{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,12]],"date-time":"2026-08-12T13:42:35Z","timestamp":1786542155001,"version":"build-2736575974"},"reference-count":51,"publisher":"MDPI AG","issue":"8","license":[{"start":{"date-parts":[[2026,8,12]],"date-time":"2026-08-12T00:00:00Z","timestamp":1786492800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Bioengineering"],"abstract":"<jats:p>Vision-language models (VLMs) are increasingly evaluated for retinal image interpretation, but end-task discrimination does not establish whether a prediction depends on the optical coherence tomography (OCT) image. We evaluated RetinaVLM, LLaVA-Med, and Qwen3.6-27B under zero-shot and parameter-efficient adaptation using an APTOS-2021 cohort (128 training, 21 validation, and 69 test eyes) and a 100-eye cross-site stress-test cohort. The primary readout compared forced-choice continue\/stop scores obtained with a real OCT B-scan and a uniform-grey image while holding the generated rationale fixed; it therefore estimates a rationale-conditioned direct image contrast rather than total image dependence. Across 22 internal cells, 21 confidence intervals included an AUC of 0.5, while one Qwen zero-shot cell was inversely aligned (AUC 0.354, 95% CI 0.224\u20130.493). No positively aligned cells survived the Benjamini\u2013Hochberg adjustment. Selected backbones produced image-responsive biomarker outputs, including pigment epithelial detachment balanced accuracy up to 0.93, whereas a five-field tabular reference model achieved decision AUC 0.731 (95% CI 0.603\u20130.846). Similar direct-contrast findings occurred in the second-site stress test. Because the generated rationales were not verified as faithful representations of latent computation, the grey image was out of distribution, and the cohorts were small, null direct contrasts cannot show that the models ignored OCT or establish equivalence to no discrimination.<\/jats:p>","DOI":"10.3390\/bioengineering13080912","type":"journal-article","created":{"date-parts":[[2026,8,12]],"date-time":"2026-08-12T12:53:48Z","timestamp":1786539228000},"page":"912","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["A Rationale-Conditioned Image-Contrast Audit of OCT Dependence in Vision-Language Models for Anti-VEGF Treatment-Response Prediction"],"prefix":"10.3390","volume":"13","author":[{"given":"Wonbong","family":"Jang","sequence":"first","affiliation":[{"name":"Clear Eye Clinic, 181-26 Jijedongsak 2-ro, Pyeongtaek 18029, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0001-1532-8194","authenticated-orcid":false,"given":"Gwon Yul","family":"Jo","sequence":"additional","affiliation":[{"name":"BioNexus Inc., Gwanggyo Business Center, 156 Gwanggyo-ro, Yeongtong-gu, Suwon 16506, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0003-1089-9443","authenticated-orcid":false,"given":"Siyun","family":"Lee","sequence":"additional","affiliation":[{"name":"Department of Ophthalmology, CHA Bundang Medical Center, CHA University School of Medicine, 59 Yatap-ro, Bundang-gu, Seongnam 13496, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Shin Jeong","family":"Yoon","sequence":"additional","affiliation":[{"name":"Department of Ophthalmology, CHA Bundang Medical Center, CHA University School of Medicine, 59 Yatap-ro, Bundang-gu, Seongnam 13496, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Gi Young","family":"Lee","sequence":"additional","affiliation":[{"name":"BioNexus Inc., Gwanggyo Business Center, 156 Gwanggyo-ro, Yeongtong-gu, Suwon 16506, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-7485-9083","authenticated-orcid":false,"given":"Hee-Eun","family":"Lee","sequence":"additional","affiliation":[{"name":"BioNexus Inc., Gwanggyo Business Center, 156 Gwanggyo-ro, Yeongtong-gu, Suwon 16506, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Tae Hyung","family":"Kim","sequence":"additional","affiliation":[{"name":"BioNexus Inc., Gwanggyo Business Center, 156 