Journals / Uludağ Üniversitesi Tıp Fakültesi Dergisi / 2025 / Cilt: 51 Sayı: 2

Evaluating the Performance of Large Language Models in Generating Impressions for Radiology Reports

Büyük Dil Modellerinin Radyoloji Raporları İçin Sonuç Bölümü Oluşturmadaki Performanslarının Değerlendirilmesi

Hasan Emin Kaya(Bursa Uludağ University, Faculty of Medicine, Department of Radiology, Bursa, Türkiye.)0000-0002-7411-4102
Dilek Sağlam(Bursa Uludağ University, Faculty of Medicine, Department of Radiology, Bursa, Türkiye.)0000-0002-5778-6847
Zeynep Yazıcı(Bursa Uludağ University, Faculty of Medicine, Department of Radiology, Bursa, Türkiye.)0000-0002-8647-5298
Gökhan Gökalp(Bursa Uludağ University, Faculty of Medicine, Department of Radiology, Bursa, Türkiye.)0000-0002-3682-2474
Pages
305–309
Publication Date
2025-08-28
🏛️Bibliographic metadata and full text access for this article are provided via DergiPark.
DergiPark Record ↗
✨

AI Summary & Key Takeaways

AI Insight
⚡ TL;DR(Too Long; Didn't Read)

In conclusion, ChatGPT, Gemini, and Copilot were found to be successful in generating impressions for radiology reports in Turkish, and no significant difference in performance was detected among the models.

📌 3 Key Takeaways

🎯
Objective & Hypothesis:The aim of the study was to evaluate and compare the performance of three popular large language models (LLMs) in generating impressions for radiology reports in Turkish.
📊
Key Findings & Data:ChatGPT, Gemini, and Copilot were used to generate impressions for 50 anonymized radiology reports using a “few-shot” prompt.
💡
Conclusion & Impact:In conclusion, ChatGPT, Gemini, and Copilot were found to be successful in generating impressions for radiology reports in Turkish, and no significant difference in performance was detected among the models.
🔬Design:Modeling & Simulation
📍Scope:Comprehensive Sample
⚙️Variables:radyoloji · yapay zeka · büyük dil modelleri

Abstract

The aim of the study was to evaluate and compare the performance of three popular large language models (LLMs) in generating impressions for radiology reports in Turkish. ChatGPT, Gemini, and Copilot were used to generate impressions for 50 anonymized radiology reports using a “few-shot” prompt. The impressions were scored by three radiologists using a Likert scale, based on whether they included all relevant information from the report, provided an appropriate summary of the report, contained no misleading information, and could be added to the report without modification. Friedman's test was used to evaluate whether there was a difference between the scores of the LLMs. The 50 reports included 32 magnetic resonance examinations, 11 computed tomography examinations, 5 ultrasound examinations, and 2 fluoroscopy examinations. Of these, 15 were neuroradiology studies, 14 were musculoskeletal studies, 13 were abdominal studies, and 8 were thoracic radiology studies. The median scores for the models’ outputs were 4 and 5. This finding indicates that the radiologists generally found the models successful in generating impressions. Furthermore, no statistically significant difference was found among the models in terms of their performance in containing all information, providing an appropriate summary, avoiding misleading information, and being suitable for inclusion in the report without modification (p = 0.607, 0.327, 0.629, 0.089, respectively). In conclusion, ChatGPT, Gemini, and Copilot were found to be successful in generating impressions for radiology reports in Turkish, and no significant difference in performance was detected among the models.

