Emek Güldoğan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Emek Güldoğan has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Cemil Colak
Department of Biostatistics and Medical Informatics, Faculty of Medicine, Inonu University, 44280 Malatya, Turkey
10 shared publications - Hasan Ucuzal
Department of Biostatistics and Medical Informatics, Faculty of Medicine, Inonu University, 44280 Malatya, Turkey
6 shared publications - Fatma Hilal Yagin
Department of Biostatistics and Medical Informatics, Faculty of Medicine, Balikesir University, Balikesir 10145, Türkiye
5 shared publications - Ramazan Paşahan
İnönü Üniversitesi
5 shared publications - Bora Tetik
İNÖNÜ ÜNİVERSİTESİ, TIP FAKÜLTESİ
3 shared publications - Abdulmohsen Algarni
Department of Computer Science, King Khalid University, Abha 61421, Saudi Arabia
2 shared publications
Latest Publications
- Open Access
- 4 November 2026
- Published by MDPI AG in Diagnostics
- Vol. 16 (21) , 3640
Development of Hybrid Deep Learning Systems Combining Convolutional Neural Networks and Vision Transformers for Medical Ultrasound Image Classification
Background: Accurate classification of thyroid nodule ultrasound images is clinically significant. This study evaluated whether hybrid architectures combining the local feature extraction of convolutional neural networks (CNNs) with the global context modelling of vision transformers (ViTs) outperform each category in isolation. Methods: Two public datasets (6905 images; 3331 benign, 3574 malignant) were used. Eleven ImageNet-pretrained architectures were compared: CNN-based (ResNet-50, EfficientNet-B4, ConvNeXt), ViT-based (Swin Transformer, DeiT, PVTv2, XCiT) and hybrid (ConViT, CoaT, MobileViT, HuviT). Backbones were frozen and only the classification head was trained; HuviT’s new projection and encoder remained frozen at random initialisation. Wilson and Newcombe 95% confidence intervals (CIs) were computed; explainability used gradient-weighted class activation mapping (Grad-CAM). Results: ConViT achieved the highest accuracy (83.30%; 95% CI 80.91–85.45) and an area under the receiver operating characteristic curve (ROC-AUC) of 89.94% (95% CI 87.8–91.7). ConvNeXt ranked second (82.63%) but had the highest ROC-AUC (90.02%; 95% CI 87.9–91.8); the intervals overlapped. HuviT reached 76.35%. Grad-CAM for a representative case highlighted lesion areas. Conclusions: While ConViT achieved the highest performance on threshold-based metrics, these findings suggest the potential of specific hybrid designs rather than establishing general superiority of hybrid architectures. Given the retrospective design and the partial association between image source and class label, prospective external validation is warranted.
- 4 November 2026
- Published by MDPI AG in Diagnostics
- Vol. 16 (21) , 3640
Open AccessDevelopment of Hybrid Deep Learning Systems Combining Convolutional Neural Networks and Vision Transformers for Medical Ultrasound Image Classification
Background: Accurate classification of thyroid nodule ultrasound images is clinically significant. This study evaluated whether hybrid architectures combining the local feature extraction of convolutional neural networks (CNNs) with the global context modelling of vision transformers (ViTs) outperform each category in isolation. Methods: Two public datasets (6905 images; 3331 benign, 3574 malignant) were used. Eleven ImageNet-pretrained architectures were compared: CNN-based (ResNet-50, EfficientNet-B4, ConvNeXt), ViT-based (Swin Transformer, DeiT, PVTv2, XCiT) and hybrid (ConViT, CoaT, MobileViT, HuviT). Backbones were frozen and only the classification head was trained; HuviT’s new projection and encoder remained frozen at random initialisation. Wilson and Newcombe 95% confidence intervals (CIs) were computed; explainability used gradient-weighted class activation mapping (Grad-CAM). Results: ConViT achieved the highest accuracy (83.30%; 95% CI 80.91–85.45) and an area under the receiver operating characteristic curve (ROC-AUC) of 89.94% (95% CI 87.8–91.7). ConvNeXt ranked second (82.63%) but had the highest ROC-AUC (90.02%; 95% CI 87.9–91.8); the intervals overlapped. HuviT reached 76.35%. Grad-CAM for a representative case highlighted lesion areas. Conclusions: While ConViT achieved the highest performance on threshold-based metrics, these findings suggest the potential of specific hybrid designs rather than establishing general superiority of hybrid architectures. Given the retrospective design and the partial association between image source and class label, prospective external validation is warranted.
