Leveraging multilingual RAG for breast cancer RCPs: AI-driven speech transcription and compliance in Darija-French clinical discussions
Open accessAbstract
• First end-to-end multilingual Voice-RAG for oncology RCPs (Darija/French) with real-time notes. • High-accuracy Darija ASR (BERTScore F1≈100 on DODa) with strong cross-corpus generalization. • Sentence-level retrieval (HNSW) plus compliance guardrails for toxicity, leakage, hallucination. • Reliable RAG across 13 LLMs with high answer relevance and groundedness on 40 clinical queries. Extensible to other dialects/specialties; next steps: clinical corpora and latency optimization. The integration of artificial intelligence (AI) into clinical decision-making has introduced new opportunities for automating and enhancing medical documentation, particularly in oncology, where multidisciplinary meetings are central to treatment planning. However, existing speech-to-text and retrieval-augmented generation (RAG) systems are not equipped to operate effectively in multilingual, dialect-rich environments such as those in North African hospitals where Moroccan Darija, Arabic, and French are frequently interwoven. These linguistic complexities, combined with the high-stakes nature of clinical dialogue, challenge transcription accuracy, contextual information retrieval, and regulatory compliance. This study presents a multilingual RAG system tailored to clinical meetings, integrating a fine-tuned Whisper ASR model with a sentence-level semantic retrieval pipeline and a compliance-aware generation framework. Evaluated on real-world clinical queries, the system demonstrates improved transc…
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In [22], a planner, retriever and verifier pipeline improved reasoning stability by mirroring clinical thought processes.
“In [22], a planner, retriever and verifier pipeline improved reasoning stability by mirroring clinical thought processes.”
From Multi-Agent Retrieval Augmented Generation for Clinical Decision Support: A Systematic Review and Integrative Conceptual Framework (2026) · cites this paper as [22]
Intermediate agent outputs such as plan, retrieved evidence, and verification results allow auditability that is not possible with monolithic ML systems.
“In [12][22]the intermediate agent outputs steps to plan, retrieved evidence, verification results were the subjects of analysis and would allow for auditability that was not possible with monolithic ML systems.”
From Multi-Agent Retrieval Augmented Generation for Clinical Decision Support: A Systematic Review and Integrative Conceptual Framework (2026) · cites this paper as [22]
Multi-agent RAG systems enhance explainability in planner, retriever, and verifier frameworks.
“The multi-agent RAG systems enhances explainability in the planner, retriever and verifier frameworks [22][12].”
From Multi-Agent Retrieval Augmented Generation for Clinical Decision Support: A Systematic Review and Integrative Conceptual Framework (2026) · cites this paper as [22]
Multi-agent systems such as planner, retriever, and verifier pipelines reduced hallucinations and improved reasoning consistency compared with single-agent LLMs.
“Multi-agent systems such as planner, retriever and verifier pipelines reduced hallucinations and improved reasoning consistency compared with single-agent LLMs [22][12]”
From Multi-Agent Retrieval Augmented Generation for Clinical Decision Support: A Systematic Review and Integrative Conceptual Framework (2026) · cites this paper as [22]
Verification agents clear unsupported outputs and prevent hallucinations, the single greatest barrier to clinician confidence in AI.
“Verification agents were vital in [22][12] clearing unsupported outputs and preventing hallucinations the single greatest barrier to building clinician confidence in AI.”
From Multi-Agent Retrieval Augmented Generation for Clinical Decision Support: A Systematic Review and Integrative Conceptual Framework (2026) · cites this paper as [22]
Addition of verification agents reduced hallucinations, but lack of such safeguards in low-resource configurations may increase clinical risk.
“The addition of verification agents in [22][12]reduced hallucinations, but the lack of such safeguards in low-resource configurations may increase clinical risk.”