Özet

Çalışmamızın amacı popüler üç büyük dil modelinin (BDM) Türkçe radyoloji raporları için sonuç bölümü oluşturma konusundaki performansını değerlendirip mukayese etmekti. Anonimize edilmiş 50 radyoloji raporu için, “few-shot” bir komut ile, ChatGPT, Gemini ve Copilot dil modellerine sonuç bölümü oluşturuldu. Sonuçlar; rapordaki tüm bilgileri içerme, raporu uygun bir şekilde özetleme, yanıltıcı bilgi içermeme ve değiştirilmeden rapora eklenebilme açısından üç radyolog tarafından bir Likert skalası kullanılarak skorlandı. Friedman testi ile BDM’lerin skorları arasında fark olup olmadığı değerlendirildi. Çalışmaya dahil edilen 50 raporun 32’si manyetik rezonans, 11’i bilgisayarlı tomografi, 5’i ultrason ve 2’si floroskopi tetkikleriydi. Bu tetkiklerden 15’i nöroradyoloji, 14’ü kas-iskelet, 13’ü abdomen ve 8’i toraks radyolojisi çalışmalarıydı. Üç radyoloğun yaptığı skorlamalarda modellerin aldığı skorların medyan değerleri 4 ve 5 idi. Bu bulgu modellerin sonuç oluşturmada radyologlar tarafından genel olarak başarılı bulunduğunu göstermekteydi. Ayrıca modeller arasında bütün bilgileri içerme, raporu uygun bir şekilde özetleme, yanıltıcı bilgi içermeme ve değiştirilmeden rapora eklenebilme performansı açısından istatistiksel bir farklılık saptanmadı (p değerleri sırasıyla 0,607; 0,327; 0,629; 0,089). Sonuç olarak ChatGPT, Gemini ve Copilot Türkçe radyoloji raporları için sonuç bölümü oluşturmada başarılı bulunmuş ve modellerin performansı arasında anlamlı bir farklılık saptanmamıştır.

🔖

Cite This Article

Kaya, H. E., Sağlam, D., Yazıcı, Z., Gökalp, G., *, H. E. K. (2025). Evaluating the Performance of Large Language Models in Generating Impressions for Radiology Reports. Uludağ Üniversitesi Tıp Fakültesi Dergisi, Cilt: 51 Sayı: 2, 305–309. https://doi.org/10.32708/uutfd.1653680

📄 Full Text PDF Viewer

📄

PDF görüntüleyici hazırlanıyor...

References (8)

  1. Elkassem AA, Smith AD. Potential Use Cases for ChatGPT in Radiology Reporting. AJR Am J Roentgenol 2023;221(3):373–6.
  2. Amin KS, Davis MA, Doshi R, Haims AH, Khosla P, Forman HP. Accuracy of ChatGPT, Google Bard, and Microsoft Bing for Simplifying Radiology Reports. Radiology 2023;309(2).
  3. Sun Z, Ong H, Kennedy P, Tang L, Chen S, Elias J, et al. Evaluating GPT4 on Impressions Generation in Radiology Reports. Radiology 2023;307(5).
  4. Zhang L, Liu M, Wang L, Zhang Y, Xu X, Pan Z, et al.Constructing a Large Language Model to Generate Impressionsfrom Findings in Radiology Reports. Radiology 2024;312(3).
  5. Doshi R, Amin KS, Khosla P, Bajaj S, Chheang S, Forman HP.Quantitative Evaluation of Large Language Models toStreamline Radiology Report Impressions: A Multimodal Retrospective Analysis. Radiology 2024;310(3).
  6. Can E, Uller W, Vogt K, Doppler MC, Busch F, Bayerl N, et al.Large Language Models for Simplified Interventional Radiology Reports: A Comparative Analysis. Acad Radiol. 2024 Sep 30:S1076-6332(24)00690-1. doi: 10.1016/j.acra.2024.09.041. Epub ahead of print. PMID: 39353826.
  7. Ueda D, Mitsuyama Y, Takita H, Horiuchi D, Walston SL,Tatekawa H, et al. ChatGPT’s Diagnostic Performance fromPatient History and Imaging Findings on the Diagnosis Please Quizzes. Radiology 2023;308(1).
  8. Horiuchi D, Tatekawa H, Shimono T, Walston SL, Takita H, Matsushita S, et al. Accuracy of ChatGPT generated diagnosis from patient’s medical history and imaging findings in neuroradiology cases. Neuroradiology 2024;66(1):73–9.