- 27 October 2026
- Published by Konuralp Medical Journal in Konuralp Tip Dergisi
- Vol. 18 (3) , 317-331
Comparison of Machine Learning and Deep Learning-Based Survival Models: An Explainable Artificial Intelligence Approach
Aim: To compare the predictive performance of ten survival models spanning statistical, tree-based, ensemble, kernel, and deep learning approaches on colon cancer survival data, and to interpret their predictions with explainable artificial intelligence methods.Material and Methods: The colon dataset from the R survival package was used (n=888; events=430; event rate 48.4%). Predictive performance was assessed by 5×3 nested cross-validation using Harrell's concordance index (C-index) and, for models with a native survival function, the integrated Brier score (IBS); time-dependent discrimination was examined with the cumulative/dynamic AUC. The stability of the point estimates was additionally summarised with 1000-iteration bootstrap 95% confidence intervals, which reflect apparent (within-sample) performance rather than generalisation. Explainability was addressed at three layers using SHAP, SurvSHAP(t), and SurvLIME.Results: In nested cross-validation the highest mean C-index values were very close across RSF (0.6600), XGBoost (0.6596), and DeepSurv (0.6578); RSF showed the most balanced performance, with the highest mean C-index, the lowest IBS (0.1740) and the highest time-dependent AUC (TD-AUC=0.7204). In the within-sample bootstrap ExtraSurvTrees and RSF attained the highest apparent C-index (0.7342 and 0.7307); however, these values are optimistically biased relative to the nested cross-validation estimates. SHAP identified the number of positive lymph nodes as the strongest prognostic factor, SurvSHAP(t) revealed time-dependent variation in feature importance, and SurvLIME produced patient-level interpretable risk profiles.Conclusion: Ensemble methods demonstrated consistent and stable cross-validated performance, whereas deep learning models, although competitive on average, carried a risk of instability on this small-to-medium dataset. The explainability analyses confirmed clinically interpretable prognostic patterns and may inform the future development of clinical decision-support tools, pending external validation.
- 27 October 2026
- Published by Konuralp Medical Journal in Konuralp Tip Dergisi
- Vol. 18 (3) , 317-331
Comparison of Machine Learning and Deep Learning-Based Survival Models: An Explainable Artificial Intelligence Approach
Aim: To compare the predictive performance of ten survival models spanning statistical, tree-based, ensemble, kernel, and deep learning approaches on colon cancer survival data, and to interpret their predictions with explainable artificial intelligence methods.Material and Methods: The colon dataset from the R survival package was used (n=888; events=430; event rate 48.4%). Predictive performance was assessed by 5×3 nested cross-validation using Harrell's concordance index (C-index) and, for models with a native survival function, the integrated Brier score (IBS); time-dependent discrimination was examined with the cumulative/dynamic AUC. The stability of the point estimates was additionally summarised with 1000-iteration bootstrap 95% confidence intervals, which reflect apparent (within-sample) performance rather than generalisation. Explainability was addressed at three layers using SHAP, SurvSHAP(t), and SurvLIME.Results: In nested cross-validation the highest mean C-index values were very close across RSF (0.6600), XGBoost (0.6596), and DeepSurv (0.6578); RSF showed the most balanced performance, with the highest mean C-index, the lowest IBS (0.1740) and the highest time-dependent AUC (TD-AUC=0.7204). In the within-sample bootstrap ExtraSurvTrees and RSF attained the highest apparent C-index (0.7342 and 0.7307); however, these values are optimistically biased relative to the nested cross-validation estimates. SHAP identified the number of positive lymph nodes as the strongest prognostic factor, SurvSHAP(t) revealed time-dependent variation in feature importance, and SurvLIME produced patient-level interpretable risk profiles.Conclusion: Ensemble methods demonstrated consistent and stable cross-validated performance, whereas deep learning models, although competitive on average, carried a risk of instability on this small-to-medium dataset. The explainability analyses confirmed clinically interpretable prognostic patterns and may inform the future development of clinical decision-support tools, pending external validation.
- Open Access
- 21 August 2026
- Published by MDPI AG in Bioengineering
- Vol. 13 (8) , 951
A Leakage-Free Survival-Modelling Benchmark for Hepatocellular Carcinoma Recurrence After Liver Transplantation: Nested Cross-Validation Against the Milan Criteria
Background: Predicting recurrence after liver transplantation (LT) for hepatocellular carcinoma (HCC) remains important for post-transplant risk stratification and surveillance planning. The Milan criteria discriminate only moderately and some machine-learning re-analyses report overly optimistic results because of information leakage. Aim: The current study aimed to re-evaluate a previously published transplant cohort under a leakage-free survival-analysis framework and to benchmark post-transplant, explant-informed survival learners against the Milan criteria as a fixed pre-transplant reference. We hypothesised moderate rather than near-perfect discrimination, similar performance across learners of differing complexity, and better discrimination than the Milan criteria. Methods: This secondary analysis included 356 patients with HCC who underwent LT. The primary endpoint was recurrence-free survival, analysed from the observed event indicator and follow-up time rather than from a derived risk label. Seven survival learners were benchmarked with repeated nested cross-validation, using three repeats of a five-fold outer loop with a three-fold inner tuning loop. All data-dependent preprocessing, including robust multivariable outlier handling and imputation, was fitted within training folds only. Performance was assessed by the concordance indices of Harrell and Uno, the time-dependent area under the curve, the integrated Brier score, calibration, decision-curve analysis and descriptive competing-risk assessment. Results: Recurrence developed in 183 of the 356 patients over a median follow-up of 52 months. Discrimination was moderate rather than near-perfect and similar across learners; the random survival forest ranked highest and the Elastic-Net Cox model performed comparably. All learners showed higher descriptive concordance than the Milan criteria, and dependency-corrected comparisons supported higher concordance for the full-feature Cox model than for the Milan criteria, whereas the random survival forest and Cox did not differ materially. Out-of-fold calibration of the Elastic-Net Cox model at 36 months was acceptable, decision-curve analysis indicated positive net benefit across clinically relevant thresholds, and tumour size and alpha-fetoprotein were the leading contributors to prediction. Findings were stable in ablation and threshold-sensitivity analyses. Conclusions: Leakage-free survival modelling gave moderate but internally validated prediction of post-transplant recurrence and higher concordance than the Milan criteria in this cohort, supporting the stated hypotheses. Careful study design may matter more than architectural complexity in this setting, and leakage-free survival analysis is a practical standard for prognostic modelling in transplant oncology.
