{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,21]],"date-time":"2026-07-21T22:17:13Z","timestamp":1784672233801,"version":"3.55.0"},"reference-count":42,"publisher":"Springer Science and Business Media LLC","issue":"1","license":[{"start":{"date-parts":[[2026,2,25]],"date-time":"2026-02-25T00:00:00Z","timestamp":1771977600000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,4,2]],"date-time":"2026-04-02T00:00:00Z","timestamp":1775088000000},"content-version":"vor","delay-in-days":36,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["npj Digit. Med."],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Considerations around model retraining are standard practice in industry and non-healthcare sectors; however, this is much less well explored in medical artificial intelligence (AI). The problem is not only that models often fail to generalise, but that academia in particular does not have a systematic science of retraining to address this gap. This matters for building trustworthy models capable of making a lasting impact, rather than compounding as research waste. In this Perspective, we highlight three common challenges that constrain model retraining in medicine, and argue that academia must evolve beyond a focus on developing proofs-of-concept and world-first innovations to also recognise model retraining as scholarship. Drawing from case examples in ophthalmology, we call on stakeholders to consider not just how we build AI models, but how we should retrain, maintain, and share them.<\/jats:p>","DOI":"10.1038\/s41746-026-02463-2","type":"journal-article","created":{"date-parts":[[2026,2,25]],"date-time":"2026-02-25T11:18:31Z","timestamp":1772018311000},"update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":3,"title":["Considering the missing science of retraining and maintenance in medical artificial intelligence, using ophthalmology as an exemplar"],"prefix":"10.1038","volume":"9","author":[{"given":"Ariel Yuhan","family":"Ong","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Robbert R.","family":"Struyven","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Alastair K.","family":"Denniston","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"David A.","family":"Merle","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Justin","family":"Engelmann","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Hyunmin","family":"Kim","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yukun","family":"Zhou","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Pearse A.","family":"Keane","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ines","family":"Lains","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2026,2,25]]},"reference":[{"key":"2463_CR1","doi-asserted-by":"publisher","DOI":"10.1186\/s12911-021-01488-9","volume":"21","author":"S Secinaro","year":"2021","unstructured":"Secinaro, S., Calandra, D., Secinaro, A., Muthurangu, V. & Biancone, P. The role of artificial intelligence in healthcare: a structured literature review. BMC Med. Inform. Decis. Mak. 21, 125 (2021).","journal-title":"BMC Med. Inform. Decis. Mak."},{"key":"2463_CR2","doi-asserted-by":"publisher","first-page":"47","DOI":"10.1038\/s41586-023-06221-2","volume":"620","author":"H Wang","year":"2023","unstructured":"Wang, H. et al. Scientific discovery in the age of artificial intelligence. Nature 620, 47\u201360 (2023).","journal-title":"Nature"},{"key":"2463_CR3","doi-asserted-by":"publisher","first-page":"40","DOI":"10.1038\/s41746-018-0048-y","volume":"1","author":"PA Keane","year":"2018","unstructured":"Keane, P. A. & Topol, E. J. With an eye to AI and autonomous diagnosis. npj Digit. Med. 1, 40 (2018).","journal-title":"npj Digit. Med."