating characteristics (ROC) curve analysis for predicting type 2 diabetes using adrenal gland volume. The area under the ROC (AUC) for type 2 diabetes prediction using cutoff values (7.2 cm3 for men and 5.5 cm3 for women) was 0.60 for men and 0.68 for women.
enm-2025-2336-Supplemental-Fig-S3.pdf

CONFLICTS OF INTEREST

Jung Hee Kim is a deputy editor of the journal. But she was not involved in the peer reviewer selection, evaluation, or decision process of this article. No other potential conflicts of interest relevant to this article were reported.

Soon Ho Yoon works in MEDICAL IP as a chief medical officer and has a stock option in the firm. Eu Jeong Ku, Seung Shin Park, Ji Won Yoon, and Jung Hee Kim declare no competing interests.

ACKNOWLEDGMENTS

This study was supported by the National Research Foundation of the Ministry of Science and ICT of Korea (Project No. NRF-2020R1C1C1010723) and a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute funded by the Ministry of Health and Welfare of the Republic of Korea (Project No. HI21C0032 and HI22C0049). The funding sources were not involved in the study design, data collection, analysis, interpretation, or decision to approve the manuscript for publication.

AUTHOR CONTRIBUTIONS

Conception or design: E.J.K., S.H.Y., S.S.P., J.W.Y., J.H.K. Acquisition, analysis, or interpretation of data: E.J.K., S.H.Y., S. S.P., J.W.Y., J.H.K. Drafting the work or revising: E.J.K., S.H.Y., S.S.P., J.W.Y., J.H.K. Final approval of the manuscript: E.J.K., S.H.Y., S.S.P., J.W.Y., J.H.K.

