Mehmet Poyrazer

Dr. Mehmet Poyrazer

orcid

Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
16 April 2026 in Frontiers in Medicine
Verified
Journal Article
Open access
PDF file
MPMehmet Poyrazer
Mehmet Poyrazer
HYHüseyin Yağcı
Hüseyin Yağcı
RERıdvan Erten
Rıdvan Erten
Background: Pretrained foundation models are increasingly adopted for diabetic retinopathy (DR) screening, yet it remains unclear how much of their performance derives from the learned representations versus the adaptation procedure. Most benchmarks report discrimination metrics alone, neglecting probability calibration. Methods: We compared the frozen representations of three pretrained encoders: MedSigLIP (medical vision–language; ViT-B/16, 448 × 448), RETFound (retinal self-supervised; ViT-L/16, 224 × 224), and EfficientNet-B0 (ImageNet-supervised; 224 × 224). All encoder weights were frozen; only an identical lightweight multilayer perceptron head was trained. Models were developed on APTOS 2019 (3,662 fundus images; five-fold cross-validation) and externally validated on MESSIDOR-2 (1,744 images). Binary referable DR detection and five-class severity grading were evaluated. AUC, expected calibration error (ECE), and Brier score served as co-primary endpoints. External-set tests used patient-level cluster-robust bootstrap to account for bilateral correlation. Results: On the development set, all three encoders achieved near-identical binary AUC (0.980–0.985). MedSigLIP showed superior calibration, with a lower Brier score than RETFound (0.044 vs. 0.049; p = 0.030) and EfficientNet-B0 (0.044 vs. 0.052; p = 0.006). External validation on MESSIDOR-2 revealed divergence: MedSigLIP maintained an AUC of 0.915 (drop 0.070), whereas RETFound fell to 0.697 (drop 0.286) and EfficientNet-B0 to 0.745 (drop 0.236). Retina-specific RETFound performed below the ImageNet baseline (ΔAUC = −0.051; p = 0.016, cluster-robust bootstrap). For five-class grading, MedSigLIP achieved an external macro-F1 of 0.450 versus 0.247 (RETFound) and 0.291 (EfficientNet-B0). Temperature scaling reduced development ECE to 0.014–0.022 but proved ineffective under domain shift (external ECE 0.086–0.149). All encoders exhibited catastrophic failure on mild DR (grade 1) externally, with RETFound and EfficientNet-B0 achieving F1 = 0.000 and MedSigLIP only 0.153. Conclusion: Under frozen transfer, the MedSigLIP encoder package produced more generalisable and better calibrated representations than both retinal self-supervised (RETFound) and ImageNet-supervised (EfficientNet-B0) encoders. Domain-specific pretraining did not guarantee domain-general frozen representations. These findings demonstrate that development-set discrimination alone is insufficient for encoder evaluation and that calibration metrics—particularly the Brier score—should be reported as standard practice.

10 April 2026 in Frontiers in Endocrinology
Verified
Journal Article
Open access
PDF file
MPMehmet Poyrazer
Mehmet Poyrazer
RERıdvan Erten
Rıdvan Erten
Background: Fine-needle aspiration biopsy (FNAB) cytology is central to thyroid nodule evaluation, yet reliable differentiation across Bethesda categories remains challenging, particularly for the indeterminate Bethesda V (Suspicious for Malignancy) class. While transfer learning with ImageNet-pretrained models is a standard approach, it remains unclear whether emerging domain-specific medical foundation models offer superior performance compared to general purpose baselines in this specialized domain. Methods: We benchmarked four frozen visual encoders—ResNet50, EfficientNet-B0, ViT-Base (ImageNet pretrained), and MedSigLIP (medical image–text pretrained)—on the ThyroidEffi 1.0 dataset (N = 1,804), comprising Bethesda II (Benign), Bethesda V (Suspicious), and Bethesda VI (Malignant) cases. A unified evaluation protocol was employed using five-fold stratified cross-validation with a lightweight multilayer perceptron head. Performance was assessed using macro-F1, balanced accuracy, Expected Calibration Error (ECE), and McNemar’s test for statistical significance. Results: EfficientNet achieved the highest macro-F1 (0.845 ± 0.021), followed closely by MedSigLIP (0.836 ± 0.019), ResNet50 (0.829 ± 0.015), and ViT (0.817 ± 0.020). Pairwise statistical testing revealed that while EfficientNet significantly outperformed ViT (p < 0.05), the difference between EfficientNet and MedSigLIP was not statistically significant after multiple comparison correction. Notably, MedSigLIP demonstrated superior reliability attributes, achieving the highest recall for the challenging Suspicious class (0.808) and the best model calibration score (ECE = 0.025) compared to the general-purpose encoders (ECE: 0.044–0.082). Conclusions: While domain-specific medical pretraining (MedSigLIP) did not yield a statistically significant advantage in aggregate classification accuracy compared to the best ImageNet-based model (EfficientNet), it provided superior calibration and sensitivity for borderline cases. These findings suggest that in thyroid cytology clinical workflow support, encoder selection should be guided by a joint view of discrimination and safety—particularly calibration and Bethesda V sensitivity—rather than aggregate accuracy alone, enabling threshold-based triage and selective expert review. In particular, well-calibrated models such as MedSigLIP suggest a potential benefit in reducing overconfident misclassification in borderline Bethesda V cases, pending prospective validation in real-world triage workflows.