- 21 August 2026
- Published by MDPI AG in Bioengineering
- Vol. 13 (8) , 951
Open AccessA Leakage-Free Survival-Modelling Benchmark for Hepatocellular Carcinoma Recurrence After Liver Transplantation: Nested Cross-Validation Against the Milan Criteria
Background: Predicting recurrence after liver transplantation (LT) for hepatocellular carcinoma (HCC) remains important for post-transplant risk stratification and surveillance planning. The Milan criteria discriminate only moderately and some machine-learning re-analyses report overly optimistic results because of information leakage. Aim: The current study aimed to re-evaluate a previously published transplant cohort under a leakage-free survival-analysis framework and to benchmark post-transplant, explant-informed survival learners against the Milan criteria as a fixed pre-transplant reference. We hypothesised moderate rather than near-perfect discrimination, similar performance across learners of differing complexity, and better discrimination than the Milan criteria. Methods: This secondary analysis included 356 patients with HCC who underwent LT. The primary endpoint was recurrence-free survival, analysed from the observed event indicator and follow-up time rather than from a derived risk label. Seven survival learners were benchmarked with repeated nested cross-validation, using three repeats of a five-fold outer loop with a three-fold inner tuning loop. All data-dependent preprocessing, including robust multivariable outlier handling and imputation, was fitted within training folds only. Performance was assessed by the concordance indices of Harrell and Uno, the time-dependent area under the curve, the integrated Brier score, calibration, decision-curve analysis and descriptive competing-risk assessment. Results: Recurrence developed in 183 of the 356 patients over a median follow-up of 52 months. Discrimination was moderate rather than near-perfect and similar across learners; the random survival forest ranked highest and the Elastic-Net Cox model performed comparably. All learners showed higher descriptive concordance than the Milan criteria, and dependency-corrected comparisons supported higher concordance for the full-feature Cox model than for the Milan criteria, whereas the random survival forest and Cox did not differ materially. Out-of-fold calibration of the Elastic-Net Cox model at 36 months was acceptable, decision-curve analysis indicated positive net benefit across clinically relevant thresholds, and tumour size and alpha-fetoprotein were the leading contributors to prediction. Findings were stable in ablation and threshold-sensitivity analyses. Conclusions: Leakage-free survival modelling gave moderate but internally validated prediction of post-transplant recurrence and higher concordance than the Milan criteria in this cohort, supporting the stated hypotheses. Careful study design may matter more than architectural complexity in this setting, and leakage-free survival analysis is a practical standard for prognostic modelling in transplant oncology.
- Open Access
- 27 May 2026
- Published by MDPI AG in Diagnostics
- Vol. 16 (11) , 1643
Explainable Boosting Machine in Sepsis Prediction Using Platelet Metabolomics: An Interpretable Machine Learning Approach
Background: Sepsis remains a leading cause of mortality in emergency and intensive care settings, with early diagnosis representing a critical determinant of patient outcomes. Despite advances in biomarker discovery, integrating platelet-derived metabolic signatures with explainable machine learning frameworks for sepsis prediction remains underexplored. The clinical adoption of predictive models has been hindered by the “black box” nature of conventional algorithms, limiting clinician trust and understanding. Objective: This study aimed to evaluate and validate an interpretable machine learning model utilizing platelet metabolomics data for accurate sepsis prediction while providing clinically meaningful explanations of the underlying metabolic disturbances that could inform therapeutic decision-making. Methods: We analyzed metabolomics data, comprising 25 sepsis patients diagnosed according to Sepsis-3 criteria and 14 age- and gender-matched non-sepsis from the emergency department. Platelet metabolite profiles were obtained via quantitative 1H-NMR spectroscopy. Five machine learning algorithms were evaluated: Explainable Boosting Machine (EBM), Support Vector Machine (SVM), Logistic Regression (LR), Gradient Boosting Machine (GBM), and AdaBoost. Three biologically motivated metabolite ratios (adenosine triphosphate/adenosine diphosphate (ATP/ADP), ATP/adenosine monophosphate (AMP), Glutamine/Glutamate) were derived as additional features, yielding 22 candidate variables. Models were evaluated using a fully nested leave-one-out cross-validation (LOOCV) framework in which log transformation, KNN imputation, BorderlineSMOTE class balancing, and hyperparameter optimisation were performed exclusively within each training fold. Global and local interpretability analyses were performed to identify discriminative metabolites. Results: EBM achieved the highest ROC-AUC (0.864; 95% CI: 0.736–1.000), the highest PR-AUC (0.902; 95% CI: 0.783–0.997), and the best Brier score (0.189; 95% CI: 0.130–0.258) among all evaluated models, with sensitivity 0.880 (95% CI: 0.640–1.000; TP = 22/25) and specificity 0.714 (95% CI: 0.357–1.000; TN = 10/14). Global feature importance identified Carnitine, myo-Inositol, ADP, and O-Phosphoethanolamine as the leading single-feature predictors, alongside three pairwise interaction terms reflecting non-additive energy–amino acid metabolic relationships. Local explanations demonstrated that the ADP–Creatine interaction, Glutamine, and myo-Inositol drove correct sepsis classification in a representative true positive case. Conclusions: The EBM model demonstrated the highest discriminative performance and best calibration among all evaluated models, providing transparent mechanistic insights through global feature importance, and patient-level local explanations. These findings position the proposed framework as a proof-of-concept warranting external validation in larger, multi-centre cohorts before any clinical application is considered.