Fig. 1.
Adrenal gland volume (AGV) based on glycemic status at baseline in men and women. AGV based on glycemic status at baseline in (A) men and (B) women. Bars represent mean±standard deviation (SD). The P value represents comparisons within each group based on glycemic status, adjusted for age and body mass index, as determined by analysis of covariance and Bonferroni post hoc analysis. NGT, normal glucose tolerance; T2D, type 2 diabetes. aP<0.05 compared with the NGT group; bP<0.05 compared with the prediabetes group; cP<0.05 compared with the T2D group.
enm-2025-2336f1.jpg
Fig. 2.
Scatterplots of adrenal gland volume (AGV) with (A) waist circumference, (B) visceral fat area (VFA), (C) triglyceride-glucose (TyG) index, (D) log-transformed glycated hemoglobin (HbA1c), (E) age, and (F) body mass index (BMI). The TyG index was calculated using the following formula: Ln (fasting triglyceride [mg/dL]×fasting plasma glucose [mg/dL]/2).
enm-2025-2336f2.jpg
Fig. 3.
Kaplan–Meier analysis shows the time to development of type 2 diabetes (T2D) based on high and low adrenal gland volumes (AGVs). High AGV is defined as AGV ≥7.2 cm3 for men and AGV ≥5.5 cm3 for women. Conversely, low AGV is defined as AGV <7.2 cm3 for men and AGV <5.5 cm3 for women. HR, hazard ratio; CI, confidence interval.
enm-2025-2336f3.jpg
Table 1.
Comparison of Clinical Characteristics in Men and Women Based on the Presence or Absence of T2D at Baseline
Characteristic Men (n=5,769)
Women (n=3,939)
T2D (+) (n=657) T2D (–) (n=5,112) P value T2D (+) (n=210) T2D (–) (n=3,729) P value
Age, yr 57.0 (51.0–63.0) 51.0 (45.0–58.0) <0.001 60.0 (54.0–67.0) 52.0 (46.0–59.0) <0.001
SBP, mm Hg 120.0 (112.0–129.0) 118.0 (110.0–127.0) <0.001 122.0 (111.0–131.0) 112.0 (103.0–123.0) <0.001
DBP, mm Hg 78.0 (72.0–84.0) 77.0 (71.0–84.0) 0.144 74.0 (67.0–80.0) 70.0 (63.0–77.0) <0.001
Height, cm 169.6 (166.1–174.0) 171.1 (167.1–174.9) <0.001 156.8 (153.0–161.0) 158.6 (155.1–162.0) <0.001
Body weight, kg 72.0 (65.9–79.6) 70.9 (65.5–77.2) 0.001 58.8 (53.4–65.0) 54.7 (50.6–59.4) <0.001
BMI, kg/m2 25.0 (23.4–27.0) 24.3 (22.7–26.0) <0.001 23.8 (22.0–26.3) 21.8 (20.1–23.7) <0.001
WC, cm 90.0 (86.0–95.5) 87.8 (83.0–92.5) <0.001 85.5 (81.0–90.5) 79.5 (74.6–85.0) <0.001
HTN 150 (23.2) 522 (10.4) <0.001 58 (27.9) 291 (7.9) <0.001
Dyslipidemia 221 (34.2) 1,258 (25.2) <0.001 70 (33.7) 750 (20.3) <0.001
HbA1c, % 6.8 (6.5–7.4) 5.6 (5.4–5.8) <0.001 6.7 (6.5–7.2) 5.6 (5.4–5.8) <0.001
Glucose, mg/dL 130.0 (116.0–149.0) 95.0 (89.0–102.0) <0.001 121.0 (106.0–136.0) 90.0 (84.0–96.0) <0.001
TC, mg/dL 182.0 (153.0–209.0) 194.0 (172.0–217.0) <0.001 185.0 (163.0–213.0) 197.0 (174.0–222.0) <0.001
TG, mg/dL 124.0 (84.0–181.0) 104.0 (73.0–150.0) <0.001 98.0 (66.0–134.0) 73.0 (52.0–103.0) <0.001
HDL-C, mg/dL 45.0 (40.0–51.0) 47.0 (42.0–54.0) <0.001 51.0 (45.0–59.0) 55.0 (48.0–63.0) <0.001
LDL-C, mg/dL 112.5 (88.0–138.0) 124.0 (105.0–145.0) <0.001 112.0 (92.0–138.0) 120.0 (100.0–143.0) 0.016
TyG index 9.0 (8.6–9.4) 8.5 (8.1–8.9) <0.001 8.7 (8.3–9.1) 8.1 (7.7–8.5) <0.001
SMA, cm2 151.1 (136.3–169.0) 149.6 (135.3–164.9) 0.088 103.1 (95.3–111.7) 99.4 (91.1–107.7) <0.001
VFA, cm2 147.3 (104.7–195.3) 115.3 (72.5–159.0) <0.001 88.2 (53.5–117.1) 40.0 (18.5–70.1) <0.001
SFA, cm2 114.5 (92.2–145.3) 118.9 (92.6–149.4) 0.165 145.6 (113.0–182.2) 126.5 (97.0–163.4) <0.001
AGV, cm3 7.3 (6.3–8.4) 6.7 (5.9–7.7) <0.001 6.3 (5.6–7.0) 5.5 (4.8–6.3) <0.001
RAGV, cm3 3.4 (2.9–3.9) 3.1 (2.7–3.6) <0.001 2.8 (2.4–3.3) 2.5 (2.1–2.9) <0.001
LAGV, cm3 3.9 (3.3–4.5) 3.6 (3.1–4.2) <0.001 3.4 (2.9–3.9) 3.0 (2.6–3.5) <0.001

Values are expressed as median (interquartile range) or number (%). P value represents the comparison between the diabetic and non-diabetic subjects within each group.

T2D, type 2 diabetes; SBP, systolic blood pressure; DBP, diastolic blood pressure; BMI, body mass index; WC, waist circumference; HTN, hypertension; HbA1c, glycated hemoglobin; TC, total cholesterol; TG, triglyceride; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TyG, triglyceride-glucose; SMA, skeletal muscle area; VFA, visceral fat area; SFA, subcutaneous fat area; AGV, adrenal gland volume; RAGV, right adrenal gland volume; LAGV, left adrenal gland volume.

Table 2.
Logistic Regression Analysis for the Association between High AGV and T2D at Baseline
OR 95% CI P value
Unadjusted model 1.919 1.667–2.212 <0.001
Model 1 2.307 1.991–2.675 <0.001
Model 2 1.905 1.629–2.229 <0.001
Model 3 1.698 1.442–2.002 <0.001

High AGV was defined as AGV ≥7.2 cm3 for men and ≥5.5 cm3 for women. Model 1: adjustment for age and sex; Model 2: model 1+body mass index, skeletal muscle area, and visceral fat area; Model 3: model 2+hypertension, dyslipidemia, and triglyceride-glucose index.

