Yusuf Yesil

Dr. Yusuf Yesil

orcid

Research Focus & Metrics

Academic Career Length

5Years (2022-Present)

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

Representative Publications

View All Publications
31 December 2024 in Journal of AI
Verified
Journal Article
Open access
PDF file
Gülşah TorkayGülşah Torkay
Gülşah Torkay
NFNouran Fadlallah
Nouran Fadlallah
AKAhmet Karagöz
Ahmet Karagöz
MCMesut Canlı
Mesut Canlı
ESEzgi Saydam
Ezgi Saydam
YYYusuf Yesil
Yusuf Yesil
Cancer, a collection of maladies that has undergone extensive examination over centuries, remains a formidable challenge. Despite the array of available pharmacological and therapeutic interventions, the intricate molecular dynamics and heterogeneity of cancer continue to challenge the scientific community. Artificial Intelligence (AI) emerges as a promising avenue, offering the potential for expedited, precise diagnostics devoid of human expertise. Additionally, AI facilitates the tailoring of patient-specific therapeutic strategies targeting various facets of cancer, spanning macroscopic to microscopic levels. Nonetheless, it is imperative to scrutinize the potential benefits and limitations of AI technologies in this context. This review undertakes a comprehensive Strengths, Weaknesses, Opportunities, and Threats (SWOT) analysis of AI's application in cancer. An extensive compilation of AI applications encompasses predictive modeling, diagnostic capabilities, prognostic assessments, and personalized therapeutic modalities, spanning genomic analyses to individualized treatment regimens. The synthesis of evidence suggests that the advantages of AI outweigh its drawbacks; nevertheless, obstacles to its widespread integration persist.

30 December 2023 in Interdisciplinary Medical Journal
Verified
Journal Article
Open access
PDF file
YYYusuf Yesil
Yusuf Yesil
MAMustafa Altindis
Mustafa Altindis
HTHande Toptan
Hande Toptan
EKElmas Pınar Kahraman Kılbaş
Elmas Pınar Kahraman Kılbaş
OBOnur Bi̇rCan
Onur Bi̇rCan
MKMehmet Koroglu
Mehmet Koroglu
Objective: In this study, it was aimed to distinguish bacterial/viral tonsillopharyngitis (TP) by scoring the symptom and throat images of pediatric patients with artificial intelligence-based mobile application. Method: Fifty-one patients who applied to Sakarya University Training and Research Hospital, Department of Pediatrics and Diseases with acute tonsillopharyngitis were included. Samples were taken from patients and mouth/throat pictures were taken so that the tonsils and pharynx were clearly visible. In the microbiology laboratory, identification with culture/MALDI-TOF MS (Biomerieux, France) from the first samples, and nucleic acid isolation from the other for molecular tests were performed. Symptoms such as fatigue, sore throat, muscle pain, cough, sneezing, and runny nose were questioned from each patient on a scale of 1 to 5. By uploading the symptom results and throat pictures to the artificial intelligence application, it was aimed to distinguish bacterial/viral tonsillopharyngitis with the developed scoring system. Results: Of the 51 samples included in the study, 21 were culture positive and 30 were negative. The artificial intelligence application was defined as 20 out of 21 culture-positive samples, 3 out of 30 culture-negative samples as bacterial tonsillopharyngitis (Sensitivity: 95.2%, specificity: 90%). Conclusion: This study is one of the first to bring together the artificial intelligence application and microbiology. AI/scoring system may have a role to play in the diagnosis of bacterial vs viral TP, and in doing so may enable more appropriate antibiotic usage targeted to only bacterial TP infections. It is important to distinguish between bacterial and viral tonsillopharyngitis in the COVID-19 pandemic.

29 May 2025 in Medicina
Verified
Journal Article
Open access
PDF file
MDMansur Dogan
Mansur Dogan
MÇMerve Çiftçi
Merve Çiftçi
YYYusuf Yesil
Yusuf Yesil
Background and Objectives: The nasopharynx, unlike other pharyngeal regions, includes an important part of the immune system, called the adenoid (nasopharyngeal tonsil); its posterior wall contains lymphoid tissue belonging to Waldeyer’s ring. Nasopharyngeal posterior wall thickness is often associated with adenoid hyperplasia in adults. The current study aimed to compare the blood lipid and metabolic profiles of adult patients with increased nasopharyngeal posterior wall thickness to those of the healthy population. Materials and Methods: This study included a cohort of 98 patients, 52 in the control group and 46 diagnosed with increased nasopharyngeal posterior wall thickness due to adenoid hyperplasia. Clinical and biochemical data were collected from medical records at Sivas Cumhuriyet University and Erbaa State Hospital between January 2024 and March 2025. The dataset consisted of the following 11 features: age, sex, total cholesterol, low-density lipoprotein (LDL), high-density lipoprotein (HDL), triglycerides, fasting blood glucose, glycated hemoglobin (HbA1C), C-reactive protein (CRP), alanine aminotransferase (ALT), and aspartate aminotransferase (AST). Results: HDL was significantly lower in the adenoid hyperplasia group (mean = 48.68, SD = 21.87) compared to the control group (mean = 51.31, SD = 11.80; Kruskal–Wallis H = 4.750, p = 0.029), with a small effect size (Cohen’s d = −0.156). ALT was higher in the adenoid hyperplasia group (mean = 26.35, SD = 16.93 vs. 20.88, SD = 11.42; permutation test p = 0.082), suggesting a trend toward significance. HbA1C had a higher mean in the adenoid hyperplasia group (7.88, SD = 9.82 vs. 6.18, SD = 1.18; p = 0.852), with high variability. Conclusions: In conclusion, this study identified HDL, HbA1C, and ALT as potential biomarkers for nasopharyngeal adenoid hyperplasia, with XGBoost and SHAP providing valuable insights despite dataset constraints.

