Journals / Süleyman Demirel Üniversitesi Fen Bilimleri Enstitüsü Dergisi / 2026 / Cilt: 29 Sayı: 2

A Hybrid Retrieval-And-Generation Framework For Radiology Report Summarization With Faiss Indexing and T5 Transformers

Faiss İndeksleme ve T5 Dönüştürücüleri ile Radyoloji Rapor Özetleme İçin Hibrit Bir Erişim ve Üretim Çerçevesi

Ayhan Arısoy(MEHMET AKİF ERSOY ÜNİVERSİTESİ)0000-0001-6754-932X
Pages
483–492
Publication Date
2025-08-25
🏛️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)

Experimental results demonstrate the model’s effectiveness in terms of both lexical fidelity and semantic consistency, achieving ROUGE-1 of 0.5299, ROUGE-2 of 0.4206, METEOR of 0.5018, and compression ratio of 0.9213.

📌 3 Key Takeaways

🎯
Objective & Hypothesis:Radiology reports contain clinically critical yet complex information that can overwhelm physicians when available in a raw narrative form.
📊
Key Findings & Data:To address this challenge, this study proposes a Retrieval-Augmented Generation (RAG) approach to medical report summarization by integrating a FAISS-based semantic search engine with a small T5 generative model.
💡
Conclusion & Impact:Experimental results demonstrate the model’s effectiveness in terms of both lexical fidelity and semantic consistency, achieving ROUGE-1 of 0.5299, ROUGE-2 of 0.4206, METEOR of 0.5018, and compression ratio of 0.9213.
🔬Design:Modeling & Simulation
📍Scope:Comprehensive Sample
⚙️Variables:Tıbbi Rapor Özetleme · Getirmeli Destekli Üretim (RAG) · FAISS Tabanlı Anlamsal Arama

Abstract

Radiology reports contain clinically critical yet complex information that can overwhelm physicians when available in a raw narrative form. To address this challenge, this study proposes a Retrieval-Augmented Generation (RAG) approach to medical report summarization by integrating a FAISS-based semantic search engine with a small T5 generative model. The methodology uses Sentence-BERT to embed the “Findings” section of radiology reports from the MIMIC-III dataset, which are then indexed using FAISS to retrieve semantically similar cases. In this approach, The retrieved context has been merged with the original 'Finding' section and subsequently fed into the T5 model to produce the respective 'Impression' summary. Experimental results demonstrate the model’s effectiveness in terms of both lexical fidelity and semantic consistency, achieving ROUGE-1 of 0.5299, ROUGE-2 of 0.4206, METEOR of 0.5018, and compression ratio of 0.9213. Domain-specific semantic matching was validated using FactCC (0.5463), CheXpert Label Agreement (0.6194), and Medical Concept Overlap (0.5552) scores. The model demonstrated stable convergence over seven epochs without overfitting, as evidenced by the steadily decreasing validation loss. Qualitative examples show that the model generates smooth and clinically consistent summaries, although occasional factual hallucinations indicate areas for improvement. Overall, the proposed FAISS+T5 process addresses key limitations of traditional summarization methods by integrating contextual retrieval and generation. This framework offers a scalable, interpretable, and domain-specific solution for summarizing clinical texts and shows promise for implementation in real-world decision support systems.

Özet

Radyoloji raporları, ham anlatı biçiminde sunulduğunda hekimleri zorlayabilecek klinik açıdan kritik ancak karmaşık bilgiler içerir. Bu sorunu çözmek için, bu çalışma FAISS tabanlı bir semantik arama motorunu küçük bir T5 üretici modelle entegre ederek tıbbi rapor özetleme için Retrieval-Augmented Generation (RAG) yaklaşımını önermektedir. Bu metodoloji, Sentence-BERT'i kullanarak MIMIC-III veri setinden radyoloji raporlarının “Bulgular” bölümünü gömmektedir. Bu bölümler daha sonra FAISS kullanılarak indekslenerek semantik olarak benzer vakalar geri getirilmektedir. Bu yaklaşımda, geri getirilen bağlam orijinal ‘Bulgular’ bölümüyle birleştirilmekte ve ardından T5 modeline beslenerek ilgili “İzlenim” özeti üretilmektedir. Deney sonuçları, modelin hem sözcüksel doğruluk hem de semantik tutarlılık açısından etkinliğini göstermektedir. Model, ROUGE-1'de 0,5299, ROUGE-2'de 0,4206, METEOR'da 0,5018 ve sıkıştırma oranında 0,9213 değerlerine ulaşmıştır. Alanına özgü anlamsal eşleştirme, FactCC (0,5463), CheXpert Label Agreement (0,6194) ve Medical Concept Overlap (0,5552) puanları kullanılarak doğrulanmıştır. Model, sürekli azalan doğrulama kaybı ile kanıtlandığı üzere, aşırı uyum olmadan yedi dönem boyunca istikrarlı bir yakınsama göstermiştir. Niteliksel örnekler, modelin düzgün ve klinik olarak tutarlı özetler ürettiğini göstermektedir, ancak ara sıra görülen gerçek dışı halüsinasyonlar iyileştirilmesi gereken alanları göstermektedir. Genel olarak, önerilen FAISS+T5 süreci, bağlamsal geri alma ve üretimi entegre ederek geleneksel özetleme yöntemlerinin temel sınırlamalarını ele almaktadır. Bu çerçeve, klinik metinleri özetlemek için ölçeklenebilir, yorumlanabilir ve alana özgü bir çözüm sunar ve gerçek dünyadaki karar destek sistemlerinde uygulanması için umut vaat etmektedir.

