{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,22]],"date-time":"2026-09-22T06:05:37Z","timestamp":1790057137314,"version":"4.0.1"},"reference-count":46,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,9,22]],"date-time":"2026-09-22T00:00:00Z","timestamp":1790035200000},"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":"<jats:sec>\n                    <jats:title>Introduction<\/jats:title>\n                    <jats:p>Medical vision-language pre-training (MedVLP) learns transferable cross-modal representations by aligning chest X-ray images with radiology reports. However, most existing CLIP-style methods have two limitations. First, they treat all unpaired image-report samples within a mini-batch as negatives, which may incorrectly penalize medically similar samples as false negatives. Second, they assume that all samples contribute equally to optimization, overlooking the heterogeneous difficulty of medical data arising from factors such as disease rarity, severity, comorbidity, and report complexity.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>To address these limitations, we propose Medical Prior-Driven Difficulty-Aware Representation Learning (MPDRL), which incorporates sample-level medical difficulty priors into multimodal representation learning. This method first constructs an eight-dimensional medical prior feature vector for each image-report pair to characterize sample heterogeneity. Based on these priors, it then incorporates difficulty-aware feature modulation, adaptive temperature scaling, medical prior-softened contrastive learning, and difficulty consistency regularization to reduce false-negative penalties among medically similar samples and preserve meaningful difficulty structures in multimodal representations.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>Extensive experiments on five medical imaging datasets across three downstream tasks demonstrate the effectiveness of MPDRL. For zero-shot classification, MPDRL achieves an AUC of 87.7% on RSNA, outperforming the strongest baseline by 0.2 percentage points, and an AUC of 84.1% on CheXpert, which is 0.1 percentage points lower than the best-performing baseline, IMITATE. On CheXpert14, MPDRL achieves the best F1-score and accuracy, reaching 26.9% and 86.5%, respectively, although its AUC of 71.8% remains lower than those of several competing methods. For zero-shot image-text retrieval on CheXpert 8 \u00d7 200, MPDRL achieves Precision@1 scores of 39.9% for image-to-text retrieval and 64.7% for text-to-image retrieval. For report generation, MPDRL achieves BLEU-1 scores of 0.508 on IU X-Ray and 0.403 on MIMIC-CXR. Importantly, clinical evaluation using CheXbert and RadGraph further shows improved medical factual consistency. On IU X-Ray, MPDRL achieves CheXbert precision, recall, and F1 scores of 0.462, 0.414, and 0.437, respectively, while obtaining an RG-F1 score of 0.256. On MIMIC-CXR, it achieves CheXbert precision, recall, and F1 scores of 0.405, 0.354, and 0.378, respectively, together with an RG-F1 score of 0.238.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>Overall, these results indicate that medical prior-driven difficulty modeling can improve the robustness of MedVLP representations across diverse downstream tasks.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/frai.2026.1906686","type":"journal-article","created":{"date-parts":[[2026,9,22]],"date-time":"2026-09-22T05:41:13Z","timestamp":1790055673000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["MPDRL: difficulty-aware representation learning with medical priors for vision\u2013language pre-training"],"prefix":"10.3389","volume":"9","author":[{"given":"Juntao","family":"Liu","sequence":"first","affiliation":[{"name":"School of Low-Altitude Technology and Engineering, Guangdong Polytechnic Normal University","place":["Heyuan, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhuoyi","family":"Tan","sequence":"additional","affiliation":[{"name":"School of Low-Altitude Technology and Engineering, Guangdong Polytechnic Normal University","place":["Heyuan, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Weiqi","family":"Chen","sequence":"additional","affiliation":[{"name":"School of Computer Science, Guangdong Polytechnic Normal University","place":["Heyuan, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiaoyong","family":"Liu","sequence":"additional","affiliation":[{"name":"School of Low-Altitude Technology and Engineering, Guangdong Polytechnic Normal