{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,30]],"date-time":"2026-06-30T10:48:14Z","timestamp":1782816494819,"version":"3.54.5"},"reference-count":62,"publisher":"Association for Computing Machinery (ACM)","issue":"3","license":[{"start":{"date-parts":[[2026,5,16]],"date-time":"2026-05-16T00:00:00Z","timestamp":1778889600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/legalcode"}],"content-domain":{"domain":["dl.acm.org"],"crossmark-restriction":true},"short-container-title":["ACM Trans. Comput. Healthcare"],"published-print":{"date-parts":[[2026,7,30]]},"abstract":"<jats:p>\n                    Radiology impression generation involves producing concise, clinically meaningful summaries from detailed imaging findings such as CT and MRI scans, serving as a critical aid in diagnosis and treatment planning. However, recent studies highlight a severe shortage of radiologists, particularly in low and middle-income countries, where there is fewer than one radiologist per 100,000 people, making timely expert interpretation a significant challenge. While advancements in AI, especially large language models (LLMs), offer promising potential to automate this task, current systems often suffer from hallucinations, omissions of key clinical details, and a lack of linguistic clarity, thereby raising serious concerns about their safety and reliability in real-world clinical settings. In this work, we attempted to address this issue by introducing\n                    <jats:italic toggle=\"yes\">RADO<\/jats:italic>\n                    , a novel framework for radiology impression generation that integrates safety, faithfulness, and linguistic refinement rewards for preference optimization. To support robust evaluation, we introduce\n                    <jats:italic toggle=\"yes\">RIB<\/jats:italic>\n                    , a real-world benchmark dataset curated and annotated by radiologists, spanning 1,429 annotated CT and MRI findings and impressions across 27 study types.\n                    <jats:italic toggle=\"yes\">RADO<\/jats:italic>\n                    enforces critical safety and factuality constraints via carefully designed reward models and achieves state-of-the-art performance across multiple automatic and human evaluation metrics. Our framework significantly outperforms existing baselines, demonstrating improved factual consistency, reduced omissions, and higher clinical relevance, thus advancing the safety and reliability of generative AI in high-stakes medical applications. The code and dataset associated with the work are made available at\n                    <jats:ext-link xmlns:xlink=\"http:\/\/www.w3.org\/1999\/xlink\" ext-link-type=\"uri\" xlink:href=\"https:\/\/github.com\/AkashGhosh\/RADO-Trustworthy-Radiology-Impression-Generation\">RADO<\/jats:ext-link>\n                    .\n                  <\/jats:p>","DOI":"10.1145\/3805803","type":"journal-article","created":{"date-parts":[[2026,3,30]],"date-time":"2026-03-30T14:40:53Z","timestamp":1774881653000},"page":"1-17","update-policy":"https:\/\/doi.org\/10.1145\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["<i>RADO<\/i>\n                    : Trustworthy Radiology Impression Generation Using Safety and Faithfulness-Based Preference Optimization"],"prefix":"10.1145","volume":"7","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-9353-7258","authenticated-orcid":false,"given":"Akash","family":"Ghosh","sequence":"first","affiliation":[{"name":"Indian Institute of Technology Patna, Patna, India"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0004-9002-0837","authenticated-orcid":false,"given":"Nishant","family":"Kumar","sequence":"additional","affiliation":[{"name":"Indian Institute of Technology