AGV, adrenal gland volume; T2D, type 2 diabetes; OR, odds ratio; CI, confidence interval.

Table 3.
Cox-Proportional Hazard Model of Subjects with High AGV for the Risk of Future T2D Development
Variable HR 95% CI P value
High AGV 1.418 1.247–1.612 <0.001
High AGVa 1.268 1.105–1.455 <0.001
Age, yr 1.050 1.044–1.056 <0.001
Male sex 1.721 1.536–1.929 <0.001
BMI, kg/m2 1.135 1.117–1.154 <0.001
SMA, cm2 1.006 1.004–1.007 <0.001
VFA, cm2 1.223 1.199–1.248 <0.001
Hypertension (yes) 1.976 1.733–2.254 <0.001
Dyslipidemia (yes) 1.286 1.148–1.441 <0.001
TyG index 5.325 4.580–6.191 <0.001
Baseline HbA1c, % 2.015 1.957–2.076 <0.001

High AGV was defined as AGV ≥7.2 cm3 for men and ≥5.5 cm3 for women.

AGV, adrenal gland volume; T2D, type 2 diabetes; HR, hazard ratio; CI, confidence interval; BMI, body mass index; SMA, skeletal muscle area; VFA, visceral fat area; TyG index, triglyceride-glucose index; HbA1c, glycated A1c.

a Adjustment for age, sex (male), BMI, SMA, VFA, hypertension, dyslipidemia, TyG index, and baseline HbA1c.

