Endocrinology and Metabolism

Articles in E-pub version are posted online ahead of regular printed publication.

Corrigendum: Correction of Figure

High Cardiorespiratory Fitness Protects against Molecular Impairments of Metabolism, Heart, and Brain with Higher Efficacy in Obesity-Induced Premature Aging: Patcharapong Pantiya, Chanisa Thonusin, Natticha Sumneang, Benjamin Ongnok, Titikorn Chunchai, Sasiwan Kerdphoo, Thidarat Jaiwongkam, Busarin Arunsak, Natthaphat Siri-Angkul, Sirawit Sriwichaiin, Nipon Chattipakorn, Siriporn C. Chattipakorn; Published online June 23, 2025; DOI: https://doi.org/10.3803/EnM.2025.401 [Epub ahead of print]

14 View
1 Download

PDF

Original Articles

Deep Learning-Based Adrenal Gland Volumetry for the Prediction of Diabetes: Eu Jeong Ku, Soon Ho Yoon, Seung Shin Park, Ji Won Yoon, Jung Hee Kim; Received February 6, 2025 Accepted April 7, 2025 Published online June 18, 2025; DOI: https://doi.org/10.3803/EnM.2025.2336 [Epub ahead of print]

111 View
9 Download

Abstract PDF: 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.

Osteoporosis Management after the Occurrence of Medication-Related Osteonecrosis of the Jaw: A 13-Year Experience at a Tertiary Center: Chun Ho Wong, Kimberly Hang Tsoi, Jingya Jane Pu, Nancy Su Jiang, Stacey Sheung Yi Chan, Connie Hong Nin Loong, Xincheng Zou, Carol Ho Yi Fong, Eunice Ka Hong Leung, Alan Chun Hong Lee, Chi Ho Lee, Kathryn Choon Beng Tan, Yu Cho Woo, Yu-xiong Su, David Tak Wai Lui; Received December 2, 2024 Accepted March 26, 2025 Published online June 13, 2025; DOI: https://doi.org/10.3803/EnM.2024.2262 [Epub ahead of print]

239 View
87 Download

Abstract PDF PubReader ePub: Background
We investigated osteoporosis management strategies and clinical outcomes following the occurrence of medicationrelated osteonecrosis of the jaw (MRONJ).
Methods
We retrospectively studied individuals diagnosed with MRONJ during osteoporosis treatment who were managed in the Osteoporosis Center or the Oral Maxillofacial Surgery & Dental Unit at Queen Mary Hospital in Hong Kong between 2010 and 2022. We examined subsequent osteoporosis management plans, fracture events, and bone mineral density (BMD).
Results
Thirty-six individuals were included (mean age, 78.5 years; 94.4% women). The estimated prevalence of MRONJ was 0.26%. All patients had been exposed to bisphosphonates, and seven had also received denosumab before MRONJ. Following MRONJ, only 14 individuals continued anti-osteoporosis treatment, a decision influenced by a higher fracture probability at MRONJ onset. The most common regimen was a teriparatide-raloxifene sequence (n=8): three patients achieved stable BMD, four achieved improving BMD, and one exhibited a mixed response. The patient with a mixed BMD response had also been treated with denosumab. Six patients sustained incident fractures after MRONJ, and these patients had lower BMD T-scores than their counterparts. Two patients experienced MRONJ recurrence, which was associated with the resumption of bisphosphonate or denosumab therapy after MRONJ. These patients had higher BMD T-scores than those who did not experience MRONJ recurrence.
Conclusion
MRONJ is challenging because high fracture risk necessitates discontinuation of antiresorptive agents. Teriparatide followed by raloxifene may be a reasonable regimen. Individualised decisions in osteoporosis management after MRONJ are required to balance fracture risk reduction with minimising MRONJ recurrence.