University","place":["Heyuan, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,9,22]]},"reference":[{"key":"B1","article-title":"Maira-2: Grounded radiology report generation","author":"Bannur","year":"2024","journal-title":"arXiv preprint arXiv:2406.04449"},{"key":"B2","doi-asserted-by":"publisher","first-page":"41","DOI":"10.1145\/1553374.1553380","article-title":"\u201cCurriculum learning,\u201d","author":"Bengio","year":"2009","journal-title":"Proceedings of the 26th annual international conference on machine learning"},{"key":"B3","first-page":"1","article-title":"\u201cMaking the most of text semantics to improve biomedical vision-language processing,\u201d","volume-title":"Computer vision-ECCV 2022, volume 13696 of Lecture Notes in Computer Science","author":"Boecking","year":"2022"},{"key":"B4","first-page":"1597","article-title":"\u201cA simple framework for contrastive learning of visual representations,\u201d","volume-title":"Proceedings of the international conference on machine learning (ICML)","author":"Chen","year":"2020"},{"key":"B5","doi-asserted-by":"publisher","first-page":"13342","DOI":"10.18653\/v1\/2023.emnlp-main.824","article-title":"\u201cVistruct: visual structural knowledge extraction via curriculum guided code-vision representation,\u201d","author":"Chen","year":"2023","journal-title":"Proceedings of the 2023 conference on empirical methods in natural language processing"},{"key":"B6","doi-asserted-by":"publisher","first-page":"5904","DOI":"10.18653\/v1\/2021.acl-long.459","article-title":"\u201cCross-modal memory networks for radiology report generation,\u201d","author":"Chen","year":"2021","journal-title":"Proceedings of the 59th annual meeting of the association for computational linguistics and the 11th international joint conference on natural language processing (volume 1: long papers)"},{"key":"B7","article-title":"Generating radiology reports via memory-driven transformer","author":"Chen","year":"2020","journal-title":"arXiv preprint arXiv:2010.16056"},{"key":"B8","doi-asserted-by":"publisher","first-page":"304","DOI":"10.1093\/jamia\/ocv080","article-title":"Preparing a collection of radiology examinations for distribution and retrieval","volume":"23","author":"Demner-Fushman","year":"2015","journal-title":"J. Am. Med. Inform. Assoc"},{"key":"B9","doi-asserted-by":"publisher","first-page":"2808","DOI":"10.1109\/CVPR52729.2023.00275","article-title":"\u201cTexts as images in prompt tuning for multi-label image recognition,\u201d","author":"Guo","year":"2023","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition"},{"key":"B10","doi-asserted-by":"publisher","first-page":"9729","DOI":"10.1109\/CVPR42600.2020.00975","article-title":"\u201cMomentum contrast for unsupervised visual representation learning,\u201d","author":"He","year":"2020","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition (CVPR)"},{"key":"B11","doi-asserted-by":"publisher","first-page":"3942","DOI":"10.1109\/ICCV48922.2021.00391","article-title":"\u201cGloria: a multimodal global-local representation learning framework for label-efficient medical image recognition,\u201d","author":"Huang","year":"2021","journal-title":"Proceedings of the IEEE\/CVF international conference on computer vision"},{"key":"B12","doi-asserted-by":"publisher","first-page":"19809","DOI":"10.1109\/CVPR52729.2023.01897","article-title":"\u201cKiut: knowledge-injected u-transformer for radiology report generation,\u201d","author":"Huang","year":"2023","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition"},{"key":"B13","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v29i1.9608","article-title":"\u201cSelf-paced curriculum learning,\u201d","author":"Jiang","year":"2015","journal-title":"Proceedings of the AAAI conference on artificial intelligence"},{"key":"B14","article-title":"Mimic-cxr-jpg, a large publicly available database of labeled chest radiographs","author":"Johnson","year":"2019","journal-title":"arXiv preprint arXiv:1901.07042"},{"key":"B15","first-page":"18661","article-title":"\u201cSupervised contrastive learning,\u201d","author":"Khosla","year":"2020","journal-title":"Advances in neural information processing systems"},{"key":"B16","first-page":"1189","article-title":"\u201cSelf-paced learning for latent variable models,\u201d","author":"Kumar","year":"2010","journal-title":"Advances in neural information processing systems (NeurIPS)"},{"key":"B17","doi-asserted-by":"publisher","first-page":"3334","DOI":"10.1109\/CVPR52729.2023.00325","article-title":"\u201cDynamic graph enhanced contrastive learning for chest x-ray report