🔖

Cite This Article

Arısoy, A. (2025). A Hybrid Retrieval-And-Generation Framework For Radiology Report Summarization With Faiss Indexing and T5 Transformers. Süleyman Demirel Üniversitesi Fen Bilimleri Enstitüsü Dergisi, Cilt: 29 Sayı: 2, 483–492. https://doi.org/10.19113/sdufenbed.1739565

📄 Full Text PDF Viewer

📄

PDF görüntüleyici hazırlanıyor...

References (17)

  1. Gauthier, L. W. et al. 2023. Assessing feasibility and risk to translate, de-identify and summarize medical reports using deep learning, medRxiv, p. 2023.07.27.23293234, Aug. 2023.
  2. Scott, D., Hallett, C., Fettiplace, R. 2013. Data-to-text summarisation of patient records: Using computer-generated summaries to access patient histories, Patient Educ. Couns., vol. 92, no. 2, pp. 153–159, Aug. 2013.
  3. Ahmed, B., Balouch, K., Hussain, F. 2023. A Transformer based approach for Abstractive Text Summarization of Radiology Reports, Int. Conf. Appl. Eng. Nat. Sci., vol. 1, no. 1, pp. 476–486, Jul. 2023.
  4. Lee, E. K., Uppa, K. l. 2020. CERC: An interactive content extraction, recognition, and construction tool for clinical and biomedical text, BMC Med. Inform. Decis. Mak., vol. 20, no. 14, pp. 1–14, Dec. 2020.
  5. Kay S. 2020. The International Patient Summary and the Summarization Requirement,” Stud. Health Technol. Inform., vol. 285, pp. 17–30, Oct. 2021.
  6. Zhang, Y. et al. 2020. When Radiology Report Generation Meets Knowledge Graph, Proc. Aaai Conf. Artif. Intell., 2020.
  7. Grewal H. et al. 2023. Radiology Gets Chatty: The ChatGPT Saga Unfolds, Cureus, vol. 15, no. 6, Jun. 2023.
  8. Wang Y. et al. 2024. Optimizing Data Extraction: Harnessing RAG and LLMs for German Medical Documents, Stud. Health Technol. Inform., vol. 316, pp. 949–950, Aug. 2024.
  9. Jony A. I., Rithin A. T., Edrish S. I. 2024. A Comparative Study and Analysis of Text Summarization Methods, Malaysian J. Sci. Adv. Technol., vol. 4, no. 2, pp. 118–129, Mar. 2024.
  10. Arora M., et al. 2023. Evaluation of text summarization techniques in healthcare domain: Pharmaceutical drug feedback, Intell. Decis. Technol., vol. 17, no. 4, pp. 1309–1322, Nov. 2023.
  11. Wang M., et al. 2021. A systematic review of automatic text summarization for biomedical literature and EHRs, J. Am. Med. Informatics Assoc., vol. 28, no. 10, pp. 2287–2297, Sep. 2021.
  12. Keszthelyi D., et al. 2023. Patient Information Summarization in Clinical Settings: Scoping Review, JMIR Med Inf. 2023;11e44639 https//medinform.jmir.org/2023/1/e44639, vol. 11, no. 1, p. e44639, Nov. 2023.
  13. Zhang Y., et al. 2020. When Radiology Report Generation Meets Knowledge Graph, Proc. AAAI Conf. Artif. Intell., vol. 34, no. 07, pp. 12910–12917, Apr. 2020.
  14. Karpukhin V. et al. 2020. Dense Passage Retrieval for Open-Domain Question Answering, EMNLP 2020 - 2020 Conf. Empir. Methods Nat. Lang. Process. Proc. Conf., pp. 6769–6781, 2020.
  15. Lewis P. et al. 2020. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, NIPS’20 Proc. 34th Int. Conf. Neural Inf. Process. Syst., pp. 9459–9474, Dec. 2020.
  16. Zhao, W., Fu, L., Huang, Z., Zhu, J., & Ma, B. (2019). Effectiveness evaluation of computer-aided diagnosis system for the diagnosis of thyroid nodules on ultrasound. Medicine, 98(32), e16379.
  17. Guo, L., Zhou, C., Xu, J., Huang, C., Yu, Y., & Lu, G. (2024). Deep learning for chest x-ray diagnosis: competition between radiologists with or without artificial intelligence assistance. Journal of Imaging Informatics in Medicine, 37(3), 922-934.