From Multi-Agent Retrieval Augmented Generation for Clinical Decision Support: A Systematic Review and Integrative Conceptual Framework (2026) · cites this paper as [22]
Cites
Cortazar, P., Zhang, L., Untch, M., Mehta, K. M., Costantino, J. P., Wolmark, N., Bonnefoi, H. R., Cameron, D., Gianni, L., Valagussa, P., et al. (2014) · The Lancet · doi:10.1016/s0140-6736(13)62422-8 · Cited by 4,777 · see citing works
Foulkes, W. D., Smith, I. E., & Reis-Filho, J. S. (2010) · New England Journal of Medicine · doi:10.1056/nejmra1001389 · Cited by 4,719 · see citing works
Pembrolizumab for Early Triple-Negative Breast Cancer
Schmid, P., Cortes, J., Pusztai, L., McArthur, H., Kümmel, S., Bergh, J., Denkert, C., Park, Y. H., Hui, R., Harbeck, N., et al. (2020) · New England Journal of Medicine · doi:10.1056/nejmoa1910549 · Cited by 3,270 · see citing works
wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations
Baevski, A., Zhou, H., Mohamed, A., & Auli, M. (2020) · arXiv · doi:10.48550/arxiv.2006.11477 · Cited by 2,369 · see citing works
Early breast cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up
Cardoso, F., Kyriakides, S., Ohno, S., Penault-Llorca, F., Poortmans, P., Rubio, I. T., Zackrisson, S., & Senkus, E. (2019) · Annals of Oncology · doi:10.1093/annonc/mdz173 · Cited by 2,214 · see citing works
Implementing Machine Learning in Health Care — Addressing Ethical Challenges
Char, D. S., Shah, N. H., & Magnus, D. (2018) · New England Journal of Medicine · doi:10.1056/nejmp1714229 · Cited by 1,857 · see citing works
Measurement of Residual Breast Cancer Burden to Predict Survival After Neoadjuvant Chemotherapy
Symmans, W. F., Peintinger, F., Hatzis, C., Rajan, R., Kuerer, H. M., Valero, V., Assad, L., Poniecka, A. W., Hennessy, B. T., Green, M. C., et al. (2007) · Journal of Clinical Oncology · doi:10.1200/jco.2007.10.6823 · Cited by 1,655 · see citing works
Medical Hallucination in Foundation Models and Their Impact on HealthcarePreprint
Kim, Y., Jeong, H., Chen, S., Li, S. S., Park, C., Lu, M., Alhamoud, K., Mun, J., Grau, C., Jung, M., et al. (2025) · medRxiv · doi:10.1101/2025.02.28.25323115 · Cited by 253 · see citing works
Malkov, Y. A., & Yashunin, D. A. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1603.09320 · Cited by 182 · see citing works
Retrieval-Augmented Generation (RAG) in Healthcare: A Comprehensive Review
Neha, F., Bhati, D., & Shukla, D. K. (2025) · AI · doi:10.3390/ai6090226 · Cited by 117 · see citing works
Thirunavukarasu, A. J., Mahmood, S., Malem, A. H., Foster, W., Sanghera, R., Hassan, R., Zhou, S. X., Wong, S. W., Wong, Y. L., Chong, Y. J., et al. (2024) · PLOS Digital Health · doi:10.1371/journal.pdig.0000341 · Cited by 95 · see citing works
Systematic Analysis of Retrieval-Augmented Generation-Based LLMs for Medical Chatbot Applications
Bora, A., & Cuayáhuitl, H. (2024) · Machine Learning and Knowledge Extraction · doi:10.3390/make6040116 · Cited by 73 · see citing works
ParlaMint II: advancing comparable parliamentary corpora across Europe
Erjavec, T., Kopp, M., Ljubešić, N., Kuzman, T., Rayson, P., Osenova, P., Ogrodniczuk, M., Çöltekin, Ç., Koržinek, D., Meden, K., et al. (2024) · Language Resources and Evaluation · doi:10.1007/s10579-024-09798-w · Cited by 19 · see citing works
Jelassi, M., Jemai, O., & Jacques Demongeot, A. (2024) · Diagnostics · doi:10.3390/diagnostics14090895 · Cited by 16 · see citing works
