Tolga Keçeci

Dr. Tolga Keçeci

orcid

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

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

Representative Publications

View All Publications
29 September 2025 in BMC Medical Informatics and Decision Making
Verified
Journal Article
Open access
PDF file
TKTolga Keçeci
Tolga Keçeci
BKBekir Karagöz
Bekir Karagöz
Artificial intelligence (AI)-based large language models (LLM) are increasingly used in healthcare education. However, the accuracy, readability, and reliability of their medical outputs remain a concern. This study aimed to compare the quality of responses generated by ChatGPT-4o and DeepSeek AI regarding the diagnosis and treatment of clavicle fractures, based on the 2022 AAOS Clinical Practice Guidelines (CPG). Fourteen clinical questions were formulated based on the AAOS CPG for clavicle fractures. Each question was independently submitted to ChatGPT-4o and DeepSeek AI. Responses were evaluated using standardized scoring tools, including DISCERN, PEMAT-P, CLEAR, Flesch-Kincaid Grade Level, Flesch Reading Ease, and Gunning-Fog Index. Two orthopedic surgeons independently rated the responses, and inter-rater scores were averaged. Statistical comparison between the two AI models was conducted using the Mann–Whitney U test. DeepSeek AI generated responses with a significantly higher word count (median: 572, IQR: 258.25 vs. 438.5, IQR: 229; p = 0.016), CLEAR score (median: 18, IQR: 0.75 vs. 16, IQR: 0.75; p < 0.001). No statistically significant differences were found in PEMAT understandability (median: 77.7 vs. 77.7; p = 0.519), PEMAT actionability (median: 0 vs. 0; p = 1.000), or PEMAT total score (median: 57.2 vs. 58; p > 0.05). Similarly, no statistically significant differences were observed in DISCERN (52.1 vs. 51.6; p > 0.05), readability indices, binary accuracy (ChatGPT: 0.93, DeepSeek: 0.89; p > 0.05), or weighted accuracy (ChatGPT: 0.83, DeepSeek: 0.79; p > 0.05). Both models demonstrated generally high accuracy levels. Both ChatGPT-4o and DeepSeek AI generated coherent and clinically relevant responses to guideline-based questions on clavicle fracture management. However, neither model achieved meaningful scores in PEMAT actionability, and occasional inaccuracies and hallucinations were observed. While DeepSeek produced longer responses, verbosity did not correspond to superior quality. These findings suggest that LLMs may serve as supplementary tools for medical education and reference, but they cannot replace evidence-based clinical judgment, underscoring the need for supervised integration and ongoing validation. Finally, the number of prompts analyzed (14) was limited, reflecting the scope of a single guideline; this sample size restricts statistical power and generalizability, and larger multi-guideline datasets will be needed in future studies.

1 September 2017 in Journal of Pediatric Orthopaedics B
Verified
Journal Article
Closed access
PDF file
SASuavi Aydoğmuş
Suavi Aydoğmuş
TDTahir Mutlu Duymuş
Tahir Mutlu Duymuş
TKTolga Keçeci
Tolga Keçeci
LALevent Adiyeke
Levent Adiyeke
AKAdnan B. Kafadar
Adnan B. Kafadar
Reports on different types of surgical procedures show that night-time surgeries are often associated with a high number of complications than daytime surgeries. However, there are no data, and hence evidence that relate the outcome of supracondylar humeral fractures in children to the time of the day when the respective surgical procedure is implemented. In this study, we analyze the clinical and radiological outcomes of pediatric supracondylar humeral fractures surgically treated during daytime and night-time hours. In total, 91 patients with Gartland type-3 fractures were included; 47 patients treated between 0800 and 1700 h were defined as the daytime group and 44 patients treated between 1701 and 0759 h were defined as the after-hours group. Age, sex, affected side, fracture type, and day and time of operation were recorded. Any preoperative neurovascular injuries, open fractures, or ipsilateral fractures were noted. The surgical method, use of a medial pin or pins, operation time, any postoperative neurovascular complications, extent of successful reduction, fixation rate, any resultant deformity, and functional loss rate were evaluated. The two groups did not differ significantly in terms of operation time, open reduction rate, rate of poor reduction, extent of poor functional outcomes, or induction of deformity (P>0.05). The poor fixation rate was significantly greater in the after-hours group than in the daytime group (P0.05). After-hours treatment of displaced supracondylar humeral fractures in children is associated with a higher poor fixation rate compared with daytime procedures. Level of evidence: therapeutic study (retrospective comparative study), Level III.