23 June 2026 in Archives of Osteoporosis
Verified
Journal Article
Open access
RERıdvan Erten
Rıdvan Erten
MPMehmet Poyrazer
Mehmet Poyrazer
Summary: In 6364 US adults aged ≥ 50 with up to 15 years of follow-up, lumbar spine–total hip T-score discordance was associated with all-cause mortality in women but not in men. Total hip T-score was the strongest single-site predictor in both sexes. Purpose: Lumbar spine bone mineral density may be influenced by degenerative and non-skeletal mineralization in older adults. Whether the discrepancy between lumbar spine and hip T-scores carries prognostic information for mortality remains unclear. We examined the sex-specific association of the lumbar spine–total hip T-score difference (ΔT_LS − TH) with all-cause mortality, alongside site-specific T-scores at the lumbar spine, femoral neck, and total hip. Methods: Prospective cohort study of 6364 adults aged ≥ 50 years from five NHANES cycles (2005–2018) linked to the National Death Index through December 31, 2019. Survey-weighted Cox models estimated hazard ratios (HR) for ΔT (LS T-score minus TH T-score), adjusting for age, sex, BMI, eGFR, smoking, and albumin. Sex-specific associations and a sex × ΔT interaction were tested. Candidate joint models were compared to evaluate whether ΔT_LS − TH provided information beyond total hip T-score. Sensitivity analyses included landmark exclusion of early deaths and additional adjustment for baseline comorbidities . Results: Over a median follow-up of 9.3 years, 1155 deaths occurred. A significant sex × ΔT_LS − TH interaction was observed ( p = 0.001). In women, ΔT_LS − TH was associated with higher all-cause mortality (HR 1.259 per unit; 95% CI 1.146–1.382; p < 0.001), and comparative models favored ΔT_LS − TH over the conventional lumbar spine–femoral neck difference. In men, the corresponding estimate was smaller and did not meet the conventional threshold for statistical significance (HR 1.084 per unit; 95% CI 0.9998–1.175; exact p = 0.0504). Total hip T-score was the strongest single-site predictor in both sexes (women HR 0.751, p < 0.001; men HR 0.854, p = 0.002). Lumbar spine T-score showed no association with mortality in women after multivariable adjustment. Parallel analyses using raw BMD values supported the same site-specific ranking, with total hip BMD showing the strongest single-site association. Conclusion: In older US adults, the lumbar spine–total hip T-score difference was associated with all-cause mortality in women but not in men, while total hip T-score remained the strongest single-site predictor in both sexes. These findings are consistent with the clinical emphasis on hip measurements when interpreting DXA results in older adults. ΔT_LS − TH may identify a higher-risk pattern of site-specific discordance at the population level, but it should not be used for individual mortality prediction.

9 October 2026 in Journal of Clinical Medicine
Verified
Journal Article
Open access
PDF file
RERıdvan Erten
Rıdvan Erten
KEKübra Erdoğan
Kübra Erdoğan
MPMehmet Poyrazer
Mehmet Poyrazer
ÖÇÖZGÜR ÇAKIR
ÖZGÜR ÇAKIR
RDRana Tuna Doğrul
Rana Tuna Doğrul
KSKamile Silay
Kamile Silay
Background/Objectives: We examined correlates of lumbar–femoral T-score differences (ΔT), retaining the triglyceride–glucose (TyG) index as the prespecified primary exposure. Methods: This retrospective cross-sectional analysis of routine records included 528 geriatric outpatients assessed in Ankara, Türkiye, in 2023; primary regression used 506 complete cases. ΔT was lumbar minus femoral-neck T-score. Device L1–L4 composites were used without vertebra-level exclusions because scan images were unavailable. Firth models and an exploratory post hoc arithmetic decomposition were secondary. Results: Male sex was associated with higher ΔT (B = +0.984; p < 0.001), uniquely explaining 9.87% of variance versus 0.88% for TyG. TyG estimates (95% CI) were +0.243 (+0.033 to +0.453; p = 0.023) in the primary analysis, +0.214 (−0.010 to +0.438; p = 0.061) retaining two records excluded by plausibility limits, and +0.070 (−0.115 to +0.255; p = 0.458) excluding influential observations. The male lumbar-worse OR was 0.12 (95% CI 0.03–0.31), based on 3/165 male events. Among patients with femoral-neck osteoporosis, lumbar T-scores exceeded −2.5 in 22/30 men versus 43/89 women (post hoc, unadjusted). Conclusions: Sex explained more unique ΔT variance than TyG. The TyG association was small and sensitive to analytic decisions. Unmeasured degeneration and selection bias limit interpretation.