- 27 May 2026
- Published by MDPI AG in Diagnostics
- Vol. 16 (11) , 1643
Open AccessExplainable Boosting Machine in Sepsis Prediction Using Platelet Metabolomics: An Interpretable Machine Learning Approach
Background: Sepsis remains a leading cause of mortality in emergency and intensive care settings, with early diagnosis representing a critical determinant of patient outcomes. Despite advances in biomarker discovery, integrating platelet-derived metabolic signatures with explainable machine learning frameworks for sepsis prediction remains underexplored. The clinical adoption of predictive models has been hindered by the “black box” nature of conventional algorithms, limiting clinician trust and understanding. Objective: This study aimed to evaluate and validate an interpretable machine learning model utilizing platelet metabolomics data for accurate sepsis prediction while providing clinically meaningful explanations of the underlying metabolic disturbances that could inform therapeutic decision-making. Methods: We analyzed metabolomics data, comprising 25 sepsis patients diagnosed according to Sepsis-3 criteria and 14 age- and gender-matched non-sepsis from the emergency department. Platelet metabolite profiles were obtained via quantitative 1H-NMR spectroscopy. Five machine learning algorithms were evaluated: Explainable Boosting Machine (EBM), Support Vector Machine (SVM), Logistic Regression (LR), Gradient Boosting Machine (GBM), and AdaBoost. Three biologically motivated metabolite ratios (adenosine triphosphate/adenosine diphosphate (ATP/ADP), ATP/adenosine monophosphate (AMP), Glutamine/Glutamate) were derived as additional features, yielding 22 candidate variables. Models were evaluated using a fully nested leave-one-out cross-validation (LOOCV) framework in which log transformation, KNN imputation, BorderlineSMOTE class balancing, and hyperparameter optimisation were performed exclusively within each training fold. Global and local interpretability analyses were performed to identify discriminative metabolites. Results: EBM achieved the highest ROC-AUC (0.864; 95% CI: 0.736–1.000), the highest PR-AUC (0.902; 95% CI: 0.783–0.997), and the best Brier score (0.189; 95% CI: 0.130–0.258) among all evaluated models, with sensitivity 0.880 (95% CI: 0.640–1.000; TP = 22/25) and specificity 0.714 (95% CI: 0.357–1.000; TN = 10/14). Global feature importance identified Carnitine, myo-Inositol, ADP, and O-Phosphoethanolamine as the leading single-feature predictors, alongside three pairwise interaction terms reflecting non-additive energy–amino acid metabolic relationships. Local explanations demonstrated that the ADP–Creatine interaction, Glutamine, and myo-Inositol drove correct sepsis classification in a representative true positive case. Conclusions: The EBM model demonstrated the highest discriminative performance and best calibration among all evaluated models, providing transparent mechanistic insights through global feature importance, and patient-level local explanations. These findings position the proposed framework as a proof-of-concept warranting external validation in larger, multi-centre cohorts before any clinical application is considered.
- Open Access
Explainable Meta-Learning Ensemble Framework for Predicting Insulin Dose Adjustments in Diabetic Patients: A Comparative Machine Learning Approach with SHAP-Based Clinical Interpretability
Background and Objectives: Diabetes mellitus represents one of the most prevalent chronic metabolic disorders worldwide, necessitating precise insulin dose management to prevent both acute and long-term complications. The optimization of insulin dosing remains a significant clinical challenge, as inappropriate dosing can lead to hypoglycemia or hyperglycemia, each carrying substantial morbidity risks. Machine learning approaches have emerged as promising tools for developing clinical decision support systems; however, their practical implementation requires both high predictive accuracy and model interpretability. This study aimed to develop and evaluate an explainable machine learning framework for predicting insulin dose adjustments in diabetic patients. We sought to compare multiple ensemble learning approaches and identify the optimal model configuration that balances predictive performance with clinical interpretability through comprehensive SHAP and LIME analyses. Materials and Methods: A comprehensive dataset comprising 10,000 patient records with 12 clinical and demographic features was utilized. We implemented and compared nine machine learning models, including gradient boosting variants (XGBoost, LightGBM, CatBoost, GradientBoosting), AdaBoost, and four ensemble strategies (Voting, Stacking, Blending, and Meta-Learning). Model interpretability was achieved through SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) analyses. Performance was evaluated using accuracy, weighted F1-score, area under the receiver operating characteristic curve (AUC-ROC), precision-recall AUC (PR-AUC), sensitivity, specificity, and cross-entropy loss. Results: The Meta-Learning Ensemble achieved superior performance across all evaluation metrics, attaining an accuracy of 81.35%, weighted F1-score of 0.8121, macro-averaged AUC-ROC of 0.9637, and PR-AUC of 0.9317. The model demonstrated exceptional sensitivity (86.61%) and specificity (91.79%), with particularly high performance in detecting dose reduction requirements (100% sensitivity for the ‘down’ class). SHAP analysis revealed insulin sensitivity, previous medications, sleep hours, weight, and body mass index as the most influential predictors across different insulin adjustment categories. The meta-model feature importance analysis indicated that LightGBM probability estimates contributed most significantly to the ensemble predictions. Conclusions: The proposed explainable Meta-Learning Ensemble framework demonstrates robust predictive capability for insulin dose adjustment recommendations while maintaining clinical interpretability. The integration of SHAP-based explanations facilitates clinician understanding of model predictions, supporting transparent and informed decision-making in diabetes management. This approach represents a significant advancement toward the clinical implementation of artificial intelligence in personalized insulin therapy.