31 July 2026 in B-ENT
Verified
Journal Article
Open access
PDF file
MÇMerve Çiftçi
Merve Çiftçi
TKTuba DOGAN KARATAS
Tuba DOGAN KARATAS
YYYusuf Yesil
Yusuf Yesil
MDMansur Dogan
Mansur Dogan
Background: Cholesteatoma detection in chronic otitis media (COM) remains challenging, requiring comprehensive diagnostic approaches. While traditional statistical analysis of individual biomarkers may show limitations, the integration of multiple analytical methods including machine learning can provide deeper insights into disease prediction. This study aimed to conduct a comprehensive analysis combining traditional statistical methods with advanced machine learning to evaluate cholesteatoma detection using hematological parameters. Methods: A retrospective analysis was conducted on 79 patients with COM (40 with cholesteatoma, 39 without cholesteatoma). Complete blood count parameters and inflammatory ratios (neutrophil-to-lymphocyte ratio [NLR], platelet-to-lymphocyte ratio [PLR], monocyte-tolymphocyte ratio [MLR]) were analyzed using (1) traditional statistical methods including Mann–Whitney U-tests, receiver oeprating characteristic (ROC) curve analysis, and correlation studies and (2) advanced machine learning algorithms including XGBoost and logistic regression with cross-validation. Results: Traditional statistical analysis revealed important baseline characteristics: no significant age differences between groups, and individual inflammatory ratios showed limited discriminative ability (NLR: P = .9414, PLR: P = .6067, MLR: P > .05). However, machine learning integration demonstrated that these same parameters could achieve clinically relevant predictive capability. XGBoost achieved 66.7% accuracy with 0.701 AUC-ROC, while ROC analysis of individual parameters yielded area under the receiver operating characteristic curve (AUC) values of 0.508- 0.545. The combination of statistical validation and machine learning optimization provided complementary insights. Conclusion: This comprehensive analysis demonstrates the value of integrating traditional statistical methods with machine learning approaches. While individual biomarker analysis confirmed the complexity of cholesteatoma diagnosis, machine learning successfully leveraged multiple parameters to achieve predictive capability. The complementary use of both analytical approaches provides a robust foundation for clinical decision-making and establishes a methodology for future diagnostic tool development.

14 February 2025 in JMIR Preprints
Verified
Preprint
Open access
YYYusuf Yesil
Yusuf Yesil
BACKGROUND Integrating evidence-based approaches in healthcare and artificial intelligence (AI) is crucial for enhancing clinical decision-making and patient safety. Yesil o1 Pro is a specialized large language model (LLM) designed to transform medical knowledge synthesis by leveraging a comprehensive, curated database of scientific literature, clinical guidelines, and medical textbooks. OBJECTIVE The system's innovative "AI Hospital" framework employs domain-specific expert agents coordinated by a central Master Agent, enabling tailored and precise medical responses across multiple disciplines. METHODS The model's advanced methodology incorporates sophisticated techniques including GraphRAG-based retrieval, extensive fine-tuning with 1.5 million question-answer pairs, and Chain of Thought (CoT) reasoning. Its robust training dataset comprises 100.5M words from high-impact journals, 96.8M words from core medical texts, and 74.4M words from international standards, ensuring a comprehensive and authoritative knowledge base. RESULTS Benchmark evaluations demonstrate Yesil o1 Pro's exceptional performance, achieving an overall accuracy of 89.1% and surpassing leading models like GPT-4o (83.9%) and Claude 3.5 Sonnet (83.0%). Domain-specific accuracies are particularly impressive, with 96.1% in Mental Health, 94.6% in Epidemiology, and 94.6% in Dentistry, highlighting the model's proficiency in handling complex, reasoning-intensive medical queries. CONCLUSIONS The model shows promising applications in clinical decision support, interdisciplinary collaboration, professional education, and medical research. While challenges remain in real-world integration and maintaining alignment with evolving medical knowledge, Yesil o1 Pro represents a significant advancement in AI-driven healthcare support. Future research will focus on validating the model's utility in clinical environments and developing strategies for seamless healthcare system integration.

5 October 2026 in Laboratory Medicine
Verified
Journal Article
Closed access
YYYusuf Yesil
Yusuf Yesil
AMAlpay Medetalibeyoglu
Alpay Medetalibeyoglu
EAEvin Ademoglu
Evin Ademoglu
Introduction: Routine laboratory panels are nearly universal, but the panels’ joint information is underused. We evaluated contemporaneous classification of International Statistical Classification of Diseases, Tenth Revision (ICD-10) code groups from same-encounter laboratory results. Methods: We developed 17 eXtreme Gradient Boosting (XGBoost) classifiers in 242 648 adult internal medicine encounters using age, sex, and results from 34 assays, with ICD-10 codes used only as outcomes. Stratified 10-fold cross-validation assessed discrimination, calibration, and decision curve performance. Leakage-free recalibration and fold-specific thresholds targeted 95% specificity. Using an 80-patient temporal holdout cohort, we compared the model with 3 large language models accessed through an application programming interface and findings from 1 specialist physician. Results: The mean cross-validated area under the curve (AUC) was 0.893. With identical preprocessing, the mean AUC was 0.892 for XGBoost and 0.809 for logistic regression. Recalibration changed the mean intercept and slope to −0.002 and 0.999, respectively, and the Brier score from 0.099 to 0.044. The holdout micro-averaged AUC was 0.880 vs 0.729 to 0.739 for language models; paired differences were 0.141 to 0.151 (all P < .005). Matched-specificity sensitivity differences favored the model but were imprecise. Discussion: Routine panels contain substantial information about contemporaneous ICD-10 coding, supporting prospective multicenter evaluation rather than a diagnostic claim.