{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,29]],"date-time":"2026-09-29T04:59:50Z","timestamp":1790657990373,"version":"4.1.0"},"reference-count":41,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,29]],"date-time":"2026-09-29T00:00:00Z","timestamp":1790640000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Artif. Intell."],"abstract":"\n Introduction<\/jats:title>\n Retrieval-Augmented Generation (RAG) has been established as a reliable paradigm for grounding large language model (LLM) outputs in factual evidence, acting as a safeguard in high-stakes domains such as biomedicine, where hallucination carries significant consequences. However, existing biomedical RAG systems rely predominantly on text-only corpora, overlooking the rich semantic information encoded in figures, charts, and other visual evidence. In this work, we investigate whether incorporating figure-based retrieval methods into a hybrid RAG pipeline meaningfully improves generation quality and retrieval effectiveness in the biomedical domain.<\/jats:p>\n <\/jats:sec>\n \n Methods<\/jats:title>\n \n We propose Figure-First Hybrid RAG (FFHRAG), a multi-stage retrieval architecture integrating multi-channel figure retrieval with hybrid text retrieval. Our architecture employs a\n figure-first<\/jats:bold>\n discovery mechanism that retrieves relevant visual evidence, and utilizes it as a guiding signal for subsequent text selection. This strategy eliminates the need for complex embedding alignment, offering a\n low-overhead<\/jats:bold>\n system that avoids\n cross-modal embedding noise<\/jats:bold>\n at the final text chunk selection stage. The system is evaluated zero-shot on summary-type questions derived from BioASQ-13b, against a corpus spanning over 1.1 million PubMed Central text chunks and 82,000 figure-captions.\n <\/jats:p>\n <\/jats:sec>\n \n Results<\/jats:title>\n A progressive, multi-seed six-variant ablation study demonstrates that while the text-only baseline components establish a strong retrieval foundation\u2014BM25 retriever yielding a Recall@10 of 0.7322 \u00b1 0.0154 and the Cross-Encoder reranking securing an MRR of 0.8331 \u00b1 0.0018, our proposed framework introduces isolated, targeted enhancements. Specifically, the localized PMCID-level boosting module refines retrieval ordering to an MRR of 0.8358 \u00b1 0.0018, while prompt-level figure caption injection substantially enriches downstream generation quality, driving ROUGE-1 from 0.4068 \u00b1 0.0037 to 0.4137 \u00b1 0.0036 and ROUGE-L from 0.2623 \u00b1 0.0018 to 0.2678 \u00b1 0.0016 over the strong text-only baseline.<\/jats:p>\n <\/jats:sec>\n \n Discussion<\/jats:title>\n Our findings suggest that figures represent an underutilized retrieval modality in biomedical RAG systems, and that even lightweight figure-aware boosting mechanisms can produce measurable gains in answer quality without requiring multimodal fusion at the embedding level.<\/jats:p>\n <\/jats:sec>","DOI":"10.3389\/frai.2026.1861611","type":"journal-article","created":{"date-parts":[[2026,9,29]],"date-time":"2026-09-29T04:42:59Z","timestamp":1790656979000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Beyond text: quantifying the impact of figure-guided retrieval in biomedical RAG systems"],"prefix":"10.3389","volume":"9","author":[{"given":"Raghav Kishore","family":"Venkatesh","sequence":"first","affiliation":[{"name":"School of Electronics Engineering, Vellore Institute of Technology","place":["Chennai, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"G.","family":"Sudhakaran","sequence":"additional","affiliation":[{"name":"School of Electronics Engineering, Vellore Institute of Technology","place":["Chennai, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"C. U. Om","family":"Kumar","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Vellore Institute of