Open AccessExplainable Meta-Learning Ensemble Framework for Predicting Insulin Dose Adjustments in Diabetic Patients: A Comparative Machine Learning Approach with SHAP-Based Clinical Interpretability
Background and Objectives: Diabetes mellitus represents one of the most prevalent chronic metabolic disorders worldwide, necessitating precise insulin dose management to prevent both acute and long-term complications. The optimization of insulin dosing remains a significant clinical challenge, as inappropriate dosing can lead to hypoglycemia or hyperglycemia, each carrying substantial morbidity risks. Machine learning approaches have emerged as promising tools for developing clinical decision support systems; however, their practical implementation requires both high predictive accuracy and model interpretability. This study aimed to develop and evaluate an explainable machine learning framework for predicting insulin dose adjustments in diabetic patients. We sought to compare multiple ensemble learning approaches and identify the optimal model configuration that balances predictive performance with clinical interpretability through comprehensive SHAP and LIME analyses. Materials and Methods: A comprehensive dataset comprising 10,000 patient records with 12 clinical and demographic features was utilized. We implemented and compared nine machine learning models, including gradient boosting variants (XGBoost, LightGBM, CatBoost, GradientBoosting), AdaBoost, and four ensemble strategies (Voting, Stacking, Blending, and Meta-Learning). Model interpretability was achieved through SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) analyses. Performance was evaluated using accuracy, weighted F1-score, area under the receiver operating characteristic curve (AUC-ROC), precision-recall AUC (PR-AUC), sensitivity, specificity, and cross-entropy loss. Results: The Meta-Learning Ensemble achieved superior performance across all evaluation metrics, attaining an accuracy of 81.35%, weighted F1-score of 0.8121, macro-averaged AUC-ROC of 0.9637, and PR-AUC of 0.9317. The model demonstrated exceptional sensitivity (86.61%) and specificity (91.79%), with particularly high performance in detecting dose reduction requirements (100% sensitivity for the ‘down’ class). SHAP analysis revealed insulin sensitivity, previous medications, sleep hours, weight, and body mass index as the most influential predictors across different insulin adjustment categories. The meta-model feature importance analysis indicated that LightGBM probability estimates contributed most significantly to the ensemble predictions. Conclusions: The proposed explainable Meta-Learning Ensemble framework demonstrates robust predictive capability for insulin dose adjustment recommendations while maintaining clinical interpretability. The integration of SHAP-based explanations facilitates clinician understanding of model predictions, supporting transparent and informed decision-making in diabetes management. This approach represents a significant advancement toward the clinical implementation of artificial intelligence in personalized insulin therapy.
- 30 August 2025
- Published by Adiyaman University in Adıyaman Üniversitesi Sağlık Bilimleri Dergisi
- Vol. 11 (2) , 191-204
Transforming healthcare and education: The impact of deep learning in medical informatics and training
Artificial intelligence (AI) and deep learning have quickly developed for image-based diagnostics, disease prediction and clinical decision support. Medical informatics/education in medicine have been huge changes caused by these technologies. To analyze such a complex biomedical data, deep learning architectures are employed. This narrative review discusses how deep learning has transformed medical education and informatics, as it is used in medical imaging, genomic analysis, predictive analytics, and smart monitoring systems. In medical education, deep learning is employed for interactive simulations, automated grading, and individual learning; in informatics, for diagnosis accuracy, decision-making, as well as patient care. The future directions show that explainable AI (XAI), federated learning and ethical frameworks are required to be integrated for enhancing transparency and reliability. Such changes are leading us into a revolution in healthcare and medical education with deep learning, where the efficiency, personalization and evidence-based are improved.