BKBekir Karagoz
Bekir Karagoz
HBHunkar Cagdas Bayrak
Hunkar Cagdas Bayrak
TKTolga Keçeci
Tolga Keçeci
Background: This study aims to compare the quality, reliability, and readability of information provided by artificial intelligence-based language models, ChatGPT-5 and DeepSeek V3, regarding foot and ankle disorders. Methods: The quality, reliability, and readability of the texts generated by both AI models were analyzed using DISCERN, the Patient Education Materials Assessment Tool for Printable Materials (PEMAT-P), the Global Quality Score (GQS), and the CLEAR scoring system. DISCERN was used to assess information reliability, PEMAT-P to evaluate understandability and actionability, GQS to assess overall quality, and CLEAR to evaluate content quality and accuracy. Standardized questions were asked to both models for 35 different foot and ankle disorders, and the generated texts were evaluated by two independent orthopedic specialists using a blinded method. Readability analysis was performed using word count, the Flesch–Kincaid Grade Level (FKGL; required reading level), and the Flesch Reading Ease (FRE; ease of readability) scoring systems. Results: ChatGPT-5 scored significantly higher than DeepSeek V3 in DISCERN, PEMAT-P, GQS, and CLEAR evaluations (p < 0.05), indicating that ChatGPT-5 provides more reliable, comprehensive, and higher-quality information. DeepSeek V3 demonstrated better readability, producing simpler and more understandable content, as reflected in its lower FKGL score and higher FRE score. Conclusions: While ChatGPT-5 delivers more detailed and reliable health information, DeepSeek V3 offers simpler and more readable texts. Both models have distinct advantages for patient education. Future research should assess the impact of AI-generated health information on patient decision-making and its clinical application potential.

1 March 2019 in Journal of Clinical Orthopaedics and Trauma
Verified
Journal Article
Closed access
TDTahir Mutlu Duymus
Tahir Mutlu Duymus
SASuavi Aydogmus
Suavi Aydogmus
IUİbRahim Ulusoy
İbRahim Ulusoy
TKTolga Keçeci
Tolga Keçeci
LALevent Adiyeke
Levent Adiyeke
SMSerhat Mutlu
Serhat Mutlu

24 February 2022 in Cureus
Verified
Journal Article
Open access
BKBekir Karagoz
Bekir Karagoz
MBMURAT BAKIR
MURAT BAKIR
TKTolga Keçeci
Tolga Keçeci
In this study, it was aimed to determine the quality and accuracy of the videos on YouTube about lateral epicondylitis. The first 100 videos were included in the study by typing the keyword "lateral epicondylitis" in the YouTube search tab without using any filters. The video power index (VPI) was used to evaluate the popularity of the videos, and the global quality score (GQS), Journal of the American Medical Association (JAMA), and DISCERN scoring systems were used to evaluate the quality. The obtained data were statistically analyzed according to these scoring systems. The mean DISCERN, JAMA, and GQS of the analyzed videos were 46.66, 3.13, and 3.85, respectively. According to these results, it was determined that the videos were of medium quality. A statistically insignificant and weak correlation was found between the VPI and DISCERN, GQS, and JAMA scores (p>0.05, intraclass correlation coefficient, ICC: −0.05, 0.09, and −0.05, respectively). While there was no significant relationship between the video source and the DISCERN, JAMA, and GQS scores (p>0.05), it was determined that the DISCERN, JAMA, and GQS scores in the exercise videos were significantly higher than in the other content types in terms of the video content (p=0.041). According to the results obtained, it was determined that YouTube videos about lateral epicondylitis were not of sufficient quality. In order to ensure standardization for quality videos, internationally acceptable guidelines should be determined and studies should be carried out to provide an adequate infrastructure for the preparation of quality medical videos that can meet the increasing needs of patients by health institutions.

14 March 2016 in Case Reports in Orthopedics
Verified
Journal Article
Open access
PDF file
SASuavi Aydoğmuş
Suavi Aydoğmuş
TDTahir Mutlu Duymuş
Tahir Mutlu Duymuş
TKTolga Keçeci
Tolga Keçeci
This study evaluated complications associated with implant depth in headless compression screw treatment of an osteochondral fracture associated with a traumatic patellar dislocation in a 21-year-old woman. Computed tomography and X-rays showed one lateral fracture fragment measuring 25 × 16 mm. Osteosynthesis was performed with two headless compression screws. Five months later, the screws were removed because of patella-femoral implant friction. We recommend that the screw heads be embedded to a depth of at least 3 mm below the cartilage surface. Further clinical studies need to examine the variation in cartilage thickness in the fracture fragment.