generation,\u201d","author":"Li","year":"2023","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition"},{"key":"B18","doi-asserted-by":"publisher","first-page":"3587","DOI":"10.1109\/JBHI.2025.3528196","article-title":"Medfilip: Medical fine-grained language-image pre-training","volume":"29","author":"Liang","year":"2025","journal-title":"IEEE J. Biomed. Health Inform"},{"key":"B19","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v40i9.37620","article-title":"\u201cAnatomical region-guided contrastive decoding: a plug-and-play strategy for mitigating hallucinations in medical vlms,\u201d","author":"Liang","year":"2026","journal-title":"Proceedings of the AAAI conference on artificial intelligence"},{"key":"B20","doi-asserted-by":"publisher","first-page":"519","DOI":"10.1109\/TMI.2024.3449690","article-title":"Imitate: clinical prior guided hierarchical vision-language pre-training","volume":"44","author":"Liu","year":"2024","journal-title":"IEEE Trans. Med. Imag"},{"key":"B21","doi-asserted-by":"publisher","first-page":"13753","DOI":"10.1109\/CVPR46437.2021.01354","article-title":"\u201cExploring and distilling posterior and prior knowledge for radiology report generation,\u201d","author":"Liu","year":"2021","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition"},{"key":"B22","doi-asserted-by":"publisher","first-page":"7206","DOI":"10.1609\/aaai.v40i9.37657","article-title":"\u201cPriorrg: prior-guided contrastive pre-training and coarse-to-fine decoding for chest x-ray report generation,\u201d","author":"Liu","year":"","journal-title":"Proceedings of the AAAI conference on artificial intelligence"},{"key":"B23","doi-asserted-by":"publisher","first-page":"10348","DOI":"10.1109\/CVPR52734.2025.00968","article-title":"\u201cEnhanced contrastive learning with multi-view longitudinal data for chest x-ray report generation,\u201d","author":"Liu","year":"2025","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition"},{"key":"B24","article-title":"Seeing like radiologists: context- and gaze-guided vision-language pretraining for chest x-rays","author":"Liu","year":"","journal-title":"arXiv preprint arXiv:2603.26049"},{"key":"B25","first-page":"8748","article-title":"\u201cLearning transferable visual models from natural language supervision,\u201d","volume-title":"International conference on machine learning","author":"Radford","year":"2021"},{"key":"B26","article-title":"\u201cContrastive learning with hard negative samples,\u201d","author":"Robinson","year":"2021","journal-title":"International conference on learning representations"},{"key":"B27","doi-asserted-by":"publisher","first-page":"e180041","DOI":"10.1148\/ryai.2019180041","article-title":"Augmenting the national institutes of health chest radiograph dataset with expert annotations of possible pneumonia","volume":"1","author":"Shih","year":"2019","journal-title":"Radiology"},{"key":"B28","doi-asserted-by":"publisher","first-page":"1399","DOI":"10.1038\/s41551-022-00936-9","article-title":"Expert-level detection of pathologies from unannotated chest x-ray images via self-supervised learning","volume":"6","author":"Tiu","year":"2022","journal-title":"Nat. Biomed. Eng"},{"key":"B29","doi-asserted-by":"publisher","first-page":"9500","DOI":"10.18653\/v1\/2024.findings-acl.566","article-title":"\u201cBootstrapping LLM-based task-oriented dialogue agents via self-talk,\u201d","author":"Ulmer","year":"2024","journal-title":"Findings of the association for computational linguistics: ACL 2024"},{"key":"B30","doi-asserted-by":"publisher","first-page":"33536","DOI":"10.52202\/068431-2430","article-title":"\u201cMulti-granularity cross-modal alignment for generalized medical visual representation learning,\u201d","author":"Wang","year":"2022","journal-title":"Advances in neural information processing systems"},{"key":"B31","doi-asserted-by":"publisher","first-page":"109583","DOI":"10.1016\/j.engappai.2024.109583","article-title":"Cross-attention interaction learning network for multi-model image fusion via transformer","volume":"139","author":"Wang","year":"2025","journal-title":"Eng. Appl. Artif. Intell"},{"key":"B32","doi-asserted-by":"publisher","first-page":"2097","DOI":"10.1109\/CVPR.2017.369","article-title":"\u201cChestx-ray8: hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases,\u201d","author":"Wang","year":"2017","journal-title":"Proceedings of the IEEE conference on computer vision and pattern recognition"},{"key":"B33","doi-asserted-by":"publisher","first-page":"100033","DOI":"10.1016\/j.metrad.2023.100033","article-title":"R2GenGPT: radiology report generation with