Patna, Patna, India"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0009-3891-2121","authenticated-orcid":false,"given":"Nitesh","family":"Patnaik","sequence":"additional","affiliation":[{"name":"Kalinga Institute of Industrial Technology, Bhubaneswar, India"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-8791-503X","authenticated-orcid":false,"given":"Adity","family":"Prakash","sequence":"additional","affiliation":[{"name":"GenX Diagnostics, Bhubaneswar, India"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6762-3892","authenticated-orcid":false,"given":"Rishi","family":"Raj","sequence":"additional","affiliation":[{"name":"Indian Institute of Management Visakhapatnam, Visakhapatnam, India"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-5458-9381","authenticated-orcid":false,"given":"Sriparna","family":"Saha","sequence":"additional","affiliation":[{"name":"Indian Institute of Technology Patna, Patna, India"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"320","published-online":{"date-parts":[[2026,5,16]]},"reference":[{"key":"e_1_3_3_2_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.findings-acl.161"},{"key":"e_1_3_3_3_2","unstructured":"Marah Abdin Jyoti Aneja Hany Awadalla Ahmed Awadallah Ammar Ahmad Awan Nguyen Bach Amit Bahree Arash Bakhtiari Jianmin Bao Harkirat Behl et al. 2024. Phi-3 technical report: A highly capable language model locally on your phone. arXiv:2404.14219. Retrieved from https:\/\/arxiv.org\/abs\/2404.14219"},{"key":"e_1_3_3_4_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.acl-long.662"},{"key":"e_1_3_3_5_2","unstructured":"Yuntao Bai Andy Jones Kamal Ndousse Amanda Askell Anna Chen Nova DasSarma Dawn Drain Stanislav Fort Deep Ganguli Tom Henighan et al. 2022. Training a helpful and harmless assistant with reinforcement learning from human feedback. arXiv:2204.05862. Retrieved from https:\/\/arxiv.org\/abs\/2204.05862"},{"key":"e_1_3_3_6_2","doi-asserted-by":"publisher","DOI":"10.1093\/nar\/gkh061"},{"key":"e_1_3_3_7_2","doi-asserted-by":"publisher","DOI":"10.1023\/A:1010933404324"},{"key":"e_1_3_3_8_2","doi-asserted-by":"publisher","unstructured":"Evan Calabrese Javier E. Villanueva-Meyer Jeffrey D. Rudie Andreas M. Rauschecker Ujjwal Baid Spyridon Bakas Soonmee Cha John T. Mongan and Christopher P. Hess. 2022. The University of California San Francisco Preoperative Diffuse Glioma MRI (UCSF-PDGM). DOI: 10.7937\/tcia.bdgf-8v37","DOI":"10.7937\/tcia.bdgf-8v37"},{"key":"e_1_3_3_9_2","article-title":"Deep reinforcement learning from human preferences","volume":"30","author":"Christiano Paul F.","year":"2017","unstructured":"Paul F. Christiano, Jan Leike, Tom Brown, Miljan Martic, Shane Legg, and Dario Amodei. 2017. Deep reinforcement learning from human preferences. In Advances in Neural Information Processing Systems, Vol. 30.","journal-title":"Advances in Neural Information Processing Systems"},{"issue":"70","key":"e_1_3_3_10_2","first-page":"1","article-title":"Scaling instruction-finetuned language models","volume":"25","author":"Chung Hyung Won","year":"2024","unstructured":"Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. 2024. Scaling instruction-finetuned language models. Journal of Machine Learning Research 25, 70 (2024), 1\u201353.","journal-title":"Journal of Machine Learning Research"},{"key":"e_1_3_3_11_2","unstructured":"d4data. 2025. Biomedical NER\u2014Recognizing 107 Biomedical Entities. Hugging Face. Retrieved from https:\/\/huggingface.co\/d4data\/biomedical-ner-all"},{"key":"e_1_3_3_12_2","doi-asserted-by":"publisher","DOI":"10.1093\/jamia\/ocv080"},{"key":"e_1_3_3_13_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2021.emnlp-main.594"},{"key":"e_1_3_3_14_2","unstructured":"Abhimanyu Dubey Abhinav Jauhri Abhinav Pandey Abhishek Kadian Ahmad Al-Dahle Aiesha Letman Akhil Mathur Alan Schelten Amy Yang Angela Fan et al. 2024. The llama 3 herd of models. arXiv:2407.21783. Retrieved from https:\/\/arxiv.org\/abs\/2407.21783"},{"key":"e_1_3_3_15_2","doi-asserted-by":"publisher","DOI":"10.1037\/h0031619"},{"key":"e_1_3_3_16_2","doi-asserted-by":"crossref","unstructured":"Soumya Suvra Ghosal Vaibhav Singh Akash Ghosh Soumyabrata Pal Subhadip Baidya Sriparna Saha and Dinesh Manocha. 