},{"key":"2463_CR4","doi-asserted-by":"publisher","unstructured":"Youssef, A. et al. External validation of AI models in health should be replaced with recurring local validation. Nat. Med. 1\u20132 https:\/\/doi.org\/10.1038\/s41591-023-02540-z (2023).","DOI":"10.1038\/s41591-023-02540-z"},{"key":"2463_CR5","doi-asserted-by":"publisher","first-page":"563","DOI":"10.9734\/jsrr\/2025\/v31i123797","volume":"31","author":"J Oduro-Gyan","year":"2025","unstructured":"Oduro-Gyan, J., Eleweke, I., Ajuwon, S., Bello, A. & Arotayo, A.-L. Embedding responsible AI into MLOps pipelines: ensuring fairness, explainability, and governance in KYC and FinTech decisioning. J. Sci. Res. Rep. 31, 563\u2013581 (2025).","journal-title":"J. Sci. Res. Rep."},{"key":"2463_CR6","unstructured":"Reddi, V. J. Chapter 13: ML Operations. In Machine Learning Systems: Principles and Practices of Engineering Artificially Intelligent Systems https:\/\/mlsysbook.ai\/book\/."},{"key":"2463_CR7","doi-asserted-by":"publisher","first-page":"421","DOI":"10.1016\/j.mcpdig.2024.06.009","volume":"2","author":"A Rajagopal","year":"2024","unstructured":"Rajagopal, A. et al. Machine learning operations in health care: a scoping review. Mayo Clin. Proc. Digit. Health 2, 421\u2013437 (2024).","journal-title":"Mayo Clin. Proc. Digit. Health"},{"key":"2463_CR8","doi-asserted-by":"publisher","DOI":"10.1038\/sdata.2016.18","volume":"3","author":"MD Wilkinson","year":"2016","unstructured":"Wilkinson, M. D. et al. The FAIR Guiding Principles for scientific data management and stewardship. Sci. Data 3, 160018 (2016).","journal-title":"Sci. Data"},{"key":"2463_CR9","doi-asserted-by":"publisher","unstructured":"Mackenzie, A., Loveland, J. & van Engen, R. Survey of image processing settings used for mammography systems in the United Kingdom: How variable is it? Preprint at arXiv.org https:\/\/arxiv.org\/abs\/2501.13595v1; https:\/\/doi.org\/10.1117\/12.3026876 (2025).","DOI":"10.1117\/12.3026876"},{"key":"2463_CR10","doi-asserted-by":"publisher","first-page":"969","DOI":"10.1016\/j.ophtha.2021.03.003","volume":"128","author":"AY Lee","year":"2021","unstructured":"Lee, A. Y. et al. Recommendations for standardization of images in ophthalmology. Ophthalmology 128, 969\u2013970 (2021).","journal-title":"Ophthalmology"},{"key":"2463_CR11","doi-asserted-by":"publisher","first-page":"3","DOI":"10.1007\/978-3-030-23937-4_1","volume":"2019","author":"S Besson","year":"2019","unstructured":"Besson, S. et al. Bringing open data to whole slide imaging. Digit. Pathol. 2019, 3\u201310 (2019).","journal-title":"Digit. Pathol."},{"key":"2463_CR12","doi-asserted-by":"publisher","first-page":"800","DOI":"10.1016\/j.echo.2023.04.016","volume":"36","author":"R Arnaout","year":"2023","unstructured":"Arnaout, R. et al. The (heart and) soul of a human creation: designing echocardiography for the big data age. J. Am. Soc. Echocardiogr. 36, 800\u2013801 (2023).","journal-title":"J. Am. Soc. Echocardiogr."},{"key":"2463_CR13","unstructured":"Heidelberg Eye Explorer 2 (HEYEX 2)\u2014Platform for Ophthalmic Image Management and Device Integration. Heidelberg Engineering. https:\/\/business-lounge.heidelbergengineering.com\/us\/en\/products\/heidelberg-eye-explorer\/heyex-2\/#product-details."},{"key":"2463_CR14","doi-asserted-by":"publisher","first-page":"283","DOI":"10.1056\/NEJMc2104626","volume":"385","author":"SG Finlayson","year":"2021","unstructured":"Finlayson, S. G. et al. The clinician and dataset shift in artificial intelligence. N. Engl. J. Med. 385, 283\u2013286 (2021).","journal-title":"N. Engl. J. Med."