Retrieval Augmented Generation for Large Language Models in Healthcare: A Systematic ReviewPreprint
Amugongo, L. M., Mascheroni, P., Brooks, S. G., Doering, S., & Seidel, J. (2024) · Preprints.org · doi:10.20944/preprints202407.0876.v1 · Cited by 15 · see citing works
Hamane, Z., Ben-Bouazza, F.-E., & Fennan, A. (2024) · International Conference on Intelligent Systems and Computer Vision (ISCV) · doi:10.1109/iscv60512.2024.10620139 · Cited by 13 · see citing works
Self-citationMoroccan Dialect -Darija- Open DatasetPreprint
Outchakoucht, A., & Es-Samaali, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.09687 · Cited by 10 · see citing works
The Evolution of Darija Open Dataset: Introducing Version 2Preprint
Outchakoucht, A., & Es-Samaali, H. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2405.13016 · Cited by 7 · see citing works
Clinical Dialogue Transcription Error Correction Using Seq2Seq Models
Nanayakkara, G., Wiratunga, N., Corsar, D., Martin, K., & Wijekoon, A. (2022) · Studies in computational intelligence · doi:10.1007/978-3-031-14771-5_4 · Cited by 5 · see citing works
RASU: Retrieval Augmented Speech Understanding through Generative Modeling
Yang, H., Zhang, M., Wang, M., & Guo, J. (2024) · Interspeech (USB) · doi:10.21437/interspeech.2024-148 · Cited by 4 · see citing works
Xu, J., Tan, D., Wang, J., & Chen, X. (2025) · IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU) · doi:10.1109/asru65441.2025.11434663 · Cited by 2 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Mugambiwa, T., & Ndlovu, B. (2026) · Journal of Applied Informatics and Computing · doi:10.30871/jaic.v10i1.11900 · Cited by 4 · see citing works
“In addition, in [21] multilingual and compliance-related retrieval showed that RAG is able to manage complex documentation environment that non classic approach of using the NLP system due to language barriers.”
Automatic Speech Recognition in Healthcare in the Post-LLM Era: A Scoping Review
Alabbad, M., & Alhoshan, W. (2026) · Healthcare · doi:10.3390/healthcare14101333 · Cited by 1 · see citing works
Stifini, D., Penna, A. D., Mihaljevic, A., Missios, P., Bitzer, M., Eickhoff, C., & Capobianco, I. (2026) · Research Square · doi:10.21203/rs.3.rs-8849187/v1
Akbulut, B., Aslam, B., & Al-Naimi, K. (2025) · Journal Islamic Economic Minangkabau · doi:10.70177/jiem.v3i2.2749
- Authors
- Emssaad, I., Ben-Bouazza, F.-E., Tafala, I., Mezali, M. C. E., & Jioudi, B.
- Affiliations
- Université Mohammed VI des Sciences et de la Santé (MA)
- Instituts Supérieurs des Professions Infirmières et Techniques de Santé (MA)
- Corresponding author
- Emssaad, Ilyass
- ORCID
- Venue
- Computer Methods and Programs in Biomedicine Update
- Publisher
- Elsevier BV
- Published
- 2025
- Volume
- 8
- Pages
- 100221
- ISSN
- 2666-9900 (ISSN-L)
- OpenAlex
- W4415218565
- Cited by
- 2
- References
- 24
- Citations by year
- 2026: 2
- FWCI
- 0.49
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Computer Methods and Programs in Biomedicine Update (publishedVersion, cc-by-nc-nd)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Physical Sciences › Computer Science › Artificial Intelligence › Speech Recognition and Synthesis
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- Moroccan Arabic, speech-to-text, hallucination detection, clinical documentation automation, cross-corpus generalization
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