  • 1. ElSayed NA, Aleppo G, Aroda VR, Bannuru RR, Brown FM, Bruemmer D, et al. 2. Classification and diagnosis of diabetes: standards of care in diabetes-2023. Diabetes Care 2023;46(Suppl 1):S19–40.
  • 2. Bae JH, Han KD, Ko SH, Yang YS, Choi JH, Choi KM, et al. Diabetes fact sheet in Korea 2021. Diabetes Metab J 2022;46:417–26.ArticlePubMedPMCPDF
  • 3. International Diabetes Federation. IDF Diabetes Atlas 2025 [Internet]. Brussels: IDF; 2025 [cited 2025 Apr 23]. Available from: https://diabetesatlas.org/resources/idf-diabetes-atlas-2025/.
  • 4. Janssen JAMJL. New insights into the role of insulin and hypothalamic-pituitary-adrenal (HPA) axis in the metabolic syndrome. Int J Mol Sci 2022;23:8178.ArticlePubMedPMC
  • 5. Chan O, Inouye K, Akirav E, Park E, Riddell MC, Vranic M, et al. Insulin alone increases hypothalamo-pituitary-adrenal activity, and diabetes lowers peak stress responses. Endocrinology 2005;146:1382–90.ArticlePubMed
  • 6. Cameron OG, Thomas B, Tiongco D, Hariharan M, Greden JF. Hypercortisolism in diabetes mellitus. Diabetes Care 1987;10:662–4.ArticlePubMedPDF
  • 7. Lentle BC, Thomas JP. Adrenal function and the complications of diabetes mellitus. Lancet 1964;2:544–9.ArticlePubMed
  • 8. Liu F, Chen Y, Xie W, Liu C, Zhu Y, Tian H, et al. Obesity might persistently increase adrenal gland volume: a preliminary study. Obes Surg 2020;30:3503–7.ArticlePubMedPDF
  • 9. Berger I, Werdermann M, Bornstein SR, Steenblock C. The adrenal gland in stress: adaptation on a cellular level. J Steroid Biochem Mol Biol 2019;190:198–206.ArticlePubMed
  • 10. Bruehl H, Rueger M, Dziobek I, Sweat V, Tirsi A, Javier E, et al. Hypothalamic-pituitary-adrenal axis dysregulation and memory impairments in type 2 diabetes. J Clin Endocrinol Metab 2007;92:2439–45.ArticlePubMedPDF
  • 11. Rosmond R, Dallman MF, Bjorntorp P. Stress-related cortisol secretion in men: relationships with abdominal obesity and endocrine, metabolic and hemodynamic abnormalities. J Clin Endocrinol Metab 1998;83:1853–9.ArticlePubMed
  • 12. Weaver JU, Kopelman PG, McLoughlin L, Forsling ML, Grossman A. Hyperactivity of the hypothalamo-pituitary-adrenal axis in obesity: a study of ACTH, AVP, beta-lipotrophin and cortisol responses to insulin-induced hypoglycaemia. Clin Endocrinol (Oxf) 1993;39:345–50.PubMed
  • 13. Badawi A, Klip A, Haddad P, Cole DE, Bailo BG, El-Sohemy A, et al. Type 2 diabetes mellitus and inflammation: prospects for biomarkers of risk and nutritional intervention. Diabetes Metab Syndr Obes 2010;3:173–86.ArticlePubMedPMC
  • 14. Velikova TV, Kabakchieva PP, Assyov YS, Georgiev TА. Targeting inflammatory cytokines to improve type 2 diabetes control. Biomed Res Int 2021;2021:7297419.ArticlePubMedPMCPDF
  • 15. Pivonello R, Isidori AM, De Martino MC, Newell-Price J, Biller BM, Colao A. Complications of Cushing’s syndrome: state of the art. Lancet Diabetes Endocrinol 2016;4:611–29.ArticlePubMed
  • 16. Kahl KG, Schweiger U, Pars K, Kunikowska A, Deuschle M, Gutberlet M, et al. Adrenal gland volume, intra-abdominal and pericardial adipose tissue in major depressive disorder. Psychoneuroendocrinology 2015;58:1–8.ArticlePubMed
  • 17. Luo G, Yang Q, Chen T, Zheng T, Xie W, Sun H. An optimized two-stage cascaded deep neural network for adrenal segmentation on CT images. Comput Biol Med 2021;136:104749.ArticlePubMed
  • 18. Robinson-Weiss C, Patel J, Bizzo BC, Glazer DI, Bridge CP, Andriole KP, et al. Machine learning for adrenal gland segmentation and classification of normal and adrenal masses at CT. Radiology 2023;306:e220101.ArticlePubMed
  • 19. Kim TM, Choi SJ, Ko JY, Kim S, Jeong CW, Cho JY, et al. Fully automatic volume measurement of the adrenal gland on CT using deep learning to classify adrenal hyperplasia. Eur Radiol 2023;33:4292–302.ArticlePubMedPDF
  • 20. Ku EJ, Yoon SH, Park SS, Yoon JW, Kim JH. Deep learning-based adrenal gland volumetry for the prediction of diabetes mellitus: an observational study. Supplementary Materials [Internet]. Figshare Digital Repository; 2024 [cited 2025 Apr 23]. Available from: www.doi.org/10.6084/m9.figshare.26880124.
