Towards Automating Lung-RADS Classification: Insights from Portuguese Radiology Reports

• Tarcísio Lima Ferreira UFAL
• Marcelo Costa Oliveira UFAL
• Juliana Simon Petruceli UFAL

DOI: https://doi.org/10.5753/sbcas.2026.20778

Resumo

Early lung cancer detection requires efficient nodule classification. This study compares automated strategies for assigning Lung-RADS categories to Portuguese CT reports using DL and LLMs. Analyzing 963 reports, we employed NER (BiLSTM-CRF, BioBERTpt) and QA (GPT-4o, Gemini 1.5 Flash, Llama 3 70B) with prompt engineering. Gemini 1.5 Flash achieved the highest evaluation metrics (macro-F1: 0.58; weighted-F1: 0.67). Results demonstrate a scalable, adaptable pathway for automating radiology workflows in non-English clinical settings, enhancing diagnostic efficiency through structured NLP post-processing.

Referências

Alsentzer et al. (2019). Publicly available clinical bert embeddings.

Bedi, Jain and Shah (2024). Evaluating the clinical benefits of llms. Nature Medicine.

Beyer et al. (2017). Automatic Lung-RADS™ classification with a natural language processing system. J Thorac Dis, 9(9):3114–3122.

Brown et al. (2020). Language models are few-shot learners.

Chowdhery et al. (2022). Palm: Scaling language modeling with pathways.

da Rocha et al. (2023). Natural language processing to extract information from portuguese-language medical records. Data, 8(1).

Deffebach and Humphrey (2015). Lung cancer screening. Surg Clin North Am, 95(5):967–978.

Devlin et al. (2019). Bert: Pre-training of deep bidirectional transformers for language understanding.

Fei et al. (2022). Quality management of pulmonary nodule radiology reports based on natural language processing. Bioengineering (Basel), 9(6).

Ferreira, Oliveira and De Almeida Vieira (2023). Lung-rads + ai: A tool for quantifying the risk of lung cancer in computed tomography reports. In 2023 IEEE 23rd International Conference on Bioinformatics and Bioengineering (BIBE), pages 292–297.

Gershanik, Lacson and Khorasani (2011). Critical finding capture in the impression section of radiology reports. AMIA Annu Symp Proc, 2011:465–469.

Gu et al. (2021). Domain-specific language model pretraining for biomedical natural language processing. ACM Transactions on Computing for Healthcare, 3(1):1–23.

Hirschman and Gaizauskas (2001). Natural language question answering: The view from here. Natural Language Engineering, 7:275 – 300.

Hu et al. (2024a). Zero-shot information extraction from radiological reports using chat-gpt. International Journal of Medical Informatics, 183:105321.

Hu et al. (2024b). Improving large language models for clinical named entity recognition via prompt engineering.

Kreimeyer et al. (2017). Natural language processing systems for capturing and standardizing unstructured clinical information: A systematic review. Journal of Biomedical Informatics, 73:14–29.

Lee et al. (2019). BioBERT: a pre-trained biomedical language representation model for biomedical text mining. Bioinformatics, 36(4):1234–1240.

Li et al. (2022). A survey on deep learning for named entity recognition. IEEE Transactions on Knowledge and Data Engineering, 34(1):50–70.

MacMahon et al. (2017). Guidelines for management of incidental pulmonary nodules detected on CT images: From the fleischner society 2017. Radiology, 284(1):228–243.

Mathias et al. (2020). Lung cancer in brazil. Journal of Thoracic Oncology, 15(2):170–175.

Nakayama et al. (2018). doccano: Text annotation tool for human. Software available from [link].

National Lung Screening Trial Research Team (2019). Lung cancer incidence and mortality with extended follow-up in the national lung screening trial. J Thorac Oncol, 14(10):1732–1742.

Nobel et al. (2020). Natural language processing in dutch free text radiology reports: Challenges in a small language area staging pulmonary oncology. Journal of Digital Imaging, 33(4):1002–1008.

of Radiology (2022). Lung-rads® v2022. [link]. Accessed: 2023-05-01.

OpenAI et al. (2024). Gpt-4 technical report.

Pandey et al. (2020). Extraction of radiographic findings from unstructured thoracoabdominal computed tomography reports using convolutional neural network based natural language processing. PLoS One, 15(7):e0236827.

Rainio, Teuho and Klén (2024). Evaluation metrics and statistical tests for machine learning. Scientific Reports, 14(1):6086.

Reimers and Gurevych (2020). Making monolingual sentence embeddings multilingual using knowledge distillation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.

Schneider et al. (2020). BioBERTpt - a Portuguese neural language model for clinical named entity recognition. In Rumshisky et al., editors, Proceedings of the 3rd Clinical Natural Language Processing Workshop, pages 65–72, Online. Association for Computational Linguistics.

Siegel et al. (2023). Cancer statistics, 2023. CA Cancer J Clin, 73(1):17–48.

Singhal et al. (2023). Towards expert-level medical question answering with large language models.

Vaswani et al. (2023). Attention is all you need.

Wu et al. (2023). Pmc-llama: Towards building open-source language models for medicine.

Zheng et al. (2021). Natural language processing to identify pulmonary nodules and extract nodule characteristics from radiology reports. Chest, 160(5):1902–1914.