Gwanggyo-ro, Yeongtong-gu, Suwon 16506, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jong Won","family":"Baek","sequence":"additional","affiliation":[{"name":"BioNexus Inc., Gwanggyo Business Center, 156 Gwanggyo-ro, Yeongtong-gu, Suwon 16506, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3717-2223","authenticated-orcid":false,"given":"Joonhyung","family":"Kim","sequence":"additional","affiliation":[{"name":"Department of Ophthalmology, CHA Bundang Medical Center, CHA University School of Medicine, 59 Yatap-ro, Bundang-gu, Seongnam 13496, Republic of Korea"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1968","published-online":{"date-parts":[[2026,8,12]]},"reference":[{"key":"ref_1","doi-asserted-by":"crossref","first-page":"723","DOI":"10.1016\/j.ophtha.2018.11.025","article-title":"Tolerating Subretinal Fluid in Neovascular Age-Related Macular Degeneration Treated with Ranibizumab Using a Treat-and-Extend Regimen: FLUID Study 24-Month Results","volume":"126","author":"Guymer","year":"2019","journal-title":"Ophthalmology"},{"key":"ref_2","doi-asserted-by":"crossref","first-page":"291","DOI":"10.1016\/j.oret.2021.12.007","article-title":"Impact of Intra- and Subretinal Fluid on Vision Based on Volume Quantification in the HARBOR Trial","volume":"6","author":"Riedl","year":"2022","journal-title":"Ophthalmol. Retin."},{"key":"ref_3","doi-asserted-by":"crossref","unstructured":"Bogunovi\u0107, H., Mares, V., Reiter, G.S., and Schmidt-Erfurth, U. (2022). Predicting treat-and-extend outcomes and treatment intervals in neovascular age-related macular degeneration from retinal optical coherence tomography using artificial intelligence. Front. Med., 9.","DOI":"10.3389\/fmed.2022.958469"},{"key":"ref_4","doi-asserted-by":"crossref","first-page":"24","DOI":"10.1016\/j.oret.2017.03.015","article-title":"Machine Learning to Analyze the Prognostic Value of Current Imaging Biomarkers in Neovascular Age-Related Macular Degeneration","volume":"2","author":"Sadeghipour","year":"2018","journal-title":"Ophthalmol. Retin."},{"key":"ref_5","doi-asserted-by":"crossref","first-page":"4173","DOI":"10.1167\/iovs.17-21878","article-title":"Analyzing and Predicting Visual Acuity Outcomes of Anti-VEGF Therapy by a Longitudinal Mixed Effects Model of Imaging and Clinical Data","volume":"58","author":"Vogl","year":"2017","journal-title":"Investig. Ophthalmol. Vis. Sci."},{"key":"ref_6","doi-asserted-by":"crossref","first-page":"15434","DOI":"10.1038\/s41598-020-72359-y","article-title":"Prediction of age-related macular degeneration disease using a sequential deep learning approach on longitudinal SD-OCT imaging biomarkers","volume":"10","author":"Banerjee","year":"2020","journal-title":"Sci. Rep."},{"key":"ref_7","doi-asserted-by":"crossref","first-page":"e106","DOI":"10.1016\/S2214-109X(13)70145-1","article-title":"Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: A systematic review and meta-analysis","volume":"2","author":"Wong","year":"2014","journal-title":"Lancet Glob. Health"},{"key":"ref_8","doi-asserted-by":"crossref","first-page":"1858","DOI":"10.1109\/TMI.2019.2901398","article-title":"RETOUCH: The Retinal OCT Fluid Detection and Segmentation Benchmark and Challenge","volume":"38","author":"Venhuizen","year":"2019","journal-title":"IEEE Trans. Med. Imaging"},{"key":"ref_9","doi-asserted-by":"crossref","first-page":"3627","DOI":"10.1364\/BOE.8.003627","article-title":"ReLayNet: Retinal Layer and Fluid Segmentation of Macular Optical Coherence Tomography using Fully Convolutional Network","volume":"8","author":"Roy","year":"2017","journal-title":"Biomed. Opt. Express"},{"key":"ref_10","doi-asserted-by":"crossref","first-page":"5871","DOI":"10.1038\/s41598-022-09642-7","article-title":"Prediction of treatment outcome in neovascular age-related macular degeneration using a novel convolutional neural network","volume":"12","author":"Yeh","year":"2022","journal-title":"Sci. Rep."