27 December 2025 in JMIR Preprints
Verified
Preprint
Open access
MPMehmet Poyrazer
Mehmet Poyrazer
AMAyşe Münevver Mühürdaroğlu
Ayşe Münevver Mühürdaroğlu
MÇMahmut Emin Çelik
Mahmut Emin Çelik
PGPüren Gökbulut
Püren Gökbulut
SFSevde Nur Fırat
Sevde Nur Fırat
BACKGROUND Hypothyroidism affects approximately 5% of the global population. While YouTube has emerged as a dominant health information platform, the specific network dynamics and information flow regarding chronic thyroid disease remain largely unexplored in digital health research. OBJECTIVE The objective of this study was to conduct a comprehensive infodemiological analysis of hypothyroidism content on YouTube, evaluating temporal trends, network structures, and audience sentiment across different creator categories. METHODS We identified 331 hypothyroidism-related YouTube videos (2015-2025). For creator-type comparisons, we excluded 18 videos from channels with unknown creator classification, yielding a filtered dataset of 313 videos (medical n=251; influencer n=62). Descriptive analyses used the full dataset (N=331), whereas inferential comparisons used the filtered dataset (n=313).The methodology integrated engagement analytics, The Valence Aware Dictionary and sEntiment Reasoner(VADER) sentiment analysis (validated via manual coding, κ=0.87), and social network analysis using NetworkX to construct bipartite graphs of the information ecosystem. Statistical comparisons were performed using Mann-Whitney U tests and Cohen d effect sizes. RESULTS Medical channels dominated content production (80.2%). Despite numerical differences in raw metrics, medical and influencer content demonstrated equivalent engagement rates (2.24% vs 2.36%; P=.31; d=-0.091), suggesting that engagement quality is independent of creator credentials. The COVID-19 pandemic catalyzed a 234% increase in content volume, resulting in a sustained elevation of post-pandemic engagement (2.47% vs 1.70% pre-pandemic). Network analysis revealed distinct structural models: medical networks followed a distributed expertise model (density=0.0151), whereas influencer networks utilized hub-centric approaches (density=0.0368). Sentiment analyses included 40,260 comments in the full dataset; creator-type stratified analyses included 38,495 comments after excluding unknown channels. Overall sentiment was predominantly positive (47.3%), though mean sentiment scores significantly declined from pre-pandemic optimism (0.270) to post-pandemic skepticism (0.183). Key information hubs identified included Dr. Westin Childs (medical centrality=0.189) and Fit Tuber (8.1M subscribers). CONCLUSIONS This study demonstrates that engagement quality in thyroid health communication transcends traditional creator credentials. The pandemic has permanently transformed digital health consumption patterns, establishing YouTube as a resilient resource for chronic disease information while challenging prevailing assumptions regarding the dominance of health misinformation.

30 November 2022 in Balıkesir Medical Journal
Verified
Journal Article
Open access
PDF file
MPMehmet Poyrazer
Mehmet Poyrazer
FEFatma Kaplan Efe
Fatma Kaplan Efe
BYBusragul Yilmaz
Busragul Yilmaz
RERıdvan Erten
Rıdvan Erten
GGGül Gürsoy
Gül Gürsoy
AMAÇ: Prediyabet (PDM) diyabetin kendisi gibi enflamatuar bir süreçtir. Bu çalışmada PDM hastalarında endojen nonenzimatik antioksidan olan total bilirubin ile kardiak iskemi ve enflamasyon markırı olan iskemi modifiye albümin arasındaki ilişkinin değerlendirilmesi amaçlandı. MATERYAL-METOD: Çalışmaya kronik hastalıkları olmayan sigara ve alkol kullanmayan ve rutin biyokimyasal analizleri için başvuran 100 olgu dahil edildi. 50 olgu yeni tanı PDM, 50 olgu kontrol grubu olarak belirlendi. Olguların açlık kan örnekleri alındı. 4000 RPM de santrifüj edildi ve plazma kısmı -80 ° C’de saklandı. IMA absorbansı ise Bar-Or’un spektrofotometrik metoduyla analiz edildi. Çalışmamızda IMA düzeyleri bireylerin albümin sonuçlarına göre düzeltildi. BULGULAR: İskemi modifiye albümin (IMA) ve albümine göre düzeltilmiş İskemi modifiye albümin (D-IMA) PDM hastalarında kontrol grubuna göre istatiksel anlamlı olarak yüksekti Total bilirubin ve IMA ve D-IMA arasında korelasyon saptanmadı. IMA ve D-IMA albümin dışında diğer metabolik parametreler ile de korelasyon saptanmadı. Albümin ile D-IMA ve IMA arasında negatif korelasyon saptandı. SONUÇ: IMA ve D-IMA düzeylerinin prediyabetik grupta daha yüksek olması prediyabetik hastaların hem aterosklerotik risk faktörleri hem de oksidatif stres açısından kontrol grubuna göre daha fazla risk altında oldugunu düsündürmektedir.