  • 21. Hur KY, Moon MK, Park JS, Kim SK, Lee SH, Yun JS, et al. 2021 Clinical practice guidelines for diabetes mellitus of the Korean Diabetes Association. Diabetes Metab J 2021;45:461–81.PubMedPMC
  • 22. Lee YS, Hong N, Witanto JN, Choi YR, Park J, Decazes P, et al. Deep neural network for automatic volumetric segmentation of whole-body CT images for body composition assessment. Clin Nutr 2021;40:5038–46.ArticlePubMed
  • 23. Isensee F, Jaeger PF, Kohl SA, Petersen J, Maier-Hein KH. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods 2021;18:203–11.ArticlePubMedPDF
  • 24. Taha AA, Hanbury A. Metrics for evaluating 3D medical image segmentation: analysis, selection, and tool. BMC Med Imaging 2015;15:29.ArticlePubMedPMCPDF
  • 25. Synapse. Multi-atlas labeling beyond the cranial vault: workshop and challenge [Internet]. Seattle: Sage Bionetworks; 2013 [cited 2025 Apr 23]. Available from: https://www.synapse.org/#!Synapse:syn3193805/wiki/217780.
  • 26. Chen Y, Yang J, Zhang Y, Sun Y, Zhang X, Wang X. Age-related morphometrics of normal adrenal glands based on deep learning-aided segmentation. Heliyon 2023;9:e16810.ArticlePubMedPMC
  • 27. Carsin-Vu A, Oubaya N, Mule S, Janvier A, Delemer B, Soyer P, et al. MDCT linear and volumetric analysis of adrenal glands: normative data and multiparametric assessment. Eur Radiol 2016;26:2494–501.ArticlePubMedPDF
  • 28. Askani E, Rospleszcz S, Lorbeer R, Kulka C, von Kruchten R, Muller-Peltzer K, et al. Association of MRI-based adrenal gland volume and impaired glucose metabolism in a population-based cohort study. Diabetes Metab Res Rev 2022;38:e3528.PubMed
  • 29. Golden SH, Wand GS, Malhotra S, Kamel I, Horton K. Reliability of hypothalamic-pituitary-adrenal axis assessment methods for use in population-based studies. Eur J Epidemiol 2011;26:511–25.ArticlePubMedPMCPDF
  • 30. Kong SH, Kim JH, Shin CS. Contralateral adrenal thinning as a distinctive feature of mild autonomous cortisol excess of the adrenal tumors. Eur J Endocrinol 2020;183:325–33.ArticlePubMed
  • 31. Kim TM, Kim JH, Jang HN, Choi MH, Cho JY, Kim SY. Adrenal morphology as an indicator of long-term disease control in adults with classic 21-hydroxylase deficiency. Endocrinol Metab (Seoul) 2022;37:124–37.ArticlePubMedPMCPDF
  • 32. Melmed S, Koenig R, Rosen CJ, Auchus RJ, Goldfine AB. Williams textbook of endocrinology; 14th ed. Philadelphia: Elsevier; 2019.
  • 33. Chan O, Inouye K, Riddell MC, Vranic M, Matthews SG. Diabetes and the hypothalamo-pituitary-adrenal (HPA) axis. Minerva Endocrinol 2003;28:87–102.PubMed
  • 34. Vermes I, Steinmetz E, Schoorl J, van der Veen EA, Tilders FJ. Increased plasma levels of immunoreactive beta-endorphin and corticotropin in non-insulin-dependent diabetes. Lancet 1985;2:725–6.
  • 35. Fruehwald-Schultes B, Kern W, Bong W, Wellhoener P, Kerner W, Born J, et al. Supraphysiological hyperinsulinemia acutely increases hypothalamic-pituitary-adrenal secretory activity in humans. J Clin Endocrinol Metab 1999;84:3041–6.ArticlePubMed
  • 36. van Raalte DH, Ouwens DM, Diamant M. Novel insights into glucocorticoid-mediated diabetogenic effects: towards expansion of therapeutic options? Eur J Clin Invest 2009;39:81–93.ArticlePubMed
  • 37. Di Dalmazi G, Pagotto U, Pasquali R, Vicennati V. Glucocorticoids and type 2 diabetes: from physiology to pathology. J Nutr Metab 2012;2012:525093.PubMedPMC
  • 38. Bechmann N, Moskopp ML, Constantinescu G, Stell A, Ernst A, Berthold F, et al. Asymmetric adrenals: sexual dimorphism of adrenal tumors. J Clin Endocrinol Metab 2024;109:471–82.ArticlePubMedPDF
  • 39. Schneller J, Reiser M, Beuschlein F, Osswald A, Pallauf A, Riester A, et al. Linear and volumetric evaluation of the adrenal gland: MDCT-based measurements of the adrenals. Acad Radiol 2014;21:1465–74.ArticlePubMed
  • 40. Ludescher B, Najib A, Baar S, Machann J, Thamer C, Schick F, et al. Gender specific correlations of adrenal gland size and body fat distribution: a whole body MRI study. Horm Metab Res 2007;39:515–8.ArticlePubMed
  • 41. Gurun E, Kaya M, Gurun KH. Evaluation of normal adrenal gland volume and morphometry and relationship with waist circumference in an adult population using multidetector computed tomography. Sisli Etfal Hastan Tip Bul 2021;55:333–8.PubMedPMC