},{"key":"ref_11","doi-asserted-by":"crossref","first-page":"28253","DOI":"10.1038\/s41598-024-79034-6","article-title":"Anti-VEGF treatment outcome prediction based on optical coherence tomography images in neovascular age-related macular degeneration using a deep neural network","volume":"14","author":"Han","year":"2024","journal-title":"Sci. Rep."},{"key":"ref_12","doi-asserted-by":"crossref","first-page":"103942","DOI":"10.1016\/j.media.2026.103942","article-title":"Predicting Diabetic Macular Edema Treatment Responses Using OCT: Dataset and Methods of APTOS Competition","volume":"109","author":"Zhang","year":"2026","journal-title":"Med. Image Anal."},{"key":"ref_13","doi-asserted-by":"crossref","first-page":"532","DOI":"10.1038\/s41746-025-01893-8","article-title":"Specialized curricula for training vision-language models in retinal image analysis","volume":"8","author":"Holland","year":"2025","journal-title":"npj Digit. Med."},{"key":"ref_14","doi-asserted-by":"crossref","unstructured":"Li, C., Wong, C., Zhang, S., Usuyama, N., Liu, H., Yang, J., Naumann, T., Poon, H., and Gao, J. (2023). LLaVA-Med: Training a Large Language-and-Vision Assistant for Biomedicine in One Day. Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), Curran Associates, Inc.","DOI":"10.52202\/075280-1240"},{"key":"ref_15","first-page":"353","article-title":"Med-Flamingo: A Multimodal Medical Few-Shot Learner","volume":"Volume 225","author":"Moor","year":"2023","journal-title":"Proceedings of the 3rd Machine Learning for Health Symposium (ML4H)"},{"key":"ref_16","doi-asserted-by":"crossref","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":"ref_17","doi-asserted-by":"crossref","unstructured":"Wang, M.H. (2025). Explainable Artificial Intelligence Framework for Predicting Treatment Outcomes in Age-Related Macular Degeneration. Sensors, 25.","DOI":"10.3390\/s25226879"},{"key":"ref_18","unstructured":"Qwen Team (2026, July 29). Qwen3.6-27B Model Card and Checkpoint. Hugging Face. Available online: https:\/\/huggingface.co\/Qwen\/Qwen3.6-27B."},{"key":"ref_19","doi-asserted-by":"crossref","first-page":"e55825","DOI":"10.2196\/55825","article-title":"Diagnosis of Chronic Kidney Disease Using Retinal Imaging and Urine Dipstick Data: Multimodal Deep Learning Approach","volume":"13","author":"Bhak","year":"2025","journal-title":"JMIR Med. Inform."},{"key":"ref_20","doi-asserted-by":"crossref","first-page":"234","DOI":"10.1001\/jamaophthalmol.2022.5928","article-title":"Neurologic Dysfunction Assessment in Parkinson Disease Based on Fundus Photographs Using Deep Learning","volume":"141","author":"Ahn","year":"2023","journal-title":"JAMA Ophthalmol."},{"key":"ref_21","doi-asserted-by":"crossref","unstructured":"Xia, P., Chen, Z., Tian, J., Gong, Y., Hou, R., Xu, Y., Wu, Z., Fan, Z., Zhou, Y., and Zhu, K. (2024). CARES: A Comprehensive Benchmark of Trustworthiness in Medical Vision Language Models. Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track, Curran Associates, Inc.","DOI":"10.52202\/079017-4455"},{"key":"ref_22","unstructured":"Chang, A., Huang, L., Bhatia, P., Kass-Hout, T., Ma, F., and Xiao, C. (2025). MedHEval: Benchmarking Hallucinations and Mitigation Strategies in Medical Large Vision-Language Models. arXiv."},{"key":"ref_23","unstructured":"Buckley, T., Diao, J.A., Rajpurkar, P., Rodman, A., and Manrai, A.K. (2023). Multimodal Foundation Models Exploit Text to Make Medical Image Predictions. arXiv."},{"key":"ref_24","doi-asserted-by":"crossref","first-page":"74952","DOI":"10.52202\/075280-3275","article-title":"Language Models Don\u2019t Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting","volume":"Volume 36","author":"Turpin","year":"2023","journal-title":"Proceedings of the Advances in Neural Information Processing Systems (NeurIPS)"},{"key":"ref_25","unstructured":"Lanham, T., Chen, A., Radhakrishnan, A., Steiner, B., Denison, C., Hernandez, D., Li, D., Durmus, E., Hubinger, E., and Kernion, J. (2023). Measuring Faithfulness in Chain-of-Thought Reasoning. arXiv."