- 30 August 2025
- Published by Adiyaman University in Adıyaman Üniversitesi Sağlık Bilimleri Dergisi
- Vol. 11 (2) , 191-204
Transforming healthcare and education: The impact of deep learning in medical informatics and training
Artificial intelligence (AI) and deep learning have quickly developed for image-based diagnostics, disease prediction and clinical decision support. Medical informatics/education in medicine have been huge changes caused by these technologies. To analyze such a complex biomedical data, deep learning architectures are employed. This narrative review discusses how deep learning has transformed medical education and informatics, as it is used in medical imaging, genomic analysis, predictive analytics, and smart monitoring systems. In medical education, deep learning is employed for interactive simulations, automated grading, and individual learning; in informatics, for diagnosis accuracy, decision-making, as well as patient care. The future directions show that explainable AI (XAI), federated learning and ethical frameworks are required to be integrated for enhancing transparency and reliability. Such changes are leading us into a revolution in healthcare and medical education with deep learning, where the efficiency, personalization and evidence-based are improved.
- Open Access
Interpretable Machine Learning for Serum-Based Metabolomics in Breast Cancer Diagnostics: Insights from Multi-Objective Feature Selection-Driven LightGBM-SHAP Models
Background and Objectives: Breast cancer accounts for 12.5% of all new cancer cases in women worldwide. Early detection significantly improves survival rates, but traditional biomarkers like CA 15-3 and HER2 lack sensitivity and specificity, particularly for early-stage disease. Advances in metabolomics and machine learning, particularly explainable artificial intelligence (XAI), offer new opportunities for identifying robust biomarkers and improving diagnostic accuracy. This study aimed to identify and validate serum-based metabolic biomarkers for breast cancer using advanced metabolomic profiling techniques and a Light Gradient Boosting Machine (LightGBM) model. Additionally, SHapley Additive exPlanations (SHAP) were applied to enhance model interpretability and biological insight. Materials and Methods: The study included 103 breast cancer patients and 31 healthy controls. Serum samples underwent liquid and gas chromatography–time-of-flight mass spectrometry (LC-TOFMS and GC-TOFMS). Mutual Information (MI), Sparse Partial Least Squares (sPLS), Boruta, and Multi-Objective Feature Selection (MOFS) approaches were applied to the data for biomarker discovery. LightGBM, AdaBoost, and Random Forest were employed for classification and to identify class imbalance with the Synthetic Minority Oversampling Technique (SMOTE). SHAP analysis ranked metabolites based on their contribution to model predictions. Results: Compared to other feature selection approaches, the MOFS approach was more robust in terms of predictive performance, and metabolites identified by this method were used in subsequent analyses for biomarker discovery. LightGBM outperformed the AdaBoost and Random Forest models, achieving 86.6% accuracy, 89.1% sensitivity, 84.2% specificity, and an F1-score of 87.0%. SHAP analysis identified 2-Aminobutyric acid, choline, and coproporphyrin as the most influential metabolites, with dysregulation of these markers associated with breast cancer risk. Conclusions: This study is among the first to integrate SHAP explainability with metabolomic profiling, bridging computational predictions and biological insights for improved clinical adoption. This study demonstrates the effectiveness of combining metabolomics with XAI-driven machine learning for breast cancer diagnostics. The identified biomarkers not only improve diagnostic accuracy but also reveal critical metabolic dysregulations associated with disease progression.
Open AccessInterpretable Machine Learning for Serum-Based Metabolomics in Breast Cancer Diagnostics: Insights from Multi-Objective Feature Selection-Driven LightGBM-SHAP Models
Background and Objectives: Breast cancer accounts for 12.5% of all new cancer cases in women worldwide. Early detection significantly improves survival rates, but traditional biomarkers like CA 15-3 and HER2 lack sensitivity and specificity, particularly for early-stage disease. Advances in metabolomics and machine learning, particularly explainable artificial intelligence (XAI), offer new opportunities for identifying robust biomarkers and improving diagnostic accuracy. This study aimed to identify and validate serum-based metabolic biomarkers for breast cancer using advanced metabolomic profiling techniques and a Light Gradient Boosting Machine (LightGBM) model. Additionally, SHapley Additive exPlanations (SHAP) were applied to enhance model interpretability and biological insight. Materials and Methods: The study included 103 breast cancer patients and 31 healthy controls. Serum samples underwent liquid and gas chromatography–time-of-flight mass spectrometry (LC-TOFMS and GC-TOFMS). Mutual Information (MI), Sparse Partial Least Squares (sPLS), Boruta, and Multi-Objective Feature Selection (MOFS) approaches were applied to the data for biomarker discovery. LightGBM, AdaBoost, and Random Forest were employed for classification and to identify class imbalance with the Synthetic Minority Oversampling Technique (SMOTE). SHAP analysis ranked metabolites based on their contribution to model predictions. Results: Compared to other feature selection approaches, the MOFS approach was more robust in terms of predictive performance, and metabolites identified by this method were used in subsequent analyses for biomarker discovery. LightGBM outperformed the AdaBoost and Random Forest models, achieving 86.6% accuracy, 89.1% sensitivity, 84.2% specificity, and an F1-score of 87.0%. SHAP analysis identified 2-Aminobutyric acid, choline, and coproporphyrin as the most influential metabolites, with dysregulation of these markers associated with breast cancer risk. Conclusions: This study is among the first to integrate SHAP explainability with metabolomic profiling, bridging computational predictions and biological insights for improved clinical adoption. This study demonstrates the effectiveness of combining metabolomics with XAI-driven machine learning for breast cancer diagnostics. The identified biomarkers not only improve diagnostic accuracy but also reveal critical metabolic dysregulations associated with disease progression.