Editorials

Calorie Restriction and Bone Health: Lessons from Bariatric Metabolic Surgery: Bukyung Kim; Received May 15, 2025 Accepted May 20, 2025 Published online June 11, 2025; DOI: https://doi.org/10.3803/EnM.2025.2452 [Epub ahead of print]

104 View
11 Download

PDF PubReader ePub

Long-Term Outcomes of Radiofrequency Ablation in Patients with Low-Risk Papillary Thyroid Microcarcinoma: Hunjong Lim, Min Joo Kim; Received May 5, 2025 Accepted May 19, 2025 Published online June 11, 2025; DOI: https://doi.org/10.3803/EnM.2025.2432 [Epub ahead of print]

120 View
11 Download

PDF PubReader ePub

Original Articles

Comparison of Ultrasensitive and Highly Sensitive Assay to Predict Stimulated Thyroglobulin Levels Using Unstimulated Levels in Differentiated Thyroid Cancer Patients: Jinsun Jang, Hyun Joo Kim, Seunggyun Ha, Kyong Yeun Jung, Gyeongseo Jung, Sun Wook Cho, Do Joon Park, Gi Jeong Cheon, Young Joo Park; Received January 6, 2025 Accepted March 20, 2025 Published online June 5, 2025; DOI: https://doi.org/10.3803/EnM.2025.2302 [Epub ahead of print]

269 View
14 Download

Abstract PDF Supplementary Material PubReader ePub: Background
Thyroglobulin (Tg) measurement is an essential aspect of monitoring for differentiated thyroid cancer (DTC) patients. This study compared the performances of ultrasensitive Tg (ultraTg) and highly sensitive Tg (hsTg) assays in predicting stimulated Tg levels without thyroid-stimulating hormone stimulation.
Methods
Overall, 268 DTC patients who had undergone total thyroidectomy and either radioiodine treatment or I-123 diagnostic scanning were included. Unstimulated and stimulated Tg levels were measured using hsTg (BRAHMS Dynotest Tg-plus) and ultraTg (RIAKEY Tg immunoradiometric assay) assays. Correlations of each assay with the ability of unstimulated Tg levels to predict stimulated Tg ≥1 ng/mL were analyzed.
Results
hsTg and ultraTg showed a strong correlation (R=0.79, P<0.01); the correlation was weaker in Tg antibody-positive patients (R=0.52). UltraTg demonstrated higher sensitivity in predicting stimulated Tg ≥1 ng/mL compared with hsTg. The optimal cut-off for ultraTg was 0.12 ng/mL (sensitivity, 72.0%; specificity, 67.2%). hsTg at 0.105 ng/mL had lower sensitivity (39.8%) but higher specificity (91.5%). Eight discordant cases with low hsTg (<0.2 ng/mL) but elevated ultraTg (>0.23 ng/mL) were identified; three developed structural recurrence within 3.4 to 5.8 years. Two patients had an excellent response according to hsTg but an indeterminate or biochemical incomplete response according to ultraTg.
Conclusion
UltraTg demonstrated higher sensitivity in predicting positive stimulated Tg levels and potential recurrence compared with hsTg. However, its lower specificity may lead to more frequent classifications of biochemical incomplete response. UltraTg may be beneficial in clinically suspicious cases where hsTg falls below the cut-off, but its broader applicability requires further investigation.

Fatty Liver Index Dynamics as a Predictor of Hepatocellular Carcinoma in Patients with Type 2 Diabetes Mellitus and Non-Cirrhotic Livers: Eun-Hee Cho, Min Gu Kang, Chang Hun Lee, Shinyoung Oh, Chen Shen, Ha Ram Oh, Young Ran Park, Hyun Lee, Jong Seung Kim, Ji Hyun Park; Received December 15, 2024 Accepted March 13, 2025 Published online May 29, 2025; DOI: https://doi.org/10.3803/EnM.2024.2286 [Epub ahead of print]

356 View
19 Download

Abstract PDF Supplementary Material PubReader ePub: Background
Type 2 diabetes mellitus (T2DM) is a significant risk factor for hepatocellular carcinoma (HCC) in patients with nonalcoholic fatty liver disease; however, surveillance strategies for patients with T2DM, especially without cirrhosis, are inadequate. This study examined whether the fatty liver index (FLI) and its dynamic changes can effectively identify patients with T2DM at increased risk for HCC.
Methods
Data from 92,761 individuals with T2DM aged 40 to 79 who underwent two health screenings (2012 to 2015) were analyzed. The FLI, calculated using waist circumference, body mass index, triglycerides, and gamma-glutamyl transferase, was used to stratify patients by baseline FLI and FLI changes between screenings. HCC cases were identified via International Classification of Diseases codes and reimbursement records (2016 to 2020).
Results
Patients with baseline FLI of 30 to 59.9 had a 1.90-fold higher risk (P<0.01) and those with FLI ≥60 had a 2.94-fold higher risk (P<0.01) of developing HCC compared to those with FLI <30. An increase in FLI from <30 to ≥30 resulted in a 2.10-fold higher risk of HCC (P<0.01), while a reduction in FLI from ≥30 to <30 led to a 0.64-fold lower risk (P=0.03). Protective benefits of FLI reduction took approximately 3 years to manifest.
Conclusion
Baseline and dynamic monitoring of FLI effectively identified HCC risk in T2DM patients with non-cirrhotic livers, supporting early detection and intervention.