Technology","place":["Chennai, India"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,29]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"e0000877","DOI":"10.1371\/journal.pdig.0000877","article-title":"Retrieval augmented generation (RAG) for large language models (LLMs) in healthcare: a systematic review","volume":"4","author":"Amugongo","year":"2025","journal-title":"PLoS Digit. Health"},{"key":"B2","doi-asserted-by":"publisher","first-page":"bbaf539","DOI":"10.1093\/bib\/bbaf539","article-title":"BioRAGent: natural language biomedical querying with retrieval-augmented multiagent systems","volume":"26","author":"Bi","year":"2025","journal-title":"Brief. Bioinform"},{"key":"B3","doi-asserted-by":"crossref","first-page":"335","DOI":"10.1145\/290941.291025","article-title":"\u201cThe use of MMR, diversity-based re-ranking for reordering documents and producing summaries,\u201d","volume-title":"Proceedings of the 21st annual international ACM SIGIR conference on research and development in information retrieval","author":"Carbonell","year":"1998"},{"key":"B4","doi-asserted-by":"publisher","first-page":"152682","DOI":"10.1109\/ACCESS.2025.3604351","article-title":"Benchmarking large language models for biomedical literature summarization: abstractive versus extractive paradigms","volume":"13","author":"Celikten","year":"2025","journal-title":"IEEE Access"},{"key":"B5","doi-asserted-by":"crossref","first-page":"758","DOI":"10.1145\/1571941.1572114","article-title":"\u201cReciprocal rank fusion outperforms Condorcet and individual rank learning methods,\u201d","volume-title":"Proceedings of the 32nd international ACM SIGIR conference on research and development in information retrieval","author":"Cormack","year":"2009"},{"key":"B6","doi-asserted-by":"publisher","first-page":"474","DOI":"10.26421\/JDI3.4-5","article-title":"Improved methods to aid unsupervised evidence-based fact checking for online health news","volume":"3","author":"Deka","year":"2022","journal-title":"J. Data Intell"},{"key":"B7","doi-asserted-by":"publisher","first-page":"28191","DOI":"10.1007\/s00521-025-11666-9","article-title":"A survey on retrieval-augmentation generation (RAG) models for healthcare applications","volume":"37","author":"El-Enen","year":"2025","journal-title":"Neural Comput. Appl"},{"key":"B8","article-title":"\u201cColPali: efficient document retrieval with vision language models,\u201d","volume-title":"Proceedings of the international conference on learning representations (ICLR)","author":"Faysse","year":"2025"},{"key":"B9","article-title":"\u201cEnhancing biomedical question answering with parameter-efficient fine-tuning and hierarchical retrieval augmented generation,\u201d","volume-title":"Proceedings of the conference and labs of the evaluation forum","author":"Gao","year":"2024"},{"key":"B10","doi-asserted-by":"publisher","first-page":"7612","DOI":"10.1093\/nar\/gky515","article-title":"Casein kinase 2 mediated phosphorylation of Spt6 modulates histone dynamics and regulates spurious transcription","volume":"46","author":"Gouot","year":"2018","journal-title":"Nucleic Acids Res"},{"key":"B11","doi-asserted-by":"publisher","first-page":"1270","DOI":"10.2337\/db12-0533","article-title":"Dissociation of Bcl-2-Beclin1 complex by activated AMPK enhances cardiac autophagy and protects against cardiomyocyte apoptosis in diabetes","volume":"62","author":"He","year":"2013","journal-title":"Diabetes"},{"key":"B12","unstructured":"MRAG-suite: a diagnostic evaluation platform for visual retrieval-augmented generation\n \n \n \n Ji\n Y.\n \n \n Lan\n W.\n \n \n Ng\n P.