- Open Access
- 27 June 2024
- Published by MDPI AG in Diagnostics
- Vol. 14 (13) , 1364
Hybrid Explainable Artificial Intelligence Models for Targeted Metabolomics Analysis of Diabetic Retinopathy
Background: Diabetic retinopathy (DR) is a prevalent microvascular complication of diabetes mellitus, and early detection is crucial for effective management. Metabolomics profiling has emerged as a promising approach for identifying potential biomarkers associated with DR progression. This study aimed to develop a hybrid explainable artificial intelligence (XAI) model for targeted metabolomics analysis of patients with DR, utilizing a focused approach to identify specific metabolites exhibiting varying concentrations among individuals without DR (NDR), those with non-proliferative DR (NPDR), and individuals with proliferative DR (PDR) who have type 2 diabetes mellitus (T2DM). Methods: A total of 317 T2DM patients, including 143 NDR, 123 NPDR, and 51 PDR cases, were included in the study. Serum samples underwent targeted metabolomics analysis using liquid chromatography and mass spectrometry. Several machine learning models, including Support Vector Machines (SVC), Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), and Multilayer Perceptrons (MLP), were implemented as solo models and in a two-stage ensemble hybrid approach. The models were trained and validated using 10-fold cross-validation. SHapley Additive exPlanations (SHAP) were employed to interpret the contributions of each feature to the model predictions. Statistical analyses were conducted using the Shapiro–Wilk test for normality, the Kruskal–Wallis H test for group differences, and the Mann–Whitney U test with Bonferroni correction for post-hoc comparisons. Results: The hybrid SVC + MLP model achieved the highest performance, with an accuracy of 89.58%, a precision of 87.18%, an F1-score of 88.20%, and an F-beta score of 87.55%. SHAP analysis revealed that glucose, glycine, and age were consistently important features across all DR classes, while creatinine and various phosphatidylcholines exhibited higher importance in the PDR class, suggesting their potential as biomarkers for severe DR. Conclusion: The hybrid XAI models, particularly the SVC + MLP ensemble, demonstrated superior performance in predicting DR progression compared to solo models. The application of SHAP facilitates the interpretation of feature importance, providing valuable insights into the metabolic and physiological markers associated with different stages of DR. These findings highlight the potential of hybrid XAI models combined with explainable techniques for early detection, targeted interventions, and personalized treatment strategies in DR management.
- 27 June 2024
- Published by MDPI AG in Diagnostics
- Vol. 14 (13) , 1364
Open AccessHybrid Explainable Artificial Intelligence Models for Targeted Metabolomics Analysis of Diabetic Retinopathy
Background: Diabetic retinopathy (DR) is a prevalent microvascular complication of diabetes mellitus, and early detection is crucial for effective management. Metabolomics profiling has emerged as a promising approach for identifying potential biomarkers associated with DR progression. This study aimed to develop a hybrid explainable artificial intelligence (XAI) model for targeted metabolomics analysis of patients with DR, utilizing a focused approach to identify specific metabolites exhibiting varying concentrations among individuals without DR (NDR), those with non-proliferative DR (NPDR), and individuals with proliferative DR (PDR) who have type 2 diabetes mellitus (T2DM). Methods: A total of 317 T2DM patients, including 143 NDR, 123 NPDR, and 51 PDR cases, were included in the study. Serum samples underwent targeted metabolomics analysis using liquid chromatography and mass spectrometry. Several machine learning models, including Support Vector Machines (SVC), Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), and Multilayer Perceptrons (MLP), were implemented as solo models and in a two-stage ensemble hybrid approach. The models were trained and validated using 10-fold cross-validation. SHapley Additive exPlanations (SHAP) were employed to interpret the contributions of each feature to the model predictions. Statistical analyses were conducted using the Shapiro–Wilk test for normality, the Kruskal–Wallis H test for group differences, and the Mann–Whitney U test with Bonferroni correction for post-hoc comparisons. Results: The hybrid SVC + MLP model achieved the highest performance, with an accuracy of 89.58%, a precision of 87.18%, an F1-score of 88.20%, and an F-beta score of 87.55%. SHAP analysis revealed that glucose, glycine, and age were consistently important features across all DR classes, while creatinine and various phosphatidylcholines exhibited higher importance in the PDR class, suggesting their potential as biomarkers for severe DR. Conclusion: The hybrid XAI models, particularly the SVC + MLP ensemble, demonstrated superior performance in predicting DR progression compared to solo models. The application of SHAP facilitates the interpretation of feature importance, providing valuable insights into the metabolic and physiological markers associated with different stages of DR. These findings highlight the potential of hybrid XAI models combined with explainable techniques for early detection, targeted interventions, and personalized treatment strategies in DR management.