2025. Relic: Enhancing reward model generalization for low-resource Indic languages with few-shot examples. arXiv:2506.16502. Retrieved from https:\/\/arxiv.org\/abs\/2506.16502","DOI":"10.18653\/v1\/2025.findings-emnlp.79"},{"key":"e_1_3_3_17_2","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v38i20.30206"},{"key":"e_1_3_3_18_2","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-56069-9_8"},{"key":"e_1_3_3_19_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-emnlp.675"},{"key":"e_1_3_3_20_2","doi-asserted-by":"publisher","DOI":"10.1002\/widm.70045"},{"key":"e_1_3_3_21_2","unstructured":"Akash Ghosh Srivarshinee Sridhar Raghav Kaushik Ravi Muhsin Muhsin Sriparna Saha and Chirag Agarwal. 2025. CLINIC: Evaluating multilingual trustworthiness in language models for healthcare. arXiv:2512.11437. Retrieved from https:\/\/arxiv.org\/abs\/2512.11437"},{"key":"e_1_3_3_22_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.acl-long.708"},{"key":"e_1_3_3_23_2","doi-asserted-by":"publisher","DOI":"10.1109\/TAI.2021.3086435"},{"key":"e_1_3_3_24_2","unstructured":"Danqing Hu Shanyuan Zhang Qing Liu Xiaofeng Zhu and Bing Liu. 2024. The current status of large language models in summarizing radiology report impressions. arXiv:2406.02134. Retrieved from https:\/\/arxiv.org\/abs\/2406.02134"},{"key":"e_1_3_3_25_2","unstructured":"Jinpeng Hu Jianling Li Zhihong Chen Yaling Shen Yan Song Xiang Wan and Tsung-Hui Chang. 2021. Word graph guided summarization for radiology findings. arXiv:2112.09925. Retrieved from https:\/\/arxiv.org\/abs\/2112.09925"},{"key":"e_1_3_3_26_2","unstructured":"J. Hu Z. Li Z. Chen Z. Li X. Wan and T. H. Chang. 2023. Graph enhanced contrastive learning for radiology findings summarization. arXiv:2204.00203. Retrieved from https:\/\/arxiv.org\/abs\/2204.00203"},{"key":"e_1_3_3_27_2","unstructured":"Zhiheng Huang Wei Xu and Kai Yu. 2015. Bidirectional LSTM-CRF models for sequence tagging. arXiv:1508.01991. Retrieved from https:\/\/arxiv.org\/abs\/1508.01991"},{"key":"e_1_3_3_28_2","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-88717-8_17"},{"key":"e_1_3_3_29_2","doi-asserted-by":"publisher","DOI":"10.1016\/j.jbi.2026.105009"},{"key":"e_1_3_3_30_2","volume-title":"SIIM Annual Meeting","author":"Jani Karan H.","year":"2023","unstructured":"Karan H. Jani and Anup Shetty. 2023. Retrieval augmented generation with large language models: A leap toward clinical information assistants in radiology. SIIM Annual Meeting. Retrieved from https:\/\/annualmeeting.siim.org\/wp-content\/uploads\/2024\/04\/38_Jani_Retrieval-Augmented-Generation-with-Large-Language-Models.pdf"},{"key":"e_1_3_3_31_2","doi-asserted-by":"publisher","DOI":"10.1038\/s41597-019-0322-0"},{"key":"e_1_3_3_32_2","unstructured":"Zachary Kenton Tom Everitt Laura Weidinger Iason Gabriel Vladimir Mikulik and Geoffrey Irving. 2021. Alignment of language agents. arXiv:2103.14659. Retrieved from https:\/\/arxiv.org\/abs\/2103.14659"},{"key":"e_1_3_3_33_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2020.emnlp-main.750"},{"key":"e_1_3_3_34_2","doi-asserted-by":"publisher","DOI":"10.1162\/tacl_a_00453"},{"key":"e_1_3_3_35_2","first-page":"74","volume-title":"Text Summarization Branches Out","author":"Lin Chin-Yew","year":"2004","unstructured":"Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. In Text Summarization Branches Out, 74\u201381."