{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,16]],"date-time":"2026-09-16T21:44:04Z","timestamp":1789595044459,"version":"build-2803163510"},"reference-count":42,"publisher":"Springer Science and Business Media LLC","issue":"1","license":[{"start":{"date-parts":[[2026,1,5]],"date-time":"2026-01-05T00:00:00Z","timestamp":1767571200000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2026,1,26]],"date-time":"2026-01-26T00:00:00Z","timestamp":1769385600000},"content-version":"vor","delay-in-days":21,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0"}],"funder":[{"DOI":"10.13039\/501100002806","name":"Carl Zeiss Meditec AG","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100002806","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["link.springer.com"],"crossmark-restriction":false},"short-container-title":["Sci Rep"],"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>Optical coherence tomography (OCT) enables early detection of vision-threatening diabetic retinopathy (DR) and retinal fluid accumulation, both major complications of diabetes. Despite high classification performance, deep learning models face limited clinical adoption due to poor interpretability. While attribution methods are effective for explaining predictions in the natural image domain, their applicability to medical imaging remains underexplored. To bridge this gap, our work explores how well these strong results transfer to the medical imaging domain. This study evaluates three cutting-edge attribution methods\u2014 DeepLIFT, AGI, and AttEXplore \u2014for explaining predictions of a VGG16-based deep learning model in DR classification using widefield OCTA en face images and fluid detection in OCT B-scans. We assess attribution methods\u2019 ability to highlight clinically relevant regions using quantitative (insertion and deletion scores) and qualitative (heatmap visual analysis) measures. Although the VGG16 model achieves high classification accuracy (94% for DR and 98% for fluid), attribution methods yield markedly different qualitative results due to variations in underlying assumptions and hyperparameter sensitivity. Additionally, high insertion or low deletion scores do not necessarily correlate with clinically meaningful visual attributions. In particular, insertion- and deletion-based behaviour can be more informative in pathological cases, where localized lesions can drive predictions, but tends to be less informative in normal cases, where confirming the absence of pathology requires global contextual evidence. Comparing the three approaches, we find that AGI and AttEXplore achieve similarly strong quantitative performance, whereas AttEXplore more consistently highlights clinically meaningful structures in pathological cases than AGI and DeepLIFT, making it the preferred option in our DR and fluid detection settings. However, our results also show that in any potential clinical usage, all three methods must be interpreted alongside clinical expertise, and even the result of the best-performing attribution approach cannot serve as an automatic proxy for clinical relevance. This study provides critical insights into challenges of applying attribution methods to medical imaging, using ophthalmic data, laying a foundation for improving transparency and trust in AI-assisted diagnostics.<\/jats:p>","DOI":"10.1038\/s41598-025-33120-5","type":"journal-article","created":{"date-parts":[[2026,1,5]],"date-time":"2026-01-05T16:13:04Z","timestamp":1767629584000},"update-policy":"https:\/\/doi.org\/10.1007\/springer_crossmark_policy","source":"Crossref","is-referenced-by-count":3,"title":["Applicability study of AI attribution methods for ophthalmic image classification"],"prefix":"10.1038","volume":"16","author":[{"given":"Ali","family":"Yavari","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Tilman","family":"Schmoll","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Rainer A.","family":"Leitgeb","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Kim