Current Status of Delay in Injectable Therapy among Type 2 Diabetes Mellitus Patients in South Korea: Multicenter Retrospective Study (2015–2021): Jong Han Choi, Min Kyong Moon, Hae Jin Kim, Kyung Ae Lee, Hyun Jin Kim, Jung Hae Ko, Jae-Seung Yun, Seung-Hyun Ko, Suk Chon, Nam Hoon Kim; Received December 14, 2024 Accepted March 12, 2025 Published online May 29, 2025; DOI: https://doi.org/10.3803/EnM.2024.2280 [Epub ahead of print]

284 View
19 Download

Abstract PDF Supplementary Material PubReader ePub: Background
We aimed to assess the therapeutic inertia associated with injectable therapies and the factors influencing glycemic control following these therapies in patients with type 2 diabetes mellitus (T2DM) in South Korea.
Methods
This multicenter, retrospective cohort study included 2,598 T2DM patients aged 20 to 75 years from 10 referral medical centers in South Korea. These patients had been treated with three or four oral antidiabetic drugs (OADs) and were subsequently initiated on insulin (n=1,942) or glucagon-like peptide-1 receptor agonists (GLP-1RAs, n=656) between January 2015 and December 2021. We analyzed the time to initiation of injectable therapy, changes in glycated hemoglobin (HbA1c), and associations between clinical factors and glycemic control.
Results
At the time of injectable therapy initiation, the mean HbA1c was 9.54%, with insulin users having a higher HbA1c level (9.79%) than GLP-1RA users (8.70%). The mean time from starting 3 or 4 OADs to initiating injectable therapy was 3.19 years: 53.5% of patients had started injectable therapy after 2 years, and 24.2% started after 5 years. Among insulin users, older age (P= 0.004), higher body mass index (P=0.035), and lower HbA1c levels at insulin initiation (P<0.001) were associated with better glycemic control. Among GLP-1RA users, only the HbA1c level at therapy initiation (P<0.001) was a significant factor.
Conclusion
This study highlighted significant delays in initiating injectable therapies, particularly insulin, in T2DM patients in South Korea. Early initiation of injectable therapy may improve long-term glycemic control in these patients.

Risk Stratification of Thyroid Nodules Diagnosed as Follicular Neoplasm on Core Needle Biopsy: Byeong-Joo Noh, Won Jun Kim, Jin Yub Kim, Ha Young Kim, Jong Cheol Lee, Myoung Sook Shim, Yong Jin Song, Kwang Hyun Yoon, In-Hye Jung, Hyo Sang Lee, Wooyul Paik, Dong Gyu Na; Received November 28, 2024 Accepted March 10, 2025 Published online May 28, 2025; DOI: https://doi.org/10.3803/EnM.2024.2256 [Epub ahead of print]

246 View
17 Download

Abstract PDF Supplementary Material PubReader ePub: Background
This study assessed risk stratification and diagnostic performance for malignancy in thyroid nodules diagnosed as follicular neoplasm (FN) based on core needle biopsy (CNB) subcategories.
Methods
A total of 313 consecutive nodules (>1 cm) diagnosed as FN on CNB with corresponding surgical histology were included. FN subcategories were classified retrospectively for nodules diagnosed before 2022 (retrospective dataset) and prospectively for nodules diagnosed since 2022 (prospective dataset). CNB subcategories were determined using histologic criteria based on architectural uniformity and nuclear atypia, as modified from the 2019 Korean CNB pathology guideline. The diagnostic performance of CNB subcategories, nodule size, and ultrasound risk stratification systems (RSSs) for malignancy was assessed.
Results
CNB subcategory IVb showed a significantly higher malignancy risk compared to other subcategories in both datasets (34.5%–83.7% vs. 4.2%–13.6%, P<0.001). It was also identified as an independent predictor of malignancy in both datasets (P< 0.001), whereas nodule size and all ultrasound RSSs were not predictive of malignancy, including noninvasive follicular thyroid neoplasm with papillary-like nuclear features (NIFTP) (P≥0.079). CNB subcategory IVb demonstrated higher sensitivity for malignancy and a lower surgical rate for benign nodules compared to the nodule size criterion (>2 cm). The combined criterion of CNB subcategory IVb or nodule size >3 cm identified all malignant tumors, excluding NIFTP, in the prospective dataset.
Conclusion
CNB subcategory IVb effectively stratifies malignancy risk in thyroid nodules and outperforms nodule size (>2 cm) and ultrasound RSSs in diagnostic performance. Non-IVb nodules ≤3 cm can be safely managed with ultrasound surveillance instead of immediate surgery.