},{"key":"e_1_3_3_36_2","unstructured":"Zhengliang Liu Yiwei Li Peng Shu Aoxiao Zhong Longtao Yang Chao Ju Zihao Wu Chong Ma Jie Luo Cheng Chen et al. 2023. Radiology-Llama2: Best-in-class large language model for radiology. arXiv:2309.06419. Retrieved from https:\/\/arxiv.org\/abs\/2309.06419"},{"key":"e_1_3_3_37_2","unstructured":"Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. arXiv:1711.05101. Retrieved from https:\/\/arxiv.org\/abs\/1711.05101"},{"key":"e_1_3_3_38_2","doi-asserted-by":"publisher","DOI":"10.1109\/TAI.2024.3364586"},{"key":"e_1_3_3_39_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2022.acl-long.15"},{"key":"e_1_3_3_40_2","unstructured":"Eric Onyame Akash Ghosh Subhadip Baidya Sriparna Saha Xiuying Chen and Chirag Agarwal. 2026. CURE-Med: Curriculum-informed reinforcement learning for multilingual medical reasoning. arXiv:2601.13262. Retrieved from https:\/\/arxiv.org\/abs\/2601.13262"},{"key":"e_1_3_3_41_2","doi-asserted-by":"publisher","DOI":"10.52202\/068431-2011"},{"key":"e_1_3_3_42_2","article-title":"Pytorch: An imperative style, high-performance deep learning library","volume":"32","author":"Paszke Adam","year":"2019","unstructured":"Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. Pytorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems, Vol. 32.","journal-title":"Advances in Neural Information Processing Systems"},{"key":"e_1_3_3_43_2","article-title":"Direct preference optimization: Your language model is secretly a reward model","volume":"36","author":"Rafailov Rafael","year":"2024","unstructured":"Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning, Stefano Ermon, and Chelsea Finn. 2024. Direct preference optimization: Your language model is secretly a reward model. In Advances in Neural Information Processing Systems, Vol. 36.","journal-title":"Advances in Neural Information Processing Systems"},{"key":"e_1_3_3_44_2","unstructured":"Pranab Sahoo Prabhash Meharia Akash Ghosh Sriparna Saha Vinija Jain and Aman Chadha. 2024. Unveiling hallucination in text image video and audio foundation models: A comprehensive review. arXiv:2405.09589. Retrieved from https:\/\/arxiv.org\/abs\/2405.09589"},{"key":"e_1_3_3_45_2","doi-asserted-by":"publisher","DOI":"10.1001\/archinternmed.2009.333"},{"key":"e_1_3_3_46_2","unstructured":"John Schulman Filip Wolski Prafulla Dhariwal Alec Radford and Oleg Klimov. 2017. Proximal policy optimization algorithms. arXiv:1707.06347. Retrieved from https:\/\/arxiv.org\/abs\/1707.06347"},{"key":"e_1_3_3_47_2","doi-asserted-by":"publisher","DOI":"10.1186\/s12880-024-01435-w"},{"key":"e_1_3_3_48_2","doi-asserted-by":"publisher","DOI":"10.1148\/radiol.231259"},{"key":"e_1_3_3_49_2","doi-asserted-by":"publisher","DOI":"10.1007\/s10278-024-00985-3"},{"key":"e_1_3_3_50_2","unstructured":"Unsloth. 2024. Unsloth: Accelerate Your Machine Learning Projects. Retrieved October 11 2024 from https:\/\/unsloth.ai\/."},{"key":"e_1_3_3_51_2","unstructured":"Hanyin Wang Chufan Gao and Bolun Liu. 2024. Adapting open-source large language models for cost-effective expert-level clinical note generation with on-policy reinforcement learning. arXiv:2405.00715. Retrieved from https:\/\/arxiv.org\/abs\/2405.00715"},{"key":"e_1_3_3_52_2","doi-asserted-by":"crossref","unstructured":"Xiaosong Wang Yifan Peng Le Lu Zhiyong Lu Mohammadhadi Bagheri and Ronald M. Summers. 