Lien","family":"Huber","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Heiko","family":"Stino","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Andreas","family":"Pollreisz","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Wolfgang","family":"Drexler","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Thomas","family":"Schlegl","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"297","published-online":{"date-parts":[[2026,1,5]]},"reference":[{"key":"33120_CR1","doi-asserted-by":"publisher","DOI":"10.1016\/j.eswa.2024.124710","volume":"255","author":"A Abusitta","year":"2024","unstructured":"Abusitta, A., Li, M. Q. & Fung, B. C. Survey on explainable ai: Techniques, challenges and open issues. Expert Syst. Appl. 255, 124710. https:\/\/doi.org\/10.1016\/j.eswa.2024.124710 (2024).","journal-title":"Expert Syst. Appl."},{"key":"33120_CR2","doi-asserted-by":"publisher","DOI":"10.1016\/j.compbiomed.2023.106668","volume":"156","author":"S Nazir","year":"2023","unstructured":"Nazir, S., Dickson, D. M. & Akram, M. U. Survey of explainable artificial intelligence techniques for biomedical imaging with deep neural networks. Comput. Biol. Med. 156, 106668. https:\/\/doi.org\/10.1016\/j.compbiomed.2023.106668 (2023).","journal-title":"Comput. Biol. Med."},{"key":"33120_CR3","unstructured":"Zhu, Z. et al. Attexplore: Attribution for explanation with model parameters exploration. In The Twelfth International Conference on Learning Representations (2024)."},{"key":"33120_CR4","doi-asserted-by":"publisher","DOI":"10.1016\/j.artmed.2023.102616","volume":"143","author":"J Allgaier","year":"2023","unstructured":"Allgaier, J., Mulansky, L., Draelos, R. L. & Pryss, R. How does the model make predictions? a systematic literature review on the explainability power of machine learning in healthcare. Artif. Intell. Med. 143, 102616. https:\/\/doi.org\/10.1016\/j.artmed.2023.102616 (2023).","journal-title":"Artif. Intell. Med."},{"key":"33120_CR5","unstructured":"Shrikumar, A., Greenside, P. & Kundaje, A. Learning important features through propagating activation differences. In Proceedings of the 34th International Conference on Machine Learning - Volume 70, ICML\u201917, 3145\u20133153 (JMLR.org, 2017)."},{"key":"33120_CR6","doi-asserted-by":"crossref","unstructured":"Pan, D., Li, X. & Zhu, D. Explaining deep neural network models with adversarial gradient integration. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence (IJCAI-21) (International Joint Conferences on Artificial Intelligence Organization, USA, 2021). Conference Paper.","DOI":"10.24963\/ijcai.2021\/396"},{"key":"33120_CR7","unstructured":"Simonyan, K. & Zisserman, A. Very deep convolutional networks for large-scale image recognition (2015). arXiv:1409.1556."},{"key":"33120_CR8","doi-asserted-by":"crossref","unstructured":"Ribeiro, M.\u00a0T., Singh, S. & Guestrin, C. \u201cwhy should i trust you?\u201d: Explaining the predictions of any classifier (2016). arXiv:1602.04938.","DOI":"10.18653\/v1\/N16-3020"},{"key":"33120_CR9","unstructured":"Sundararajan, M., Taly, A. & Yan, Q. Axiomatic attribution for deep networks (2017). arXiv:1703.01365."},{"key":"33120_CR10","unstructured":"Zhu, Z. et al. Mfaba: A more faithful and accelerated boundary-based attribution method for deep neural networks (2023). arXiv:2312.13630."},{"key":"33120_CR11","doi-asserted-by":"publisher","unstructured":"Lopatina, A., Ropele, S., Sibgatulin, R., Reichenbach, J.\u00a0R. & G\u00fcllmar, D. Investigation of deep-learning-driven identification of multiple sclerosis patients based on susceptibility-weighted images using relevance analysis. Frontiers in NeuroscienceVolume 14 - 2020, https:\/\/doi.org\/10.3389\/fnins.2020.609468(2020).","DOI":"10.3389\/fnins.2020.609468"},{"key":"33120_CR12","doi-asserted-by":"publisher","DOI":"10.1016\/j.compbiomed.2021.104578","volume":"135","author":"LA de Souza","year":"2021","unstructured":"de Souza, L. A. et al. Convolutional neural networks for the evaluation of cancer in barrett\u2019s esophagus: Explainable ai to lighten up the black-box. Comput. Biol. Med. 135, 104578. https:\/\/doi.org\/10.1016\/j.compbiomed.2021.104578 (2021).","journal-title":"Comput. Biol. Med."