frozen LLMs","volume":"1","author":"Wang","year":"2023","journal-title":"Meta-Radiology"},{"key":"B34","doi-asserted-by":"publisher","first-page":"3876","DOI":"10.18653\/v1\/2022.emnlp-main.256","article-title":"\u201cMedclip: contrastive learning from unpaired medical images and text,\u201d","author":"Wang","year":"2022","journal-title":"Proceedings of the conference on empirical methods in natural language processing. conference on empirical methods in natural language processing"},{"key":"B35","article-title":"Medklip: medical knowledge enhanced language-image pre-training in radiology","author":"Wu","year":"2023","journal-title":"arXiv preprint arXiv:2301.02228"},{"key":"B36","doi-asserted-by":"publisher","first-page":"2982","DOI":"10.1609\/aaai.v36i3.20204","article-title":"\u201cClinical-bert: vision-language pre-training for radiograph diagnosis and reports generation,\u201d","author":"Yan","year":"2022","journal-title":"Proceedings of the AAAI conference on artificial intelligence"},{"key":"B37","first-page":"4096","article-title":"\u201cKia: knowledge-guided implicit vision-language alignment for chest x-ray report generation,\u201d","author":"Yin","year":"2025","journal-title":"Proceedings of the 31st international conference on computational linguistics"},{"key":"B38","doi-asserted-by":"publisher","first-page":"7008","DOI":"10.1109\/ICCV51070.2023.00645","article-title":"\u201cCyclic-bootstrap labeling for weakly supervised object detection,\u201d","author":"Yin","year":"2023","journal-title":"Proceedings of the IEEE\/CVF international conference on computer vision"},{"key":"B39","first-page":"72","article-title":"\u201cAligntransformer: hierarchical alignment of visual regions and disease tags for medical report generation,\u201d","volume-title":"International conference on medical image computing and computer-assisted intervention","author":"You","year":"2021"},{"key":"B40","doi-asserted-by":"publisher","first-page":"20778","DOI":"10.1109\/CVPR52734.2025.01935","article-title":"\u201cChexworld: exploring image world modeling for radiograph representation learning,\u201d","author":"Yue","year":"2025","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition"},{"key":"B41","first-page":"35260","article-title":"\u201cMedkco: medical vision-language pretraining via knowledge-driven cognitive orchestration,\u201d","author":"Zhang","year":"2026","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition"},{"key":"B42","doi-asserted-by":"publisher","first-page":"4706","DOI":"10.1109\/TMM.2023.3325965","article-title":"Multi-task paired masking with alignment modeling for medical vision-language pre-training","volume":"26","author":"Zhang","year":"2023","journal-title":"IEEE Trans. Multim"},{"key":"B43","doi-asserted-by":"publisher","first-page":"4542","DOI":"10.1038\/s41467-023-40260-7","article-title":"Knowledge-enhanced visual-language pre-training on chest radiology images","volume":"14","author":"Zhang","year":"2023","journal-title":"Nat. Commun"},{"key":"B44","first-page":"2","article-title":"\u201cContrastive learning of medical visual representations from paired images and text,\u201d","volume-title":"Machine learning for healthcare conference","author":"Zhang","year":"2022"},{"key":"B45","doi-asserted-by":"publisher","first-page":"29744","DOI":"10.1109\/CVPR52734.2025.02769","article-title":"\u201cMedunifier: unifying vision-and-language pre-training on medical data with vision generation task using discrete visual representations,\u201d","author":"Zhang","year":"2025","journal-title":"Proceedings of the IEEE\/CVF conference on computer vision and pattern recognition"},{"key":"B46","doi-asserted-by":"publisher","first-page":"110250","DOI":"10.1016\/j.patcog.2024.110250","article-title":"Exploring low-resource medical image classification with weakly supervised prompt learning","volume":"149","author":"Zheng","year":"2024","journal-title":"Pattern Recognit"}],"container-title":["Frontiers in Artificial Intelligence"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1906686\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,22]],"date-time":"2026-09-22T05:41:16Z","timestamp":1790055676000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/frai.2026.1906686\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,22]]},"references-count":46,"alternative-id":["10.3389\/frai.2026.1906686"],"URL":"https:\/\/doi.org\/10.3389\/frai.2026.1906686","relation":{},"ISSN":["2624-8212"],"issn-type":[{"value":"2624-8212","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,22]]},"article-number":"1906686"}}