Comprehensive Evaluation of Treatment Patterns in Postmenopausal Patients with Osteoporosis without Fractures: Insights from Tertiary Care Institutions and Nationwide OMOP-CDM Data: Kyoung Jin Kim, Dachung Boo, Jimi Choi, Hyemin Yoon, Chai Young Jung, Seong Hee Ahn, Namki Hong, Beom-Jun Kim, Ji Seon Oh, Seng Chan You; Received November 22, 2024 Accepted February 27, 2025 Published online May 28, 2025; DOI: https://doi.org/10.3803/EnM.2024.2252 [Epub ahead of print]

444 View
53 Download

Abstract PDF Supplementary Material PubReader ePub: Background
Osteoporosis is a global health concern. Despite emerging treatment options for this condition, limited data are available on hospital practices in South Korea. This study addresses the need for a hospital network database that reflects changes in routine clinical practice for osteoporosis in a timely manner.
Methods
We analyzed prescription patterns for anti-osteoporosis medications (AOMs) in postmenopausal women aged ≥50 years diagnosed with osteoporosis between 2012 and 2021 using data from Osteoporosis Analysis and Surveillance Initiative using Standardized data (OASIS) (four tertiary hospitals in South Korea) and a nationwide database from the Health Insurance Review and Assessment (HIRA) Service. AOMs were categorized into antiresorptive and anabolic agents, with a focus on secular changes in the use of oral bisphosphonates, denosumab, selective estrogen receptor modulators (SERMs), and anabolic agents.
Results
In the OASIS cohort, oral bisphosphonates were the most prescribed first-line AOM (49.0%), followed by denosumab (15.7%) and SERMs (18.0%). Denosumab use increased from 2% in 2016 to 40% in 2020, while oral bisphosphonate use declined from 69% in 2012 to 22% in 2021. The use of anabolic agents, including romosozumab and teriparatide, doubled to 6% after 2019. In the HIRA cohort, parenteral bisphosphonates were most common (54.3%), with significant denosumab use (17.3%).
Conclusion
Pronounced shifts in AOM prescription patterns were observed in South Korea, marked by a notable increase in denosumab prescriptions and a decline in bisphosphonate use. These trends highlight the impact of policy changes and clinical guidelines on osteoporosis treatment and may inform future management strategies.

Clinical Outcomes of Follicular Thyroid Carcinoma Did Not Significantly Differ according to Tumor Size in an Iodine-Excessive Country: Da Eun Leem, Ji Hyun Yoo, Bo Ram Kim, Jung Sun Kim, Tae Hyuk Kim, Sun Wook Kim, Yun Jae Chung, Jae Hoon Chung, Young Lyun Oh; Received January 19, 2025 Accepted March 24, 2025 Published online May 26, 2025; DOI: https://doi.org/10.3803/EnM.2025.2324 [Epub ahead of print]

254 View
25 Download

Abstract PDF PubReader ePub: Background
Follicular thyroid carcinoma (FTC) measuring <2 cm is generally associated with good prognosis, while tumor size ≥4 cm is regarded as being associated with a poor prognosis. This study aimed to evaluate FTC prognosis by tumor size to investigate whether the 2- and 4-cm criteria are appropriate for assessing prognosis.
Methods
Data of 248 patients with FTC diagnosed between August 1995 and June 2021 were retrospectively analyzed. The population was divided into four groups according to tumor size: <2.0, 2.0–3.9, 4.0–5.9, and ≥6.0 cm. Distant metastasis (DM), recurrence and/or structural persistence (R/SP), cancer-specific death (CSD), and frequency of telomerase reverse transcriptase (TERT) promoter mutations based on tumor size were evaluated.
Results
While the rates of DM, R/SP, and CSD and the frequency of TERT promoter mutations did not differ among the size groups <6 cm, they increase sharply in tumor size ≥6 cm, although statistically insignificant (P=0.608, P=0.248, P=0.089, and P=0.165 respectively). Widely invasive subtypes, and TERT promoter mutations were significantly associated with DM (P=0.009 and P<0.001, respectively). Age ≥55 years, gross extrathyroidal extension, synchronous DM, and TERT promoter mutation were independent risk factors for CSD (P=0.005, P=0.003, P<0.001, and P=0.002, respectively).
Conclusion
DM, R/SP, CSD, and TERT promoter mutations were not uncommon in FTCs <2 cm compared to those in larger FTCs, whereas FTCs ≥6 cm showed a sharp decline in prognosis, although this was statistically insignificant.

