@inproceedings{he-etal-2026-semli,
title = "{S}e{MLI}-{RAG}: Semantics-Guided Multimodal Late Interaction for Retrieval-Augmented Radiology Report Generation",
author = {He, Jiawei and
Feng, Weibo and
Ferdowsi, Sohrab and
Yazdani, Anthony and
Zhang, Boya and
{\c{C}}inar-Kora{\c{s}}, Osman Alperen and
Khattab, Sameh and
Corbeil, Jean-Philippe and
M{\"u}ller, Henning and
Teodoro, Douglas},
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2379/",
pages = "54040--54055",
ISBN = "979-8-89176-459-0",
abstract = "The consultation of reference cases in clinical radiology reporting motivates integrating multimodal retrieval-augmented generation (RAG) into automated systems. However, traditional RAG approaches employ single-vector embeddings for indexing, resulting in loss of granular information. Moreover, a semantic gap exists whereby visually similar radiographs often encode distinct pathologies. In this paper, we propose a Semantics-guided Multi-modal Late Interaction RAG (SeMLI-RAG) framework for radiology report generation. Our approach employs a weight-based late interaction strategy for retrieval to capture fine-grained radiographic features. Furthermore, we introduce an image-adaptive RAG to selectively leverage retrieved information, mitigating performance degradation caused by retrieval-induced noise. Experiments on three radiology benchmarks demonstrate that our framework yields consistent improvements over the backbone and RAG baselines across multiple evaluation dimensions. Further analysis shows that improving retrieval granularity and retrieval robustness can both benefit radiology report generation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="he-etal-2026-semli">
<titleInfo>
<title>SeMLI-RAG: Semantics-Guided Multimodal Late Interaction for Retrieval-Augmented Radiology Report Generation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Jiawei</namePart>
<namePart type="family">He</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Weibo</namePart>
<namePart type="family">Feng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sohrab</namePart>
<namePart type="family">Ferdowsi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anthony</namePart>
<namePart type="family">Yazdani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Boya</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Osman</namePart>
<namePart type="given">Alperen</namePart>
<namePart type="family">Çinar-Koraş</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sameh</namePart>
<namePart type="family">Khattab</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jean-Philippe</namePart>
<namePart type="family">Corbeil</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Henning</namePart>
<namePart type="family">Müller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Douglas</namePart>
<namePart type="family">Teodoro</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>The consultation of reference cases in clinical radiology reporting motivates integrating multimodal retrieval-augmented generation (RAG) into automated systems. However, traditional RAG approaches employ single-vector embeddings for indexing, resulting in loss of granular information. Moreover, a semantic gap exists whereby visually similar radiographs often encode distinct pathologies. In this paper, we propose a Semantics-guided Multi-modal Late Interaction RAG (SeMLI-RAG) framework for radiology report generation. Our approach employs a weight-based late interaction strategy for retrieval to capture fine-grained radiographic features. Furthermore, we introduce an image-adaptive RAG to selectively leverage retrieved information, mitigating performance degradation caused by retrieval-induced noise. Experiments on three radiology benchmarks demonstrate that our framework yields consistent improvements over the backbone and RAG baselines across multiple evaluation dimensions. Further analysis shows that improving retrieval granularity and retrieval robustness can both benefit radiology report generation.</abstract>
<identifier type="citekey">he-etal-2026-semli</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2379/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>54040</start>
<end>54055</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T SeMLI-RAG: Semantics-Guided Multimodal Late Interaction for Retrieval-Augmented Radiology Report Generation
%A He, Jiawei
%A Feng, Weibo
%A Ferdowsi, Sohrab
%A Yazdani, Anthony
%A Zhang, Boya
%A Çinar-Koraş, Osman Alperen
%A Khattab, Sameh
%A Corbeil, Jean-Philippe
%A Müller, Henning
%A Teodoro, Douglas
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F he-etal-2026-semli
%X The consultation of reference cases in clinical radiology reporting motivates integrating multimodal retrieval-augmented generation (RAG) into automated systems. However, traditional RAG approaches employ single-vector embeddings for indexing, resulting in loss of granular information. Moreover, a semantic gap exists whereby visually similar radiographs often encode distinct pathologies. In this paper, we propose a Semantics-guided Multi-modal Late Interaction RAG (SeMLI-RAG) framework for radiology report generation. Our approach employs a weight-based late interaction strategy for retrieval to capture fine-grained radiographic features. Furthermore, we introduce an image-adaptive RAG to selectively leverage retrieved information, mitigating performance degradation caused by retrieval-induced noise. Experiments on three radiology benchmarks demonstrate that our framework yields consistent improvements over the backbone and RAG baselines across multiple evaluation dimensions. Further analysis shows that improving retrieval granularity and retrieval robustness can both benefit radiology report generation.
%U https://aclanthology.org/2026.emnlp-main.2379/
%P 54040-54055
Markdown (Informal)
[SeMLI-RAG: Semantics-Guided Multimodal Late Interaction for Retrieval-Augmented Radiology Report Generation](https://aclanthology.org/2026.emnlp-main.2379/) (He et al., EMNLP 2026)
ACL
- Jiawei He, Weibo Feng, Sohrab Ferdowsi, Anthony Yazdani, Boya Zhang, Osman Alperen Çinar-Koraş, Sameh Khattab, Jean-Philippe Corbeil, Henning Müller, and Douglas Teodoro. 2026. SeMLI-RAG: Semantics-Guided Multimodal Late Interaction for Retrieval-Augmented Radiology Report Generation. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 54040–54055, Budapest, Hungary. Association for Computational Linguistics.