},{"key":"33120_CR13","doi-asserted-by":"publisher","unstructured":"Simonyan, K., Vedaldi, A. & Zisserman, A. Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv:1312.6034, https:\/\/doi.org\/10.48550\/arXiv.1312.6034(2013).","DOI":"10.48550\/arXiv.1312.6034"},{"key":"33120_CR14","unstructured":"Springenberg, J.\u00a0T., Dosovitskiy, A., Brox, T. & Riedmiller, M.\u00a0A. Striving for simplicity: The all convolutional net. In 3rd International Conference on Learning Representations (ICLR) (2015). Workshop Track Proceedings."},{"key":"33120_CR15","doi-asserted-by":"publisher","unstructured":"Binder, A., Montavon, G., Bach, S., M\u00fcller, K.-R. & Samek, W. Layer-wise relevance propagation for neural networks with local renormalization layers. arXiv:1604.00825, https:\/\/doi.org\/10.48550\/arXiv.1604.00825(2016).","DOI":"10.48550\/arXiv.1604.00825"},{"key":"33120_CR16","doi-asserted-by":"crossref","unstructured":"Eitel, F. & Ritter, K. Testing the robustness of attribution methods for convolutional neural networks in mri-based alzheimer\u2019s disease classification. In Suzuki, K. et al. (eds.) Interpretability of Machine Intelligence in Medical Image Computing and Multimodal Learning for Clinical Decision Support, 3\u201311 (Springer International Publishing, Cham, 2019).","DOI":"10.1007\/978-3-030-33850-3_1"},{"key":"33120_CR17","doi-asserted-by":"publisher","unstructured":"Nishizawa, T. et al. Ante- and post-hoc explanations for prediction models of cisplatin-induced acute kidney injury: A comparative study. In Proceedings of the 2023 7th International Conference on Medical and Health Informatics, ICMHI \u201923, 66\u201371, https:\/\/doi.org\/10.1145\/3608298.3608311(Association for Computing Machinery, New York, NY, USA, 2023).","DOI":"10.1145\/3608298.3608311"},{"key":"33120_CR18","doi-asserted-by":"publisher","first-page":"1263","DOI":"10.1177\/00220345221106086","volume":"101","author":"J Ma","year":"2022","unstructured":"Ma, J. et al. Towards trustworthy ai in dentistry. J. Dent. Res. 101, 1263\u20131268. https:\/\/doi.org\/10.1177\/00220345221106086 (2022) (Epub 2022 Jun 23.).","journal-title":"J. Dent. Res."},{"key":"33120_CR19","doi-asserted-by":"publisher","DOI":"10.1016\/j.knosys.2022.109278","volume":"252","author":"M Li","year":"2022","unstructured":"Li, M. et al. Explainable multi-instance and multi-task learning for covid-19 diagnosis and lesion segmentation in ct images. Knowl.-Based Syst. 252, 109278. https:\/\/doi.org\/10.1016\/j.knosys.2022.109278 (2022).","journal-title":"Knowl.-Based Syst."},{"key":"33120_CR20","unstructured":"Lundberg, S.\u00a0M. & Lee, S.-I. A unified approach to interpreting model predictions. In Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS\u201917, 4768\u20134777 (Curran Associates Inc., Red Hook, NY, USA, 2017)."},{"key":"33120_CR21","doi-asserted-by":"publisher","unstructured":"Wang, W. & Lo, A. C.\u00a0Y. Diabetic retinopathy: Pathophysiology and treatments. International Journal of Molecular Sciences19, https:\/\/doi.org\/10.3390\/ijms19061816(2018).","DOI":"10.3390\/ijms19061816"},{"key":"33120_CR22","doi-asserted-by":"publisher","first-page":"98","DOI":"10.1016\/j.compbiomed.2017.09.008","volume":"90","author":"G Leontidis","year":"2017","unstructured":"Leontidis, G., Al-Diri, B. & Hunter, A. A new unified framework for the early detection of the progression to diabetic retinopathy from fundus images. Comput. Biol. Med. 90, 98\u2013115. https:\/\/doi.org\/10.1016\/j.compbiomed.2017.09.008 (2017).","journal-title":"Comput. Biol. Med."},{"key":"33120_CR23","doi-asserted-by":"publisher","first-page":"2136","DOI":"10.1016\/j.compbiomed.2013.10.007","volume":"43","author":"MRK Mookiah","year":"2013","unstructured":"Mookiah, M. R. K. et al. Computer-aided diagnosis of diabetic retinopathy: A review. Comput. Biol. Med. 43, 2136\u20132155. https:\/\/doi.org\/10.1016\/j.compbiomed.2013.10.007 (2013).","journal-title":"Comput. Biol. Med."