},{"key":"2463_CR15","doi-asserted-by":"publisher","first-page":"64","DOI":"10.1515\/jci-2021-0042","volume":"10","author":"A Subbaswamy","year":"2022","unstructured":"Subbaswamy, A., Chen, B. & Saria, S. A unifying causal framework for analyzing dataset shift-stable learning algorithms. J. Causal Inference 10, 64\u201389 (2022).","journal-title":"J. Causal Inference"},{"key":"2463_CR16","doi-asserted-by":"crossref","unstructured":"Lee, S., Yin, C. & Zhang, P. Stable clinical risk prediction against distribution shift in electronic health records. Patterns 4, 100828 (2023).","DOI":"10.1016\/j.patter.2023.100828"},{"key":"2463_CR17","doi-asserted-by":"publisher","unstructured":"Avati, A., Xue, E., Xu, Z., Lakshminarayanan, B. & Dai, A. BEDS-Bench: behavior of EHR-models under distributional shift\u2014a benchmark. In Proc. Workshop on Distribution Shifts, 35th Conference on Neural Information Processing Systems https:\/\/doi.org\/10.48550\/arXiv.2107.08189 (2021).","DOI":"10.48550\/arXiv.2107.08189"},{"key":"2463_CR18","doi-asserted-by":"publisher","first-page":"653","DOI":"10.1111\/1755-0998.13309","volume":"21","author":"R De-Kayne","year":"2021","unstructured":"De-Kayne, R. et al. Sequencing platform shifts provide opportunities but pose challenges for combining genomic data sets. Mol. Ecol. Resour. 21, 653\u2013660 (2021).","journal-title":"Mol. Ecol. Resour."},{"key":"2463_CR19","doi-asserted-by":"publisher","first-page":"e279","DOI":"10.1016\/S2589-7500(20)30102-3","volume":"2","author":"CS Lee","year":"2020","unstructured":"Lee, C. S. & Lee, A. Y. Applications of continual learning machine learning in clinical practice. Lancet Digit. Health 2, e279\u2013e281 (2020).","journal-title":"Lancet Digit. Health"},{"key":"2463_CR20","doi-asserted-by":"publisher","unstructured":"Chen, Z. & Liu, B. Lifelong Machine Learning. https:\/\/doi.org\/10.1007\/978-3-031-01581-6 (2018).","DOI":"10.1007\/978-3-031-01581-6"},{"key":"2463_CR21","doi-asserted-by":"publisher","first-page":"e51","DOI":"10.1016\/S2589-7500(20)30240-5","volume":"3","author":"SM Khan","year":"2021","unstructured":"Khan, S. M. et al. A global review of publicly available datasets for ophthalmological imaging: barriers to access, usability, and generalisability. Lancet Digit. Health 3, e51\u2013e66 (2021).","journal-title":"Lancet Digit. Health"},{"key":"2463_CR22","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1038\/s41746-025-01726-8","volume":"8","author":"AY Ong","year":"2025","unstructured":"Ong, A. Y. et al. A scoping review of artificial intelligence as a medical device for ophthalmic image analysis in Europe, Australia and America. npj Digit. Med. 8, 1\u201313 (2025).","journal-title":"npj Digit. Med."},{"key":"2463_CR23","doi-asserted-by":"publisher","first-page":"926","DOI":"10.1056\/NEJM197810262991705","volume":"299","author":"DF Ransohoff","year":"1978","unstructured":"Ransohoff, D. F. & Feinstein, A. R. Problems of spectrum and bias in evaluating the efficacy of diagnosTIC TESTs. N. Engl. J. Med. 299, 926\u2013930 (1978).","journal-title":"N. Engl. J. Med."},{"key":"2463_CR24","doi-asserted-by":"publisher","unstructured":"Ong, A. Y., Hogg, H. D. J. & Keane, P. A. Cochrane corner: artificial intelligence for diagnosing exudative age-related macular degeneration. Eye 1\u20132 https:\/\/doi.org\/10.1038\/s41433-025-03599-3 (2025).","DOI":"10.1038\/s41433-025-03599-3"},{"key":"2463_CR25","first-page":"2922","volume":"64","author":"J Engelmann","year":"2023","unstructured":"Engelmann, J., Storkey, A. & Bernabeu LLinares, M. Exclusion of poor quality fundus images biases health research linking retinal traits and systemic health. Investig. Ophthalmol. Vis. Sci. 64, 2922 (2023).","journal-title":"Investig. Ophthalmol. Vis. Sci."