請用此 Handle URI 來引用此文件：
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/104284完整後設資料紀錄
| DC 欄位 | 值 | 語言 |
|---|---|---|
| dc.contributor.advisor | 莊裕澤 | zh_TW |
| dc.contributor.advisor | Yuh-Jzer Joung | en |
| dc.contributor.author | 林怡君 | zh_TW |
| dc.contributor.author | Yi-Chun Lin | en |
| dc.date.accessioned | 2026-08-25T16:21:20Z | - |
| dc.date.available | 2026-08-26 | - |
| dc.date.copyright | 2026-08-25 | - |
| dc.date.issued | 2026 | - |
| dc.date.submitted | 2026-08-19 17:44:16 | - |
| dc.identifier.citation | [1] S. Bannur, K. Bouzid, D. C. Castro, et al. Maira-2: Grounded radiology report generation, 2024. arXiv preprint arXiv:2406.04449.
[2] S. Bannur, S. Hyland, Q. Liu, et al. Learning to exploit temporal structure for biomedical vision-language processing. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023. [3] N. Deperrois, H. Matsuo, S. Ruipérez-Campillo, et al. Radvlm: A multitask conversational vision-language model for radiology. arXiv preprint arXiv:2502.03333, 2025. [4] A. Grattafiori, A. Dubey, A. Jauhri, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024. [5] A. Heiman, X. Zhang, E. Chen, et al. Factchexcker: Mitigating measurement hallucinations in chest x-ray report generation models. arXiv preprint arXiv:2411.18672, 2025. [6] J. Irvin, P. Rajpurkar, M. Ko, et al. Chexpert: A large dataset of chest radiographs with uncertainty labels and expert comparison. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 590–597, 2019. [7] S. Jain, A. Agrawal, A. Saporta, et al. Radgraph: Extracting clinical entities and relations from radiology reports. arXiv preprint arXiv:2106.14463, 2021. [8] S. Jain, A. Agrawal, A. Saporta, et al. RadGraph: Extracting Clinical Entities and Relations from Radiology Reports. PhysioNet, June 2021. Version 1.0.0. [9] A. Johnson, M. Lungren, Y. Peng, et al. MIMIC-CXR-JPG - chest radiographs with structured labels. PhysioNet, Mar. 2024. Version 2.1.0. [10] A. E. W. Johnson, T. J. Pollard, R. G. Mark, et al. MIMIC-CXR Database, 2024. RRID:SCR_007345. [11] K. Kale, P. Bhattacharyya, M. Gune, et al. KGVL-BART: Knowledge graph augmented visual language BART for radiology report generation. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 3401–3411, Dubrovnik, Croatia, May 2023. Association for Computational Linguistics. [12] C. Li, C. Wong, S. Zhang, et al. Llava-med: Training a large language-and-vision assistant for biomedicine in one day. arXiv preprint arXiv:2306.00890, 2023. [13] Y. Li, R. M. Wehbe, F. S. Ahmad, et al. Clinical-longformer and clinical-bigbird: Transformers for long clinical sequences. arXiv preprint arXiv:2201.11838, 2022. [14] F. Liu, E. Shareghi, Z. Meng, et al. Self-alignment pretraining for biomedical entity representations. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4228–4238, 2021. [15] K. Liu, Z. Ma, X. Kang, et al. Enhanced contrastive learning with multi-view longitudinal data for chest x-ray report generation. In 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10348–10359. IEEE, June 2025. [16] R. Mahmood, D. Machado Reyes, G. Wang, et al. Fact-checking of ai-generated reports, 2025. arXiv preprint arXiv:2307.14634. [17] R. Mahmood, D. Machado-Reyes, J. Wu, et al. Phrase-grounded fact-checking for automatically generated chest x-ray reports. arXiv preprint arXiv:2509.21356, 2025. [18] R. Mahmood and T. F. Syeda-Mahmood. Inference-time correction of errors in AI-generated chest x-ray radiology reports, 2025. [19] R. Mahmood, K. C. L. Wong, D. Machado-Reyes, et al. Anatomically-grounded fact checking of automated chest x-ray reports. arXiv preprint arXiv:2412.02177, 2024. [20] R. Mahmood, P. Yan, and T. F. Syeda-Mahmood. Automatic correction of AI reports using fact-checking model-guided LLMs. In The Second Workshop on GenAI for Health: Potential, Trust, and Policy Compliance, 2025. [21] S. Ostmeier, J.-B. Jarmis, M. Purkayastha, et al. Green: Generative radiology report evaluation and error notation. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 343–356, 2024. [22] S.-J. Park, K.-S. Heo, D.-H. Shin, et al. Dart: Disease-aware image-text alignment and self-correcting re-alignment for trustworthy radiology report generation. arXiv preprint arXiv:2504.11786, 2025. [23] C. Pellegrini, E. Özsoy, B. Busam, et al. Radialog: A large vision-language model for radiology report generation and conversational assistance. arXiv preprint arXiv:2311.18681, 2025. [24] A. Radford, J. W. Kim, C. Hallacy, et al. Learning transferable visual models from natural language supervision. In Proceedings of the 38th International Conference on Machine Learning (ICML), 2021. [25] V. Ramesh, N. A. Chi, and P. Rajpurkar. Improving radiology report generation systems by removing hallucinated references to non-existent priors. arXiv preprint arXiv:2210.06340, 2022. [26] A. Sellergren, S. Kazemzadeh, T. Jaroensri, et al. Medgemma technical report. arXiv preprint arXiv:2507.05201, 2025. [27] L. Sun, J. Zhao, M. Han, et al. Fact-aware multimodal retrieval augmentation for accurate medical radiology report generation. arXiv preprint arXiv:2407.15268, 2025. [28] H. Warr, Y. Ibrahim, D. R. McGowan, et al. Quality control for radiology report generation models via auxiliary auditing components. arXiv preprint arXiv:2407.21638, 2024. [29] J. Wu, Z. Wu, R. Li, et al. Error correction in radiology reports: A knowledge distillation-based multi-stage framework. arXiv preprint arXiv:2406.15045, 2025. [30] F. Yu, M. Endo, R. Krishnan, et al. Evaluating progress in automatic chest x-ray radiology report generation. Patterns, 4(9):100802, 2023. | - |
| dc.identifier.uri | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/104284 | - |
| dc.description.abstract | 視覺語言模型（Vision-Language Models, VLMs）在自動化胸腔 X 光（Chest X-ray, CXR）報告生成任務上已展現出良好潛力，然而其臨床可靠性仍受限於關鍵發現遺漏與幻覺生成等問題。本文提出一種輕量化的報告校正框架，透過基於圖結構的檢索機制與問題導向的推理流程，以提升胸腔 X 光報告的準確性與完整性。首先，將由 VLM 生成的初始報告轉換為結構化圖表示，並據此自放射科關聯圖譜知識圖譜中檢索相關臨床關係。接著，利用檢索所得之關係生成具針對性的問題，引導視覺語言模型進行視覺推理，並逐步精煉原始報告內容，以降低遺漏現象。 | zh_TW |