},{"key":"ref_26","unstructured":"Chen, Y., Benton, J., Radhakrishnan, A., Uesato, J., Denison, C., Schulman, J., Somani, A., Hase, P., Wagner, M., and Roger, F. (2025). Reasoning Models Don\u2019t Always Say What They Think. arXiv."},{"key":"ref_27","doi-asserted-by":"crossref","unstructured":"Balasubramanian, S., Basu, S., and Feizi, S. (2025). A Closer Look at Bias and Chain-of-Thought Faithfulness of Large (Vision) Language Models. arXiv.","DOI":"10.18653\/v1\/2025.findings-emnlp.723"},{"key":"ref_28","doi-asserted-by":"crossref","unstructured":"Atanasova, P., Camburu, O.M., Lioma, C., Lukasiewicz, T., Simonsen, J.G., and Augenstein, I. (2023). Faithfulness Tests for Natural Language Explanations. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), Association for Computational Linguistics.","DOI":"10.18653\/v1\/2023.acl-short.25"},{"key":"ref_29","doi-asserted-by":"crossref","unstructured":"Jacovi, A., and Goldberg, Y. (2020). Towards Faithfully Interpretable NLP Systems: How Should We Define and Evaluate Faithfulness?. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL), Association for Computational Linguistics.","DOI":"10.18653\/v1\/2020.acl-main.386"},{"key":"ref_30","doi-asserted-by":"crossref","unstructured":"Holtzman, A., West, P., Shwartz, V., Choi, Y., and Zettlemoyer, L. (2021). Surface Form Competition: Why the Highest Probability Answer Is Not Always Right. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP), Association for Computational Linguistics.","DOI":"10.18653\/v1\/2021.emnlp-main.564"},{"key":"ref_31","first-page":"12697","article-title":"Calibrate Before Use: Improving Few-Shot Performance of Language Models","volume":"Volume 139","author":"Zhao","year":"2021","journal-title":"Proceedings of the 38th International Conference on Machine Learning (ICML)"},{"key":"ref_32","unstructured":"Jain, S., and Wallace, B.C. (2019). Attention Is Not Explanation. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT), Association for Computational Linguistics."},{"key":"ref_33","doi-asserted-by":"crossref","unstructured":"Wiegreffe, S., and Pinter, Y. (2019). Attention is not not Explanation. Proceedings of the EMNLP-IJCNLP, Association for Computational Linguistics.","DOI":"10.18653\/v1\/D19-1002"},{"key":"ref_34","doi-asserted-by":"crossref","unstructured":"Abnar, S., and Zuidema, W. (2020). Quantifying Attention Flow in Transformers. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL), Association for Computational Linguistics.","DOI":"10.18653\/v1\/2020.acl-main.385"},{"key":"ref_35","doi-asserted-by":"crossref","unstructured":"Chefer, H., Gur, S., and Wolf, L. (2021). Transformer Interpretability Beyond Attention Visualization. Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), IEEE.","DOI":"10.1109\/CVPR46437.2021.00084"},{"key":"ref_36","unstructured":"Adebayo, J., Gilmer, J., Muelly, M., Goodfellow, I., Hardt, M., and Kim, B. (2018). Sanity Checks for Saliency Maps. Proceedings of the Advances in Neural Information Processing Systems (NeurIPS), Curran Associates, Inc."},{"key":"ref_37","doi-asserted-by":"crossref","first-page":"e200267","DOI":"10.1148\/ryai.2021200267","article-title":"Assessing the Trustworthiness of Saliency Maps for Localizing Abnormalities in Medical Imaging","volume":"3","author":"Arun","year":"2021","journal-title":"Radiol. Artif. Intell."},{"key":"ref_38","doi-asserted-by":"crossref","first-page":"867","DOI":"10.1038\/s42256-022-00536-x","article-title":"Benchmarking saliency methods for chest X-ray interpretation","volume":"4","author":"Saporta","year":"2022","journal-title":"Nat. Mach. Intell."},{"key":"ref_39","doi-asserted-by":"crossref","first-page":"665","DOI":"10.1038\/s42256-020-00257-z","article-title":"Shortcut Learning in Deep Neural Networks","volume":"2","author":"Geirhos","year":"2020","journal-title":"Nat. Mach. Intell."