- Open Access
- 30 January 2024
- Published by Wiley
COVID-19 Severity Prediction in SARS-CoV-2 RNA-Positive Patients by Different Ensemble Learning Strategies
Objective: While the coronavirus persists marginally for ninety-five percent of the infected case count, the remaining five percent have been placed in a critical or vital condition. This study investigates to design an intelligent model that predicts the disease severity level by modeling the relationships between the severity of COVID-19 infection and the various demographic/clinical characteristics of individuals. Material and Methods: A public dataset of a cross-sectional study included the demographic and symptomatological characteristics of 223 COVID-19 patients. The dataset was randomly divided into training (75%) and testing (25%) datasets. During training, the class imbalance problem was solved, and the related factors with the COVID-19 severity were selected using the evolutionary method supported by a genetic algorithm. Neural Network (NN), Support Vector Machine (SVM), QUEST algorithms together with confidence weighted voting, voting, and highest confidence wins strategies (HCWS) were constructed, and the predictive power of models was evaluated by performance metrics. Results: Of the individual models, the NN model outperformed SVM and QUEST algorithms based on the performance metrics in the training and testing datasets. However, ensemble approaches gave better predictions as compared to the individual models regarding all the evaluation metrics. Conclusions: The proposed voting ensemble model outperforms other ensemble and individual machine learning approaches for the severity prediction of COVID-19 disease. The proposed ensemble learning model can be integrated into web or mobile applications in classifying the severity of COVID-19 for clinical decision support.
- 30 January 2024
- Published by Wiley
Open AccessCOVID-19 Severity Prediction in SARS-CoV-2 RNA-Positive Patients by Different Ensemble Learning Strategies
Objective: While the coronavirus persists marginally for ninety-five percent of the infected case count, the remaining five percent have been placed in a critical or vital condition. This study investigates to design an intelligent model that predicts the disease severity level by modeling the relationships between the severity of COVID-19 infection and the various demographic/clinical characteristics of individuals. Material and Methods: A public dataset of a cross-sectional study included the demographic and symptomatological characteristics of 223 COVID-19 patients. The dataset was randomly divided into training (75%) and testing (25%) datasets. During training, the class imbalance problem was solved, and the related factors with the COVID-19 severity were selected using the evolutionary method supported by a genetic algorithm. Neural Network (NN), Support Vector Machine (SVM), QUEST algorithms together with confidence weighted voting, voting, and highest confidence wins strategies (HCWS) were constructed, and the predictive power of models was evaluated by performance metrics. Results: Of the individual models, the NN model outperformed SVM and QUEST algorithms based on the performance metrics in the training and testing datasets. However, ensemble approaches gave better predictions as compared to the individual models regarding all the evaluation metrics. Conclusions: The proposed voting ensemble model outperforms other ensemble and individual machine learning approaches for the severity prediction of COVID-19 disease. The proposed ensemble learning model can be integrated into web or mobile applications in classifying the severity of COVID-19 for clinical decision support.
- 15 January 2023
- Published by LookUs Bilisim A.S. in Medical Records
- Vol. 5 (1) , 20-3
Development of Artificial Intelligence Based Clinical Decision Support System on Medical Images for the Classification of COVID-19
Aim: The first imaging method to play an vital role in the diagnosis of COVID-19 illness is the chest X-ray. Because of the abundance of large-scale annotated picture datasets, convolutional neural networks (CNNs) have shown considerable performance in image recognition/classification. The current study aims to construct a successful deep learning model that can distinguish COVID-19 from healthy controls using chest X-ray images.Material and Methods: The dataset in the study consists of subjects with 912 negative and 912 positive PCR results. A prediction model was built using VGG-16 with transfer learning for classifying COVID-19 chest X-ray images. The data set was split at random into 80% training and 20% testing groups.Results: The accuracy, F1 score, sensitivity, specificity, positive and negative values from the model that can successfully distinguish COVID-19 from healthy controls are 97.3%, 97.3%, 97.8%, 96.7%, 96.7%, and 97.8% regarding the testing dataset, respectively.Conclusion: The suggested technique might greatly improve on current radiology-based methodologies and serve as a beneficial tool for clinicians/radiologists in diagnosing and following up on COVID-19 patients.
- 15 January 2023
- Published by LookUs Bilisim A.S. in Medical Records
- Vol. 5 (1) , 20-3
Development of Artificial Intelligence Based Clinical Decision Support System on Medical Images for the Classification of COVID-19
Aim: The first imaging method to play an vital role in the diagnosis of COVID-19 illness is the chest X-ray. Because of the abundance of large-scale annotated picture datasets, convolutional neural networks (CNNs) have shown considerable performance in image recognition/classification. The current study aims to construct a successful deep learning model that can distinguish COVID-19 from healthy controls using chest X-ray images.Material and Methods: The dataset in the study consists of subjects with 912 negative and 912 positive PCR results. A prediction model was built using VGG-16 with transfer learning for classifying COVID-19 chest X-ray images. The data set was split at random into 80% training and 20% testing groups.Results: The accuracy, F1 score, sensitivity, specificity, positive and negative values from the model that can successfully distinguish COVID-19 from healthy controls are 97.3%, 97.3%, 97.8%, 96.7%, 96.7%, and 97.8% regarding the testing dataset, respectively.Conclusion: The suggested technique might greatly improve on current radiology-based methodologies and serve as a beneficial tool for clinicians/radiologists in diagnosing and following up on COVID-19 patients.