| dc.description.abstract | Vision-Language Models (VLMs) have demonstrated promising potential in automating Chest X-ray (CXR) report generation tasks; however, their clinical reliability remains limited by issues such as missing key findings and hallucination generation. This paper proposes a lightweight report correction framework that enhances the accuracy and completeness of chest X-ray reports through a graph-based retrieval mechanism and a question-oriented reasoning process. First, the initial report generated by the VLM is converted into a structured graph representation, and relevant clinical relationships are then retrieved from Radiology Association Graph. Subsequently, targeted questions are generated using the retrieved relationships to guide the VLM in visual reasoning and progressively refine the original report content to reduce omissions. | en |
| dc.description.provenance | Submitted by admin ntu (admin@lib.ntu.edu.tw) on 2026-08-25T16:21:20Z No. of bitstreams: 0 | en |
| dc.description.provenance | Made available in DSpace on 2026-08-25T16:21:20Z (GMT). No. of bitstreams: 0 | en |
| dc.description.tableofcontents | 致謝 i
摘要 iii Abstract v 目次 vii 圖次 xi 表次 xiii 第一章 緒論 1 1.1 研究背景與動機 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 研究目的與貢獻 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 第二章 文獻探討 5 2.1 醫療視覺語言模型 . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.2 放射科報告自動生成研究現況 . . . . . . . . . . . . . . . . . . . . 7 2.3 報告校正方法 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.3.1 基於事實核查的過濾型校正（Fact-Checking based Correction） . 9 2.3.2 修正報告中的測量值 . . . . . . . . . . . . . . . . . . . . . . . . 10 2.4 知識圖譜於醫療文本的應用 . . . . . . . . . . . . . . . . . . . . . 11 2.5 總結 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 第三章 研究方法 15 3.1 系統設計動機 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 3.2 系統架構總覽 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 3.3 放射科關聯圖譜建構 . . . . . . . . . . . . . . . . . . . . . . . . . 19 3.3.1 資料來源與 RadGraph 抽取結果 . . . . . . . . . . . . . . . . . . 19 3.3.2 圖資料前處理 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 3.3.2.1 比較性描述過濾（Comparative Statement Filtering） 20 3.3.2.2 Entity Token 前處理與詞彙正規化 . . . . . . . . . . 21 3.3.2.3 節點角色標記（Entity Role Feature Assignment） . . 23 3.3.3 節點與關係去重（Node & Relation Deduplication） . . . . . . . . 24 3.3.4 Supernode 建構 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 3.3.5 SUGGESTIVE_OF 邊權重計算 . . . . . . . . . . . . . . . . . . . 28 3.3.5.1 條件機率權重計算（Conditional Probability Weight Computation） . . . . . 28 3.3.5.2 NPMI 權重計算（NPMI Edge Weight Computation） 29 3.3.6 問題生成與寫入 . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 3.3.7 圖譜規模 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.4 推論時執行流程（Inference-time Pipeline） . . . . . . . . . . . . . 33 3.4.1 初步報告生成與結構化抽取 . . . . . . . . . . . . . . . . . . . . 33 3.4.2 知識圖譜檢索 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3.4.2.1 共用前置步驟 . . . . . . . . . . . . . . . . . . . . . . 35 3.4.2.2 策略一：單層展開與固定 Top-K 篩選 . . . . . . . . 40 3.4.2.3 策略二：兩層展開與逐層自適應篩選 . . . . . . . . 41 3.4.3 視覺問答推論 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 3.4.4 最終報告整合生成 . . . . . . . . . . . . . . . . . . . . . . . . . . 45 第四章 實驗過程與成果評估 51 4.1 資料集 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.2 評估指標 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4.2.1 RadGraph F1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4.2.2 CheXBert F1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4.2.3 GREEN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.3 實驗設定與版本管理 . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.3.1 Pipeline 版本設計 . . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.3.2 各版本實驗結果概覽 . . . . . . . . . . . . . . . . . . . . . . . . 57 4.4 實驗結果分析與討論 . . . . . . . . . . . . . . . . . . . . . . . . . 57 4.4.1 版本效能差異成因分析 . . . . . . . . . . . . . . . . . . . . . . . 58 4.4.2 三指標方向分歧 . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 4.4.2.1 案例一：RadGraph 下降，CheXBert Micro 上升 . . . 61 4.4.2.2 案例二：GREEN 下降，CheXBert Micro 大幅上升 . 64 4.4.3 Macro F1 與 Micro F1 提升幅度之不對稱性分析 . . . . . . . . . 66 4.4.4 量化指標退步的失敗模式分析 . . . . . . . . . . . . . . . . . . . 67 4.4.4.1 模式一：QA 步驟的不準確回答引入幻覺 . . . . . . 68 4.4.4.2 模式二：錯誤基礎引發的幻覺疊加 . . . . . . . . . . 69 4.4.5 知識注入格式對照實驗 . . . . . . . . . . . . . . . . . . . . . . . 69 4.4.6 綜合討論 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 4.5 延伸實驗與改善侷限分析 . . . . . . . . . . . . . . . . . . . . . . 75 4.5.1 改善動機與設計邏輯 . . . . . . . . . . . . . . . . . . . . . . . . 75 4.5.2 延伸實驗設計 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.5.3 實驗結果 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 4.5.4 失敗原因分析 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 第五章 結論 81 5.1 研究總結 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 5.2 研究限制 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 5.2.1 視覺問答步驟的準確度上限 . . . . . . . . . . . . . . . . . . . . 83 5.2.2 錯誤傳播的累積效應 . . . . . . . . . . . . . . . . . . . . . . . . 84 5.2.3 幻覺後處理的侷限 . . . . . . . . . . . . . . . . . . . . . . . . . . 84 5.3 未來研究方向 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 5.3.1 改善視覺問答步驟的準確度 . . . . . . . . . . . . . . . . . . . . 84 5.3.2 強化 Pipeline 的錯誤容忍機制 . . . . . . . . . . . . . . . . . . . 85 5.3.3 更精確的幻覺偵測與移除 . . . . . . . . . . . . . . . . . . . . . . 85 參考文獻 87 | - |
| dc.language.iso | zh_TW | - |
| dc.subject | 胸腔 X 光 | - |
| dc.subject | 報告生成 | - |
| dc.subject | 視覺語言模型 | - |
| dc.subject | 放射科關聯圖譜 | - |
| dc.subject | 圖式檢索 | - |
| dc.subject | 醫學影像人工智慧 | - |
| dc.subject | Chest X-ray | - |
| dc.subject | Report Generation | - |
| dc.subject | Vision-Language Models | - |
| dc.subject | Radiology Association Graph | - |
| dc.subject | Graph-based Retrieval | - |
| dc.subject | Medical Imaging AI | - |
| dc.title | 基於視覺語言模型與放射科關聯圖譜之胸腔 X 光報告生成校正系統 | zh_TW |
| dc.title | A Chest X-ray Report Refinement System Based on Vision–Language Models and Radiology Association Graph | en |
| dc.type | Thesis | - |
| dc.date.schoolyear | 114-2 | - |
| dc.description.degree | 碩士 | - |
| dc.contributor.oralexamcommittee | 楊立偉;簡睿哲;莊豐源 | zh_TW |
| dc.contributor.oralexamcommittee | Li-Wei Yang;Ruey-Jer Jean;Feng-Yuan Chuang | en |
| dc.subject.keyword | 胸腔 X 光; 報告生成; 視覺語言模型; 放射科關聯圖譜; 圖式檢索; 醫學影像人工智慧 | zh_TW |
| dc.subject.keyword | Chest X-ray; Report Generation; Vision-Language Models; Radiology Association Graph; Graph-based Retrieval; Medical Imaging AI | en |
| dc.relation.page | 90 | - |
| dc.identifier.doi | 10.6342/NTU202604189 | - |
| dc.rights.note | 未授權 | - |
| dc.date.accepted | 2026-08-20 | - |
| dc.contributor.author-college | 管理學院 | - |
| dc.contributor.author-dept | 資訊管理學系 | - |
| dc.date.embargo-lift | N/A | - |
| 顯示於系所單位： | 資訊管理學系 | |
文件中的檔案：
| 檔案 | 大小 | 格式 | |
|---|---|---|---|
| ntu-114-2.pdf 未授權公開取用 | 4.58 MB | Adobe PDF |
系統中的文件，除了特別指名其著作權條款之外，均受到著作權保護，並且保留所有的權利。