},{"key":"ref_40","doi-asserted-by":"crossref","first-page":"610","DOI":"10.1038\/s42256-021-00338-7","article-title":"AI for Radiographic COVID-19 Detection Selects Shortcuts over Signal","volume":"3","author":"DeGrave","year":"2021","journal-title":"Nat. Mach. Intell."},{"key":"ref_41","doi-asserted-by":"crossref","first-page":"206","DOI":"10.1038\/s42256-019-0048-x","article-title":"Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead","volume":"1","author":"Rudin","year":"2019","journal-title":"Nat. Mach. Intell."},{"key":"ref_42","doi-asserted-by":"crossref","first-page":"27730","DOI":"10.52202\/068431-2011","article-title":"Training Language Models to Follow Instructions with Human Feedback","volume":"Volume 35","author":"Ouyang","year":"2022","journal-title":"Proceedings of the Advances in Neural Information Processing Systems (NeurIPS)"},{"key":"ref_43","unstructured":"Biderman, D., Portes, J., Ortiz, J.J.G., Paul, M., Greengard, P., Jennings, C., King, D., Havens, S., Chiley, V., and Frankle, J. (2024). LoRA Learns Less and Forgets Less. arXiv."},{"key":"ref_44","doi-asserted-by":"crossref","first-page":"109","DOI":"10.1016\/S0079-7421(08)60536-8","article-title":"Catastrophic Interference in Connectionist Networks: The Sequential Learning Problem","volume":"Volume 24","author":"McCloskey","year":"1989","journal-title":"Psychology of Learning and Motivation"},{"key":"ref_45","doi-asserted-by":"crossref","first-page":"3521","DOI":"10.1073\/pnas.1611835114","article-title":"Overcoming Catastrophic Forgetting in Neural Networks","volume":"114","author":"Kirkpatrick","year":"2017","journal-title":"Proc. Natl. Acad. Sci. USA"},{"key":"ref_46","unstructured":"Luo, Y., Yang, Z., Meng, F., Li, Y., Zhou, J., and Zhang, Y. (2023). An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuning. arXiv."},{"key":"ref_47","unstructured":"Chu, T., Zhai, Y., Yang, J., Tong, S., Xie, S., Schuurmans, D., Le, Q.V., Levine, S., and Ma, Y. (2025). SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training. Proceedings of the International Conference on Machine Learning (ICML), PMLR."},{"key":"ref_48","doi-asserted-by":"crossref","unstructured":"Li, K., Wu, Z., Peng, K.C., Ernst, J., and Fu, Y. (2018). Tell Me Where to Look: Guided Attention Inference Network. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), IEEE Computer Society.","DOI":"10.1109\/CVPR.2018.00960"},{"key":"ref_49","doi-asserted-by":"crossref","unstructured":"Ross, A.S., Hughes, M.C., and Doshi-Velez, F. (2017). Right for the Right Reasons: Training Differentiable Models by Constraining Their Explanations. Proceedings of the Twenty-Sixth International Joint Conference on Artificial Intelligence (IJCAI), AAAI Press.","DOI":"10.24963\/ijcai.2017\/371"},{"key":"ref_50","doi-asserted-by":"crossref","first-page":"20","DOI":"10.21561\/jor.2025.10.1.20","article-title":"Short-term Efficacy and Safety of Faricimab Switching in Neovascular Age-related Macular Degeneration Previously Treated with Anti-vascular Endothelial Growth Factor","volume":"10","author":"Lee","year":"2025","journal-title":"J. Retin."},{"key":"ref_51","first-page":"103686","article-title":"Batch Image Encryption Using Cross Image Permutation and Diffusion","volume":"80","author":"Song","year":"2024","journal-title":"J. Inf. Secur. Appl."}],"container-title":["Bioengineering"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/2306-5354\/13\/8\/912\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,8,12]],"date-time":"2026-08-12T13:23:05Z","timestamp":1786540985000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/2306-5354\/13\/8\/912"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,8,12]]},"references-count":51,"journal-issue":{"issue":"8","published-online":{"date-parts":[[2026,8]]}},"alternative-id":["bioengineering13080912"],"URL":"https:\/\/doi.org\/10.3390\/bioengineering13080912","relation":{},"ISSN":["2306-5354"],"issn-type":[{"value":"2306-5354","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,8,12]]}}}