Vision–language foundation models for multimodal medical image understanding and report generation: a narrative review with quantitative synthesis
5 affiliations
- 1Department of Computer Science, Bahria University, Lahore, Pakistan
- 2College of Pharmacy, University of Sharjah, Sharjah, United Arab Emirates
- 3Information Systems Department, College of Computer and Information Sciences, Imam Mohammad Ibn Saud Islamic University (IMSIU), Riyadh, Saudi Arabia
- 4Department of Software, Faculty of Artificial Intelligence and Software, Gachon University, Seongnam-si, Republic of Korea
- 5Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA, Shah Alam, Selangor, Malaysia
Abstract
Vision-language foundation models (VLFMs) for medical image understanding have advanced rapidly, yet evidence supporting clinical deployment remains fragmented. This narrative review synthesizes architectures, training paradigms, benchmarks, public datasets, and translation barriers. Drawing on more than 50 original studies published between 2020 and 2025, we provide a structured descriptive comparison rather than a statistical meta-analysis. In a focused audit of 13 representative studies, BLEU/ROUGE was reported by 31% (4/13), RadGraph F1 by 0% (0/13), CheXpert- or CheXbert-derived label evaluation by 46% (6/13), BERTScore by 8% (1/13), human expert evaluation by 31% (4/13), and inference latency or throughput by 0% (0/13). Because datasets, splits, candidate pools, and metrics differ, cross-task results are presented as exploratory benchmark indicators and not as a clinical-readiness scale. The synthesis identifies persistent concerns involving hallucination, limited interpretability, distribution shift, and narrow demographic and geographic coverage. We propose a minimum clinical validation and reporting standard aligned with the distinct purposes of FDA/Health Canada/MHRA and IMDRF Good Machine Learning Practice, the EU AI Act, TRIPOD+AI, and CLAIM, and identify eleven priority research directions. Medical VLFMs are promising, but progress should be judged by source-traceable, clinically grounded, and externally validated evidence.
Related
- Advances in Multimodal Artificial Intelligence in Radiology: Data Integration, Foundation Models, and Clinical Applications-A Narrative Review.
- Foundation models in medical image analysis: A systematic review and quantitative analysis
- Understanding Venous Remodeling After Minimally Invasive and Noninvasive Treatments for Varicose Veins: A Mechanistic Narrative Review
- Understanding underserved and underrepresented populations in European clinical studies: a narrative review with systematic approach
- Understanding the role of the hippocampus in representing multimodal stimuli
BibTeX
@article{mehmood2026visionlanguage,
title = {{Vision–language foundation models for multimodal medical image understanding and report generation: a narrative review with quantitative synthesis}},
author = {Shahid Mehmood and Farman Matloob Khan and Abeer A. K. Alharbi and Abdul Khader Jilani Saudagar and Mona Alkhattabi and Muhammad Adnan Khan and Muhammad Amir Khan},
journal = {Frontiers in Medicine},
year = {2026},
month = oct,
doi = {10.3389/fmed.2026.1958283},
url = {https://doi.org/10.3389/fmed.2026.1958283},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Vision–language foundation models for multimodal medical image understanding and report generation: a narrative review with quantitative synthesis
AU - Shahid Mehmood
AU - Farman Matloob Khan
AU - Abeer A. K. Alharbi
AU - Abdul Khader Jilani Saudagar
AU - Mona Alkhattabi
AU - Muhammad Adnan Khan
AU - Muhammad Amir Khan
PY - 2026
DA - 2026/10/26
JO - Frontiers in Medicine
DO - 10.3389/fmed.2026.1958283
UR - https://doi.org/10.3389/fmed.2026.1958283
AB - Vision-language foundation models (VLFMs) for medical image understanding have advanced rapidly, yet evidence supporting clinical deployment remains fragmented. This narrative review synthesizes architectures, training paradigms, benchmarks, public datasets, and translation barriers. Drawing on more than 50 original studies published between 2020 and 2025, we provide a structured descriptive comparison rather than a statistical meta-analysis. In a focused audit of 13 representative studies, BLEU/ROUGE was reported by 31% (4/13), RadGraph F1 by 0% (0/13), CheXpert- or CheXbert-derived label evaluation by 46% (6/13), BERTScore by 8% (1/13), human expert evaluation by 31% (4/13), and inference latency or throughput by 0% (0/13). Because datasets, splits, candidate pools, and metrics differ, cross-task results are presented as exploratory benchmark indicators and not as a clinical-readiness scale. The synthesis identifies persistent concerns involving hallucination, limited interpretability, distribution shift, and narrow demographic and geographic coverage. We propose a minimum clinical validation and reporting standard aligned with the distinct purposes of FDA/Health Canada/MHRA and IMDRF Good Machine Learning Practice, the EU AI Act, TRIPOD+AI, and CLAIM, and identify eleven priority research directions. Medical VLFMs are promising, but progress should be judged by source-traceable, clinically grounded, and externally validated evidence.
SN - 2296-858X
ER -