},{"key":"33120_CR24","doi-asserted-by":"publisher","DOI":"10.1016\/j.jdiacomp.2019.107417","volume":"33","author":"RP Singh","year":"2019","unstructured":"Singh, R. P., Elman, M. J., Singh, S. K., Fung, A. E. & Stoilov, I. Advances in the treatment of diabetic retinopathy. J. Diabetes Complications 33, 107417. https:\/\/doi.org\/10.1016\/j.jdiacomp.2019.107417 (2019).","journal-title":"J. Diabetes Complications"},{"key":"33120_CR25","doi-asserted-by":"publisher","first-page":"106","DOI":"10.1016\/j.ins.2007.07.020","volume":"178","author":"WL Yun","year":"2008","unstructured":"Yun, W. L. et al. Identification of different stages of diabetic retinopathy using retinal optical images. Inf. Sci. 178, 106\u2013121. https:\/\/doi.org\/10.1016\/j.ins.2007.07.020 (2008).","journal-title":"Inf. Sci."},{"key":"33120_CR26","doi-asserted-by":"publisher","first-page":"1122","DOI":"10.1016\/j.cell.2018.02.010","volume":"172","author":"DS Kermany","year":"2018","unstructured":"Kermany, D. S. et al. Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell 172, 1122-1131.e9. https:\/\/doi.org\/10.1016\/j.cell.2018.02.010 (2018).","journal-title":"Cell"},{"key":"33120_CR27","doi-asserted-by":"publisher","unstructured":"Elsharkawy, M. et al. A novel computer-aided diagnostic system for early detection of diabetic retinopathy using 3d-oct higher-order spatial appearance model. Diagnostics12, https:\/\/doi.org\/10.3390\/diagnostics12020461(2022).","DOI":"10.3390\/diagnostics12020461"},{"key":"33120_CR28","doi-asserted-by":"publisher","first-page":"145","DOI":"10.4103\/0974-9233.151887","volume":"22","author":"DA Salz","year":"2015","unstructured":"Salz, D. A. & Witkin, A. J. Imaging in diabetic retinopathy. Middle East Afr. J. Ophthalmol. 22, 145\u2013150. https:\/\/doi.org\/10.4103\/0974-9233.151887 (2015).","journal-title":"Middle East Afr. J. Ophthalmol."},{"key":"33120_CR29","doi-asserted-by":"publisher","first-page":"459","DOI":"10.1097\/IAE.0000000000003070","volume":"41","author":"LM Jampol","year":"2021","unstructured":"Jampol, L. M., Tadayoni, R. & Ip, M. Need for a new classification of diabetic retinopathy. Retina 41, 459\u2013460. https:\/\/doi.org\/10.1097\/IAE.0000000000003070 (2021).","journal-title":"Retina"},{"key":"33120_CR30","doi-asserted-by":"publisher","first-page":"12636","DOI":"10.1364\/OE.15.012636","volume":"15","author":"J Fingler","year":"2007","unstructured":"Fingler, J., Schwartz, D., Yang, C. & Fraser, S. E. Mobility and transverse flow visualization using phase variance contrast with spectral domain optical coherence tomography. Opt. Express 15, 12636\u201312653. https:\/\/doi.org\/10.1364\/OE.15.012636 (2007).","journal-title":"Opt. Express"},{"key":"33120_CR31","doi-asserted-by":"publisher","first-page":"22190","DOI":"10.1364\/OE.17.022190","volume":"17","author":"J Fingler","year":"2009","unstructured":"Fingler, J., Zawadzki, R. J., Werner, J. S., Schwartz, D. & Fraser, S. E. Volumetric microvascular imaging of human retina using optical coherence tomography with a novel motion contrast technique. Opt. Express 17, 22190\u201322200. https:\/\/doi.org\/10.1364\/OE.17.022190 (2009).","journal-title":"Opt. Express"},{"key":"33120_CR32","doi-asserted-by":"publisher","first-page":"1467","DOI":"10.1364\/OL.35.001467","volume":"35","author":"RK Wang","year":"2010","unstructured":"Wang, R. K., An, L., Francis, P. & Wilson, D. J. Depth-resolved imaging of capillary networks in retina and choroid using ultrahigh sensitive optical microangiography. Opt. Lett. 35, 1467\u20131469. https:\/\/doi.org\/10.1364\/OL.35.001467 (2010).","journal-title":"Opt. Lett."},{"key":"33120_CR33","unstructured":"Matten, P. et al. A clinical mhz swept-source oct prototype for ultra-widefield imaging. Investigative Ophthalmology & Visual Science61, PB00107\u2013PB00107 (2020). This is a 2020 Imaging in the Eye Conference abstract."},{"key":"33120_CR34","doi-asserted-by":"publisher","first-page":"1009","DOI":"10.1109\/TMI.2022.3222638","volume":"42","author":"M Niederleithner","year":"2023","unstructured":"Niederleithner, M. et al. Ultra-widefield oct angiography. IEEE Trans. Med. Imaging 42, 1009\u20131020. https:\/\/doi.org\/10.1109\/TMI.2022.3222638 (2023).","journal-title":"IEEE Trans. Med. Imaging"},{"key":"33120_CR35","doi-asserted-by":"publisher","first-page":"91","DOI":"10.1136\/bjo-2022-322134","volume":"108","author":"H Stino","year":"2023","unstructured":"Stino, H. et al. Detection of diabetic neovascularisation using single-capture 65$$^{\\circ }$$-widefield optical coherence tomography angiography. Br. J. Ophthalmol. 