Exploring Sex Differences in Type 2 Diabetes via a Male-Dominant Beta-Cell Cluster from Single-Cell Pancreatic Sequencing of Public Datasets: Nguyen Van Anh, Hyo-Wook Gil, Samel Park, Seongho Ryu; Received December 27, 2024 Accepted February 18, 2025 Published online May 19, 2025; DOI: https://doi.org/10.3803/EnM.2025.2297 [Epub ahead of print]

598 View
15 Download

Abstract PDF PubReader ePub: Background
Type 2 diabetes is a complex metabolic disorder characterized by insulin resistance and progressive beta-cell dysfunction. Although sex differences in type 2 diabetes prevalence, progression, and complications have been reported, the molecular mechanisms underlying these differences remain largely unknown. We aimed to utilize single-cell RNA sequencing to identify a beta-cell cluster that is more prevalent in males than in females and exhibits distinct gene expression patterns, gene set enrichment profiles, and cell-cell communication compared to other clusters.
Methods
FASTQ files from four public datasets were preprocessed, aligned to the human genome (GRCh38), and integrated into a high-quality matrix to mitigate batch effects. We focused on beta-cells from type 2 diabetes patients, performed trajectory inference to identify clusters, and conducted differential gene expression and gene set enrichment analyses. These findings were validated using bulk RNA-seq datasets. Additionally, cell-cell communication analysis was performed to identify ligand-receptor interactions, followed by a sensitivity analysis to assess sex-specific differences.
Results
We identified a male-dominant beta-cell cluster (adjusted P value=4.2×10^–6) that displayed unique gene expression patterns and downregulation of pathways associated with protein metabolism and insulin synthesis. Differentially expressed genes (e.g., interleukin 24 [IL24], regulator of G protein signaling like 1 [RGSL1]) were confirmed through bulk analysis. Moreover, the cluster demonstrated distinct communication patterns with other cell types, underscoring sex-specific differences.
Conclusion
We have identified a male-dominant beta-cell cluster characterized by distinct gene expression, signaling pathways, and cell interactions. These findings provide insights into the pathophysiology of type 2 diabetes and may inform the development of more effective, sex-specific therapeutic strategies in the future.

The Triglyceride-Glucose Index and Risk of End-Stage Renal Disease across Different Durations of Type 2 Diabetes Mellitus: A Longitudinal Cohort Study: Mi-sook Kim, Kyu-Na Lee, Jeongmin Lee, Jeongeun Kwak, Seung-Hwan Lee, Hyuk-Sang Kwon, Jing Hughes, Kyung-Do Han, Eun Young Lee; Received December 9, 2024 Accepted March 4, 2025 Published online May 19, 2025; DOI: https://doi.org/10.3803/EnM.2024.2271 [Epub ahead of print]

362 View
17 Download

Abstract PDF Supplementary Material PubReader ePub: Background
This study investigated the association between the triglyceride-glucose (TyG) index, a marker of insulin resistance, and the risk of end-stage renal disease (ESRD) in individuals with type 2 diabetes mellitus (T2DM), focusing on variations by diabetes duration.
Methods
We analyzed 1,219,148 Korean adults with T2DM from National Health Insurance Service data who underwent biennial health evaluations (2015 to 2016). ESRD was defined using specific procedural codes (V codes), and Cox proportional hazard models were employed to estimate hazard ratios (HRs) for ESRD across TyG index quartiles and diabetes duration categories, adjusting for various confounders.
Results
Over 6,967,381 person-years of follow-up, 7,548 participants developed ESRD. Higher TyG index quartiles were independently associated with increased risk of ESRD, which was more pronounced with longer diabetes duration. The adjusted HR for ESRD in the highest TyG quartile (Q4) compared to the lowest quartile (Q1) was 1.235 (95% confidence interval [CI], 0.995 to 1.533) in new-onset diabetes, and 1.592 (95% CI, 1.465 to 1.730) in those with diabetes for ≥10 years. Compared to the lowest TyG quartile in new-onset diabetes, the adjusted HR for ESRD in the highest quartile with diabetes duration ≥10 years increased to 10.239 (95% CI, 8.440 to 12.422). Subgroup analysis revealed that a higher TyG index consistently increased the risk of ESRD, with stronger associations observed in younger individuals and those without comorbidities.
Conclusion
The TyG index is a significant predictor of ESRD in T2DM, particularly in those with prolonged diabetes duration. Targeting insulin resistance early may mitigate the risk of ESRD in this population.