\n \n \n arXiv\n \n 2025"},{"key":"B13","doi-asserted-by":"crossref","first-page":"2567","DOI":"10.18653\/v1\/D19-1259","article-title":"\u201cPubMedQA: a dataset for biomedical research question answering,\u201d","volume-title":"Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP)","author":"Jin","year":"2019"},{"key":"B14","doi-asserted-by":"publisher","first-page":"btad651","DOI":"10.1093\/bioinformatics\/btad651","article-title":"Medcpt: Contrastive pre-trained transformers with large-scale pubmed search logs for zero-shot biomedical information retrieval","volume":"39","author":"Jin","year":"","journal-title":"Bioinformatics"},{"key":"B15","doi-asserted-by":"publisher","first-page":"104988","DOI":"10.1016\/j.ebiom.2024.104988","article-title":"PubMed and beyond: biomedical literature search in the age of artificial intelligence","volume":"100","author":"Jin","year":"2024","journal-title":"eBioMedicine"},{"key":"B16","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/3490238","article-title":"Biomedical question answering: a survey of approaches and challenges","volume":"55","author":"Jin","year":"","journal-title":"ACM Comput. Surv"},{"key":"B17","doi-asserted-by":"publisher","first-page":"160035","DOI":"10.1038\/sdata.2016.35","article-title":"MIMIC-III, a freely accessible critical care database","volume":"3","author":"Johnson","year":"2016","journal-title":"Sci. Data"},{"key":"B18","doi-asserted-by":"publisher","first-page":"535","DOI":"10.1109\/TBDATA.2019.2921572","article-title":"Billion-scale similarity search with GPUs","volume":"7","author":"Johnson","year":"2021","journal-title":"IEEE Trans. Big Data"},{"key":"B19","doi-asserted-by":"publisher","first-page":"89","DOI":"10.3390\/make7030089","article-title":"AlzheimerRAG: multimodal retrieval-augmented generation for clinical use cases","volume":"7","author":"Lahiri","year":"2025","journal-title":"Mach. Learn. Knowl. Extr"},{"key":"B20","first-page":"228","article-title":"\u201cMETEOR: an automatic metric for MT evaluation with high levels of correlation with human judgments,\u201d","volume-title":"Proceedings of the second workshop on statistical machine translation (WMT 2007)","author":"Lavie","year":"2007"},{"key":"B21","first-page":"74","article-title":"\u201cROUGE: a package for automatic evaluation of summaries,\u201d","volume-title":"Text Summarization Branches Out","author":"Lin","year":"2004"},{"key":"B22","doi-asserted-by":"publisher","first-page":"605","DOI":"10.1093\/jamia\/ocaf008","article-title":"Improving large language model applications in biomedicine with retrieval-augmented generation: a systematic review, meta-analysis, and clinical development guidelines","volume":"32","author":"Liu","year":"2025","journal-title":"J. Am. Med. Inform. Assoc"},{"key":"B23","first-page":"1108","article-title":"\u201cA hybrid approach for biomedical question answering: Combining sparse and dense retrieval with LLM-based reranking,\u201d","volume-title":"Proceedings of the international conference on recent advancements and modernisations in sustainable intelligent technologies and applications (RAMSITA 2025)","author":"Makhija","year":"2025"},{"key":"B24","unstructured":"PMC Open Access Subset\n \n 2003"},{"key":"B25","doi-asserted-by":"crossref","first-page":"145","DOI":"10.1007\/978-3-032-04354-2_12","article-title":"\u201cOverview of BioASQ 2025: the thirteenth BioASQ challenge on large-scale biomedical semantic indexing and question answering,\u201d","volume-title":"Experimental IR meets multilinguality, multimodality, and interaction","author":"Nentidis","year":"2026"},{"key":"B26","first-page":"311","article-title":"\u201cBLEU: a method for automatic evaluation of machine translation,\u201d","volume-title":"Proceedings of the 40th annual meeting of the association for computational linguistics (ACL 2002)","author":"Papineni","year":"2002"},{"key":"B27","doi-asserted-by":"crossref","first-page":"3982","DOI":"10.18653\/v1\/D19-1410","article-title":"\u201cSentence-bert: sentence embeddings using siamese bert-networks,\u201d","volume-title":"Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP)","author":"Reimers","year":"2019"},{"key":"B28","unstructured":"ClinicalEncoder26AM: a multlilingual diagnosable ColBERT model; evidences from the MultiClinNER shared task\n \n \n \n Remy\n F.