Figure & Data

References

    Citations

    Citations to this article as recorded by  

      • PubReader PubReader
      • ePub LinkePub Link
      • Cite
        Cite
        export Copy Download
        Close
        Download Citation
        Download a citation file in RIS format that can be imported by all major citation management software, including EndNote, ProCite, RefWorks, and Reference Manager.

        Format:
        • RIS — For EndNote, ProCite, RefWorks, and most other reference management software
        • BibTeX — For JabRef, BibDesk, and other BibTeX-specific software
        Include:
        • Citation for the content below
        Deep Learning-Based Adrenal Gland Volumetry for the Prediction of Diabetes
        Close
      • XML DownloadXML Download
      Figure
      • 0
      • 1
      • 2
      Deep Learning-Based Adrenal Gland Volumetry for the Prediction of Diabetes
      Image Image Image
      Fig. 1. Adrenal gland volume (AGV) based on glycemic status at baseline in men and women. AGV based on glycemic status at baseline in (A) men and (B) women. Bars represent mean±standard deviation (SD). The P value represents comparisons within each group based on glycemic status, adjusted for age and body mass index, as determined by analysis of covariance and Bonferroni post hoc analysis. NGT, normal glucose tolerance; T2D, type 2 diabetes. aP<0.05 compared with the NGT group; bP<0.05 compared with the prediabetes group; cP<0.05 compared with the T2D group.
      Fig. 2. Scatterplots of adrenal gland volume (AGV) with (A) waist circumference, (B) visceral fat area (VFA), (C) triglyceride-glucose (TyG) index, (D) log-transformed glycated hemoglobin (HbA1c), (E) age, and (F) body mass index (BMI). The TyG index was calculated using the following formula: Ln (fasting triglyceride [mg/dL]×fasting plasma glucose [mg/dL]/2).
      Fig. 3. Kaplan–Meier analysis shows the time to development of type 2 diabetes (T2D) based on high and low adrenal gland volumes (AGVs). High AGV is defined as AGV ≥7.2 cm3 for men and AGV ≥5.5 cm3 for women. Conversely, low AGV is defined as AGV <7.2 cm3 for men and AGV <5.5 cm3 for women. HR, hazard ratio; CI, confidence interval.
      Deep Learning-Based Adrenal Gland Volumetry for the Prediction of Diabetes
      Characteristic Men (n=5,769)
      Women (n=3,939)
      T2D (+) (n=657) T2D (–) (n=5,112) P value T2D (+) (n=210) T2D (–) (n=3,729) P value
      Age, yr 57.0 (51.0–63.0) 51.0 (45.0–58.0) <0.001 60.0 (54.0–67.0) 52.0 (46.0–59.0) <0.001
      SBP, mm Hg 120.0 (112.0–129.0) 118.0 (110.0–127.0) <0.001 122.0 (111.0–131.0) 112.0 (103.0–123.0) <0.001
      DBP, mm Hg 78.0 (72.0–84.0) 77.0 (71.0–84.0) 0.144 74.0 (67.0–80.0) 70.0 (63.0–77.0) <0.001
      Height, cm 169.6 (166.1–174.0) 171.1 (167.1–174.9) <0.001 156.8 (153.0–161.0) 158.6 (155.1–162.0) <0.001
      Body weight, kg 72.0 (65.9–79.6) 70.9 (65.5–77.2) 0.001 58.8 (53.4–65.0) 54.7 (50.6–59.4) <0.001
      BMI, kg/m2 25.0 (23.4–27.0) 24.3 (22.7–26.0) <0.001 23.8 (22.0–26.3) 21.8 (20.1–23.7) <0.001
      WC, cm 90.0 (86.0–95.5) 87.8 (83.0–92.5) <0.001 85.5 (81.0–90.5) 79.5 (74.6–85.0) <0.001
      HTN 150 (23.2) 522 (10.4) <0.001 58 (27.9) 291 (7.9) <0.001
      Dyslipidemia 221 (34.2) 1,258 (25.2) <0.001 70 (33.7) 750 (20.3) <0.001
      HbA1c, % 6.8 (6.5–7.4) 5.6 (5.4–5.8) <0.001 6.7 (6.5–7.2) 5.6 (5.4–5.8) <0.001
      Glucose, mg/dL 130.0 (116.0–149.0) 95.0 (89.0–102.0) <0.001 121.0 (106.0–136.0) 90.0 (84.0–96.0) <0.001
      TC, mg/dL 182.0 (153.0–209.0) 194.0 (172.0–217.0) <0.001 185.0 (163.0–213.0) 197.0 (174.0–222.0) <0.001
      TG, mg/dL 124.0 (84.0–181.0) 104.0 (73.0–150.0) <0.001 98.0 (66.0–134.0) 73.0 (52.0–103.0) <0.001
      HDL-C, mg/dL 45.0 (40.0–51.0) 47.0 (42.0–54.0) <0.001 51.0 (45.0–59.0) 55.0 (48.0–63.0) <0.001
      LDL-C, mg/dL 112.5 (88.0–138.0) 124.0 (105.0–145.0) <0.001 112.0 (92.0–138.0) 120.0 (100.0–143.0) 0.016
      TyG index 9.0 (8.6–9.4) 8.5 (8.1–8.9) <0.001 8.7 (8.3–9.1) 8.1 (7.7–8.5) <0.001
      SMA, cm2 151.1 (136.3–169.0) 149.6 (135.3–164.9) 0.088 103.1 (95.3–111.7) 99.4 (91.1–107.7) <0.001
      VFA, cm2 147.3 (104.7–195.3) 115.3 (72.5–159.0) <0.001 88.2 (53.5–117.1) 40.0 (18.5–70.1) <0.001
      SFA, cm2 114.5 (92.2–145.3) 118.9 (92.6–149.4) 0.165 145.6 (113.0–182.2) 126.5 (97.0–163.4) <0.001
      AGV, cm3 7.3 (6.3–8.4) 6.7 (5.9–7.7) <0.001 6.3 (5.6–7.0) 5.5 (4.8–6.3) <0.001
      RAGV, cm3 3.4 (2.9–3.9) 3.1 (2.7–3.6) <0.001 2.8 (2.4–3.3) 2.5 (2.1–2.9) <0.001
      LAGV, cm3 3.9 (3.3–4.5) 3.6 (3.1–4.2) <0.001 3.4 (2.9–3.9) 3.0 (2.6–3.5) <0.001
      OR 95% CI P value
      Unadjusted model 1.919 1.667–2.212 <0.001
      Model 1 2.307 1.991–2.675 <0.001
      Model 2 1.905 1.629–2.229 <0.001
      Model 3 1.698 1.442–2.002 <0.001
      Variable HR 95% CI P value
      High AGV 1.418 1.247–1.612 <0.001
      High AGVa 1.268 1.105–1.455 <0.001
      Age, yr 1.050 1.044–1.056 <0.001
      Male sex 1.721 1.536–1.929 <0.001
      BMI, kg/m2 1.135 1.117–1.154 <0.001
      SMA, cm2 1.006 1.004–1.007 <0.001
      VFA, cm2 1.223 1.199–1.248 <0.001
      Hypertension (yes) 1.976 1.733–2.254 <0.001
      Dyslipidemia (yes) 1.286 1.148–1.441 <0.001
      TyG index 5.325 4.580–6.191 <0.001
      Baseline HbA1c, % 2.015 1.957–2.076 <0.001
      Table 1. Comparison of Clinical Characteristics in Men and Women Based on the Presence or Absence of T2D at Baseline