Brief Report

Causal Associations of Cardiovascular Proteins and Diabetic Nephropathy Revealed by Mediation and Phenome-Wide Mendelian Randomization: Lei Chen, Yongdi Zuo, Manrong He, Jingxue Du, Wanxin Tang; Received October 15, 2024 Accepted January 23, 2025 Published online May 19, 2025; DOI: https://doi.org/10.3803/EnM.2024.2207 [Epub ahead of print]

340 View
10 Download

Abstract PDF Supplementary Material PubReader ePub: Cardiovascular factors play a critical role in the progression of diabetic nephropathy (DN). This study utilized Mendelian randomization analysis to explore the causal relationships among cardiovascular proteins, risk factors, and DN. Using cis-protein quantitative trait loci (cis-pQTL) data for 90 cardiovascular proteins, we identified C-X3-C motif chemokine ligand 1 (CX3CL1) and colony-stimulating factor-1 (CSF-1) as potential therapeutic targets for DN. CX3CL1 was found to increase DN risk through mechanisms involving body mass index, fasting insulin, and hypertension. Conversely, CSF-1 appeared to protect against DN by reducing the number of monocytes expressing high levels of human leukocyte antigen. Additionally, targeting CX3CL1 could lower the risk of DN as well as other conditions, including acute renal injury, pituitary disorders, and necrotizing vasculopathies. These results highlight CX3CL1 and CSF-1 as promising novel therapeutic targets for DN.

Original Article

Synthetic Data-Enhanced Classification of Prevalent Osteoporotic Fractures Using Dual-Energy X-Ray Absorptiometry-Based Geometric and Material Parameters: Luca Quagliato, Jiin Seo, Jiheun Hong, Taeyong Lee, Yoon-Sok Chung; Received October 22, 2024 Accepted January 8, 2025 Published online May 14, 2025; DOI: https://doi.org/10.3803/EnM.2024.2211 [Epub ahead of print]

248 View
11 Download

Abstract PDF PubReader ePub: Background
Bone fracture risk assessment for osteoporotic patients is essential for implementing early countermeasures and preventing discomfort and hospitalization. Current methodologies, such as Fracture Risk Assessment Tool (FRAX), provide a risk assessment over a 5- to 10-year period rather than evaluating the bone’s current health status.
Methods
The database was collected by Ajou University Medical Center from 2017 to 2021. It included 9,260 patients, aged 55 to 99, comprising 242 femur fracture (FX) cases and 9,018 non-fracture (NFX) cases. To model the association of the bone’s current health status with prevalent FXs, three prediction algorithms—extreme gradient boosting (XGB), support vector machine, and multilayer perceptron—were trained using two-dimensional dual-energy X-ray absorptiometry (2D-DXA) analysis results and subsequently benchmarked. The XGB classifier, which proved most effective, was then further refined using synthetic data generated by the adaptive synthetic oversampler to balance the FX and NFX classes and enhance boundary sharpness for better classification accuracy.
Results
The XGB model trained on raw data demonstrated good prediction capabilities, with an area under the curve (AUC) of 0.78 and an F1 score of 0.71 on test cases. The inclusion of synthetic data improved classification accuracy in terms of both specificity and sensitivity, resulting in an AUC of 0.99 and an F1 score of 0.98.
Conclusion
The proposed methodology demonstrates that current bone health can be assessed through post-processed results from 2D-DXA analysis. Moreover, it was also shown that synthetic data can help stabilize uneven databases by balancing majority and minority classes, thereby significantly improving classification performance.

First
Prev
Page of 3
Next
Last

Ahead-of print