},{"key":"2463_CR26","doi-asserted-by":"publisher","first-page":"e220081","DOI":"10.1148\/ryai.220081","volume":"4","author":"K Venkatesh","year":"2022","unstructured":"Venkatesh, K., Santomartino, S. M., Sulam, J. & Yi, P. H. Code and data sharing practices in the radiology artificial intelligence literature: a meta-research study. Radio. Artif. Intell. 4, e220081 (2022).","journal-title":"Radio. Artif. Intell."},{"key":"2463_CR27","doi-asserted-by":"publisher","unstructured":"Lee, T., Lee, J. H., Yoon, S. H., Park, S. H. & Kim, H. Availability and transparency of artificial intelligence models in radiology: a meta-research study. Eur. Radiol. https:\/\/doi.org\/10.1007\/s00330-025-11492-6 (2025).","DOI":"10.1007\/s00330-025-11492-6"},{"key":"2463_CR28","doi-asserted-by":"publisher","unstructured":"Zhou, Y. et al. A foundation model for generalizable disease detection from retinal images. Nature 1\u20138 https:\/\/doi.org\/10.1038\/s41586-023-06555-x (2023).","DOI":"10.1038\/s41586-023-06555-x"},{"key":"2463_CR29","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-025-62123-z","volume":"16","author":"J Engelmann","year":"2025","unstructured":"Engelmann, J. & Bernabeu, M. O. Training a high-performance retinal foundation model with half-the-data and 400 times less compute. Nat. Commun. 16, 6862 (2025).","journal-title":"Nat. Commun."},{"key":"2463_CR30","unstructured":"vLLM. https:\/\/docs.vllm.ai\/en\/latest\/."},{"key":"2463_CR31","doi-asserted-by":"crossref","unstructured":"Hintersdorf, D., Struppek, L. & Kersting, K. Balancing Transparency and Risk: An Overview of the Security and Privacy Risks of Open-Source Machine Learning Models. In Bridging the Gap Between AI and Reality. AISoLA 2023. Lecture Notes in Computer Science, (eds Steff en, B.) Vol. 14129 (Springer, Cham, 2023).","DOI":"10.1007\/978-3-031-73741-1_16"},{"key":"2463_CR32","first-page":"98:1","volume":"58","author":"J Wei","year":"2025","unstructured":"Wei, J. et al. Memorization in deep learning: a survey. ACM Comput. Surv. 58, 98:1\u201398:35 (2025).","journal-title":"ACM Comput. Surv."},{"key":"2463_CR33","doi-asserted-by":"crossref","unstructured":"Mohammadi, M. et al. Differential privacy for medical deep learning: methods, tradeoffs, and deployment implications. npj Digit. Med. 9, 93 (2026).","DOI":"10.1038\/s41746-025-02280-z"},{"key":"2463_CR34","doi-asserted-by":"publisher","first-page":"1132","DOI":"10.1038\/s41592-021-01256-7","volume":"18","author":"BJ Heil","year":"2021","unstructured":"Heil, B. J. et al. Reproducibility standards for machine learning in the life sciences. Nat. Methods 18, 1132\u20131135 (2021).","journal-title":"Nat. Methods"},{"key":"2463_CR35","doi-asserted-by":"publisher","first-page":"111610","DOI":"10.1016\/j.knosys.2024.111610","volume":"293","author":"A Mahadevan","year":"2024","unstructured":"Mahadevan, A. & Mathioudakis, M. Cost-aware retraining for machine learning. Knowl.-Based Syst. 293, 111610 (2024).","journal-title":"Knowl.-Based Syst."},{"key":"2463_CR36","unstructured":"Sculley, D. et al. Hidden technical debt in machine learning systems. In Proc. Advances in Neural Information Processing Systems Vol. 28 (Curran Associates, Inc., 2015)."},{"key":"2463_CR37","doi-asserted-by":"publisher","first-page":"111907","DOI":"10.1016\/j.jss.2023.111907","volume":"209","author":"A Serban","year":"2024","unstructured":"Serban, A., van der Blom, K., Hoos, H. & Visser, J. Software engineering practices for machine learning\u2014adoption, effects, and team assessment. J. Syst. Softw. 