2017. ChestX-ray8: Hospital-scale chest X-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases. arXiv:1705.02315. Retrieved from https:\/\/arxiv.org\/abs\/1705.02315","DOI":"10.1109\/CVPR.2017.369"},{"key":"e_1_3_3_53_2","unstructured":"Benjamin Warner Antoine Chaffin Benjamin Clavi\u00e9 Orion Weller Oskar Hallstr\u00f6m Said Taghadouini Alexis Gallagher Raja Biswas Faisal Ladhak Tom Aarsen et al. 2024. Smarter better faster longer: A modern bidirectional encoder for fast memory efficient and long context finetuning and inference. arXiv:2412.13663. Retrieved from https:\/\/arxiv.org\/abs\/2412.13663"},{"key":"e_1_3_3_54_2","unstructured":"Thomas Wolf Lysandre Debut Victor Sanh Julien Chaumond Clement Delangue Anthony Moi Pierric Cistac Tim Rault R\u00e9mi Louf Morgan Funtowicz et al. 2019. Huggingface\u2019s transformers: State-of-the-art natural language processing. arXiv:1910.03771. Retrieved from https:\/\/arxiv.org\/abs\/1910.03771"},{"key":"e_1_3_3_55_2","doi-asserted-by":"publisher","DOI":"10.1109\/BigData59044.2023.10386916"},{"key":"e_1_3_3_56_2","unstructured":"An Yang Baosong Yang Beichen Zhang Binyuan Hui Bo Zheng Bowen Yu Chengyuan Li Dayiheng Liu Fei Huang Haoran Wei et al. 2024. Qwen2.5 technical report. arXiv:2412.15115. Retrieved from https:\/\/arxiv.org\/abs\/2412.15115"},{"key":"e_1_3_3_57_2","unstructured":"Fang Zeng Zhiliang Lyu Quanzheng Li and Xiang Li. 2024. Enhancing LLMs for impression generation in radiology reports through a multi-agent system. arXiv:2412.06828. Retrieved from https:\/\/arxiv.org\/abs\/2412.06828"},{"key":"e_1_3_3_58_2","doi-asserted-by":"publisher","DOI":"10.1148\/radiol.240885"},{"key":"e_1_3_3_59_2","doi-asserted-by":"crossref","unstructured":"Nan Zhang Yusen Zhang Wu Guo Prasenjit Mitra and Rui Zhang. 2023. FaMeSumm: Investigating and improving faithfulness of medical summarization. arXiv:2311.02271. Retrieved from https:\/\/arxiv.org\/abs\/2311.02271","DOI":"10.18653\/v1\/2023.emnlp-main.673"},{"key":"e_1_3_3_60_2","volume-title":"Proceedings of the International Conference on Learning Representations (ICLR)","author":"Zhang Tianyi","year":"2020","unstructured":"Tianyi Zhang, Varsha Kishore, Felix Wu, Kilian Q. Weinberger, and Yoav Artzi. 2020. BERTScore: Evaluating text generation with BERT. In Proceedings of the International Conference on Learning Representations (ICLR). Retrieved from https:\/\/openreview.net\/forum?id=SkeHuCVFDr."},{"key":"e_1_3_3_61_2","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/W18-5623"},{"key":"e_1_3_3_62_2","unstructured":"Yuhao Zhang Derek Merck Emily Bao Tsai Christopher D. Manning and Curtis P. Langlotz. 2019. Optimizing the factual correctness of a summary: A study of summarizing radiology reports. arXiv:1911.02541. Retrieved from https:\/\/arxiv.org\/abs\/1911.02541"},{"key":"e_1_3_3_63_2","first-page":"15763","article-title":"Consequences of misaligned AI","volume":"33","author":"Zhuang Simon","year":"2020","unstructured":"Simon Zhuang and Dylan Hadfield-Menell. 2020. Consequences of misaligned AI. In Advances in Neural Information Processing Systems, Vol. 33, 15763\u201315773.","journal-title":"Advances in Neural Information Processing Systems"}],"container-title":["ACM Transactions on Computing for Healthcare"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3805803","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,30]],"date-time":"2026-06-30T10:36:42Z","timestamp":1782815802000},"score":1,"resource":{"primary":{"URL":"https:\/\/dl.acm.org\/doi\/10.1145\/3805803"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,5,16]]},"references-count":62,"journal-issue":{"issue":"3","published-print":{"date-parts":[[2026,7,30]]}},"alternative-id":["10.1145\/3805803"],"URL":"https:\/\/doi.org\/10.1145\/3805803","relation":{},"ISSN":["2691-1957","2637-8051"],"issn-type":[{"value":"2691-1957","type":"print"},{"value":"2637-8051","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,5,16]]},"assertion":[{"value":"2025-07-18","order":0,"name":"received","label":"Received","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-03-13","order":2,"name":"accepted","label":"Accepted","group":{"name":"publication_history","label":"Publication History"}},{"value":"2026-05-16","order":3,"name":"published","label":"Published","group":{"name":"publication_history","label":"Publication History"}}]}}