\n \n \n arXiv\n \n 2026"},{"key":"B29","unstructured":"Modernbert + colbert: enhancing biomedical rag through an advanced re-ranking retriever\n \n \n \n Rivera\n E. M.\n \n \n Menolascina\n F.\n \n \n arXiv\n \n 2025"},{"key":"B30","doi-asserted-by":"publisher","first-page":"333","DOI":"10.1561\/1500000019","article-title":"The probabilistic relevance framework: BM25 and beyond","volume":"3","author":"Robertson","year":"2009","journal-title":"Found. Trends Inf. Retr"},{"key":"B31","doi-asserted-by":"publisher","first-page":"vbad074","DOI":"10.1093\/bioadv\/vbad074","article-title":"FigureReach: retrieval of figures from the biomedical literature using figure modality","volume":"3","author":"Trabucco","year":"2023","journal-title":"Bioinform. Adv"},{"key":"B32","doi-asserted-by":"publisher","first-page":"138","DOI":"10.1186\/s12859-015-0564-6","article-title":"An overview of the BIOASQ large-scale biomedical semantic indexing and question answering competition","volume":"16","author":"Tsatsaronis","year":"2015","journal-title":"BMC Bioinform"},{"key":"B33","doi-asserted-by":"crossref","first-page":"1","DOI":"10.1002\/9780471462422.eoct979","article-title":"\u201cWilcoxon signed-rank test,\u201d","volume-title":"Wiley Encyclopedia of Clinical Trials","author":"Woolson","year":"2008"},{"key":"B34","unstructured":"Google's neural machine translation system: bridging the gap between human and machine translation\n \n \n \n Wu\n Y.\n \n \n Schuster\n M.\n \n \n Chen\n Z.\n \n \n Le\n Q. V.\n \n \n Norouzi\n M.\n \n \n Macherey\n W.\n \n \n arXiv\n \n 2016"},{"key":"B35","article-title":"\u201cMMed-RAG: versatile multimodal RAG system for medical vision language models,\u201d","volume-title":"Proceedings of the international conference on learning representations (ICLR 2025)","author":"Xia","year":"2025"},{"key":"B36","doi-asserted-by":"crossref","first-page":"1081","DOI":"10.18653\/v1\/2024.emnlp-main.62","article-title":"\u201cRULE: reliable multimodal RAG for factuality in medical vision language models,\u201d","volume-title":"Proceedings of the conference on empirical methods in natural language processing (EMNLP 2024)","author":"Xia","year":"2024"},{"key":"B37","unstructured":"A survey for biomedical text summarization: from pre-trained to large language models\n \n \n \n Xie\n Q.\n \n \n Luo\n Z.\n \n \n Wang\n B.\n \n \n Ananiadou\n S.\n \n \n arXiv\n \n 2023"},{"key":"B38","article-title":"\u201cVisRAG: vision-based retrieval-augmented generation on multi-modality documents,\u201d","volume-title":"Proceedings of the international conference on learning representations (ICLR 2025)","author":"Yu","year":"2025"},{"key":"B39","doi-asserted-by":"publisher","first-page":"547","DOI":"10.1145\/3581783.3611830","article-title":"\u201cRAMM: retrieval-augmented biomedical visual question answering with multi-modal pre-training,\u201d","author":"Yuan","year":"2023","journal-title":"Proceedings of the 31st ACM international conference on multimedia"},{"key":"B40","doi-asserted-by":"publisher","first-page":"AIoa2400640","DOI":"10.1056\/AIoa2400640","article-title":"A multimodal biomedical foundation model trained from fifteen million image-text pairs","volume":"2","author":"Zhang","year":"2024","journal-title":"NEJM AI"},{"key":"B41","article-title":"\u201cBertScore: evaluating text generation with BERT,\u201d","volume-title":"Proceedings of the international conference on learning representations (ICLR 2020)","author":"Zhang","year":"2020"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1861611\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,29]],"date-time":"2026-09-29T04:43:24Z","timestamp":1790657004000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1861611\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,29]]},"references-count":41,"alternative-id":["10.3389\/frai.2026.1861611"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1861611","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,29]]},"article-number":"1861611"}}