      Values are expressed as median (interquartile range) or number (%). P value represents the comparison between the diabetic and non-diabetic subjects within each group.

      T2D, type 2 diabetes; SBP, systolic blood pressure; DBP, diastolic blood pressure; BMI, body mass index; WC, waist circumference; HTN, hypertension; HbA1c, glycated hemoglobin; TC, total cholesterol; TG, triglyceride; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; TyG, triglyceride-glucose; SMA, skeletal muscle area; VFA, visceral fat area; SFA, subcutaneous fat area; AGV, adrenal gland volume; RAGV, right adrenal gland volume; LAGV, left adrenal gland volume.

      Table 2. Logistic Regression Analysis for the Association between High AGV and T2D at Baseline

      High AGV was defined as AGV ≥7.2 cm3 for men and ≥5.5 cm3 for women. Model 1: adjustment for age and sex; Model 2: model 1+body mass index, skeletal muscle area, and visceral fat area; Model 3: model 2+hypertension, dyslipidemia, and triglyceride-glucose index.

      AGV, adrenal gland volume; T2D, type 2 diabetes; OR, odds ratio; CI, confidence interval.

      Table 3. Cox-Proportional Hazard Model of Subjects with High AGV for the Risk of Future T2D Development

      High AGV was defined as AGV ≥7.2 cm3 for men and ≥5.5 cm3 for women.

      AGV, adrenal gland volume; T2D, type 2 diabetes; HR, hazard ratio; CI, confidence interval; BMI, body mass index; SMA, skeletal muscle area; VFA, visceral fat area; TyG index, triglyceride-glucose index; HbA1c, glycated A1c.

      Adjustment for age, sex (male), BMI, SMA, VFA, hypertension, dyslipidemia, TyG index, and baseline HbA1c.


      Endocrinol Metab : Endocrinology and Metabolism
      TOP