108, 91\u201397. https:\/\/doi.org\/10.1136\/bjo-2022-322134 (2023).","journal-title":"Br. J. Ophthalmol."},{"key":"33120_CR36","doi-asserted-by":"publisher","first-page":"23024","DOI":"10.1038\/s41598-021-02479-6","volume":"11","author":"G Ryu","year":"2021","unstructured":"Ryu, G. et al. A deep learning model for identifying diabetic retinopathy using optical coherence tomography angiography. Sci. Rep. 11, 23024. https:\/\/doi.org\/10.1038\/s41598-021-02479-6 (2021) (Received 22 March 2021, Accepted 10 November 2021, Published 26 November 2021.).","journal-title":"Sci. Rep."},{"key":"33120_CR37","doi-asserted-by":"publisher","DOI":"10.1016\/j.xops.2022.100245","volume":"3","author":"P Zang","year":"2023","unstructured":"Zang, P. et al. Deep-learning-aided diagnosis of diabetic retinopathy, age-related macular degeneration, and glaucoma based on structural and angiographic oct. Ophthalmology Science 3, 100245. https:\/\/doi.org\/10.1016\/j.xops.2022.100245 (2023).","journal-title":"Ophthalmology Science"},{"key":"33120_CR38","doi-asserted-by":"publisher","first-page":"1042","DOI":"10.1016\/j.oret.2023.07.020","volume":"7","author":"H Stino","year":"2023","unstructured":"Stino, H. et al. Association of diabetic lesions and retinal nonperfusion using widefield multimodal imaging. Ophthalmology Retina 7, 1042\u20131050. https:\/\/doi.org\/10.1016\/j.oret.2023.07.020 (2023).","journal-title":"Ophthalmology Retina"},{"key":"33120_CR39","unstructured":"Kingma, D. P. & Ba, J (A method for stochastic optimization, Adam, 2017) arXiv:1412.6980."},{"key":"33120_CR40","doi-asserted-by":"crossref","unstructured":"Lin, T.-Y., Goyal, P., Girshick, R., He, K. & Doll\u00e1r, P. Focal loss for dense object detection (2018). arXiv:1708.02002.","DOI":"10.1109\/ICCV.2017.324"},{"key":"33120_CR41","unstructured":"Petsiuk, V., Das, A. & Saenko, K. Rise: Randomized input sampling for explanation of black-box models. arXiv:1806.07421 (2018)."},{"key":"33120_CR42","unstructured":"Baniecki, H. et al. Be careful when evaluating explanations regarding ground truth (2023). arXiv:2311.04813."}],"container-title":["Scientific Reports"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.nature.com\/articles\/s41598-025-33120-5","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s41598-025-33120-5.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.nature.com\/articles\/s41598-025-33120-5.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,1,27]],"date-time":"2026-01-27T01:04:05Z","timestamp":1769475845000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.nature.com\/articles\/s41598-025-33120-5"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,1,5]]},"references-count":42,"journal-issue":{"issue":"1","published-online":{"date-parts":[[2026,12]]}},"alternative-id":["33120"],"URL":"https:\/\/doi.org\/10.1038\/s41598-025-33120-5","relation":{},"ISSN":["2045-2322"],"issn-type":[{"value":"2045-2322","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,1,5]]},"assertion":[{"value":"18 August 2025","order":1,"name":"received","label":"Received","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"16 December 2025","order":2,"name":"accepted","label":"Accepted","group":{"name":"ArticleHistory","label":"Article History"}},{"value":"5 January 2026","order":3,"name":"first_online","label":"First Online","group":{"name":"ArticleHistory","label":"Article History"}},{"order":1,"name":"Ethics","group":{"name":"EthicsHeading","label":"Declarations"}},{"value":"TSchm is an employee of Carl Zeiss Meditec AG (Oberkochen, Germany). WD and RAL are consultants for Carl Zeiss Meditec AG and have received financial compensation. All other authors declare no conflicts of interest.","order":2,"name":"Ethics","group":{"name":"EthicsHeading","label":"Competing interests"}}],"article-number":"3296"}}