{"paper":{"title":"FunduSegmenter: Leveraging the RETFound Foundation Model for Joint Optic Disc and Optic Cup Segmentation in Retinal Fundus Images","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Adapting RETFound with new adapters and a decoder enables accurate joint optic disc and optic cup segmentation in fundus images.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Emanuele Trucco, Muthu Rama Krishnan Mookiah, Zhenyi Zhao","submitted_at":"2025-08-15T09:43:49Z","abstract_excerpt":"Purpose: This study introduces the first adaptation of RETFound for joint optic disc (OD) and optic cup (OC) segmentation. RETFound is a well-known foundation model developed for fundus camera and optical coherence tomography images, which has shown promising performance in disease diagnosis. Methods: We propose FunduSegmenter, a model integrating a series of novel modules with RETFound, including a Pre-adapter, a Decoder, a Post-adapter, skip connections with Convolutional Block Attention Module and a Vision Transformer block adapter. The model is evaluated on a proprietary dataset, GoDARTS, "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"This study introduces the first adaptation of RETFound for joint optic disc (OD) and optic cup (OC) segmentation. ... An average Dice similarity coefficient of 90.51% was achieved in internal verification, which outperformed all baselines, some substantially (nnU-Net: 82.91%; DUNet: 89.17%; TransUNet: 87.91%). In all external verification experiments, the average results were about 3% higher than those of the best baseline.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that the proposed modules (Pre-adapter, Decoder, Post-adapter, CBAM skip connections, and ViT block adapter) transfer usefully to other foundation models and that performance on the tested mix of proprietary and public datasets indicates reliable behavior on unseen clinical data sources.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"FunduSegmenter adapts RETFound with custom adapters and attention modules to reach 90.51% average Dice score for optic disc and cup segmentation, outperforming baselines on internal and external tests across five datasets.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Adapting RETFound with new adapters and a decoder enables accurate joint optic disc and optic cup segmentation in fundus images.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"94b18fce8205646671bcd692dda7c983016e030af31718523ba4c1b85b9de8ef"},"source":{"id":"2508.11354","kind":"arxiv","version":3},"verdict":{"id":"3846198f-9400-4fb7-97b0-44e050ed009d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-18T22:55:49.636355Z","strongest_claim":"This study introduces the first adaptation of RETFound for joint optic disc (OD) and optic cup (OC) segmentation. ... An average Dice similarity coefficient of 90.51% was achieved in internal verification, which outperformed all baselines, some substantially (nnU-Net: 82.91%; DUNet: 89.17%; TransUNet: 87.91%). In all external verification experiments, the average results were about 3% higher than those of the best baseline.","one_line_summary":"FunduSegmenter adapts RETFound with custom adapters and attention modules to reach 90.51% average Dice score for optic disc and cup segmentation, outperforming baselines on internal and external tests across five datasets.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that the proposed modules (Pre-adapter, Decoder, Post-adapter, CBAM skip connections, and ViT block adapter) transfer usefully to other foundation models and that performance on the tested mix of proprietary and public datasets indicates reliable behavior on unseen clinical data sources.","pith_extraction_headline":"Adapting RETFound with new adapters and a decoder enables accurate joint optic disc and optic cup segmentation in fundus images."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2508.11354/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":49,"sample":[{"doi":"","year":2023,"title":"Zhou, Y ., Chia, M. A., Wagner, S. K., Ayhan, M. S., Williamson, D. J., Struyven, R. R., ... & Keane, P . A. (2023). A foundation model for generalizable disease detection from retinal images. Nature,","work_id":"e73ca5f7-9c7a-4c80-8497-215abc4ecb76","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Mookiah, M. R. K., Hogg, S., MacGillivray, T., & Trucco, E. (2021). On the quantitative effects of compression of retinal fundus images on morphometric vascular measurements in VAMPIRE. Computer Metho","work_id":"605c7c6b-0b7d-44d9-8f11-76a8a4bce199","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"W., & Heng, P","work_id":"ec2ca960-11e2-4f6c-ac61-00bdcdec58e7","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Porwal, P ., Pachade, S., Kamble, R., Kokare, M., Deshmukh, G., Sahasrabuddhe, V ., & Meriaudeau, F . (2018). Indian diabetic retinopathy image dataset (IDRiD): a database for diabetic retinopathy scr","work_id":"b3c5877c-768d-4db2-b747-787ad649b11d","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Sivaswamy, J., Krishnadas, S., Chakravarty, A., Joshi, G., & Tabish, A. S. (2015). A comprehensive retinal image dataset for the assessment of glaucoma from the optic nerve head analysis. JSM Biomedic","work_id":"fd2adacc-66c6-4940-88ad-a52b94a0a467","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":49,"snapshot_sha256":"5f0fa7e10644fe7607a7975c7417adfea90cca3cce2f2fd9062977c322c31c9d","internal_anchors":5},"formal_canon":{"evidence_count":2,"snapshot_sha256":"0c2a7ec534532f7fc722244f8990f0cb0e7c53ad9dd7a5b2c6806c3c3c14c4f4"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}