209, 111907 (2024).","journal-title":"J. Syst. Softw."},{"key":"2463_CR38","doi-asserted-by":"publisher","first-page":"AIoa2500522","DOI":"10.1056\/AIoa2500522","volume":"3","author":"E Chen","year":"2025","unstructured":"Chen, E. et al. International retrospective observational study of continual learning for AI on endotracheal tube placement from chest radiographs. NEJM AI 3, AIoa2500522 (2025).","journal-title":"NEJM AI"},{"key":"2463_CR39","doi-asserted-by":"publisher","first-page":"6","DOI":"10.1167\/iovs.65.6.6","volume":"65","author":"J Engelmann","year":"2024","unstructured":"Engelmann, J. et al. Choroidalyzer: an open-source, end-to-end pipeline for choroidal analysis in optical coherence tomography. Investig. Ophthalmol. Vis. Sci. 65, 6 (2024).","journal-title":"Investig. Ophthalmol. Vis. Sci."},{"key":"2463_CR40","doi-asserted-by":"publisher","first-page":"1342","DOI":"10.1038\/s41591-018-0107-6","volume":"24","author":"J De Fauw","year":"2018","unstructured":"De Fauw, J. et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat. Med. 24, 1342\u20131350 (2018).","journal-title":"Nat. Med."},{"key":"2463_CR41","unstructured":"Tazbaz, T. & Nicol, J. Blog: a lifecycle management approach toward delivering safe, effective AI-enabled health care. FDA. https:\/\/www.fda.gov\/medical-devices\/digital-health-center-excellence\/blog-lifecycle-management-approach-toward-delivering-safe-effective-ai-enabled-health-care."},{"key":"2463_CR42","doi-asserted-by":"crossref","unstructured":"Xie, Q., Luong, M.-T., Hovy, E. & Le, Q. V. Self-training with Noisy Student improves ImageNet classification. 2020 IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 10684\u201310695, (Seattle, WA, USA, 2020).","DOI":"10.1109\/CVPR42600.2020.01070"}],"container-title":["npj Digital Medicine"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.nature.com\/articles\/s41746-026-02463-2","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s41746-026-02463-2.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s41746-026-02463-2.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,4,2]],"date-time":"2026-04-02T10:39:17Z","timestamp":1775126357000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.nature.com\/articles\/s41746-026-02463-2"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,2,25]]},"references-count":42,"journal-issue":{"issue":"1","published-online":{"date-parts":[[2026,12]]}},"alternative-id":["2463"],"URL":"https:\/\/doi.org\/10.1038\/s41746-026-02463-2","relation":{},"ISSN":["2398-6352"],"issn-type":[{"value":"2398-6352","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,2,25]]},"assertion":[{"value":"8 November 2025","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"11 February 2026","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"25 February 2026","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"P.A.K. is a cofounder of Cascader Ltd. and has acted as a consultant for Retina Consultants of America, Roche, Boehringer-Ingelheim, and Bitfount and is an equity owner in Big Picture Medical. He has received speaker fees from Zeiss, Thea, Apellis, and Roche. He has received travel support from Bayer and Roche. He has attended advisory boards for Topcon, Bayer, Boehringer-Ingelheim, and Roche. I.L. has attended advisory boards for EyePoint and 4D Molecular Therapeutics, Inc. The remaining authors do not have any conflicts of interest to declare.","order":1,"name":"Ethics","group":{"name":"EthicsHeading","label":"Competing interests"}}],"article-number":"277"}}