{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,8]],"date-time":"2026-10-08T21:49:26Z","timestamp":1791496166226,"version":"4.3.4"},"reference-count":0,"publisher":"SPIE-Intl Soc Optical Eng","issue":"10","license":[{"start":{"date-parts":[[2026,10,8]],"date-time":"2026-10-08T00:00:00Z","timestamp":1791417600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"name":"Guangxi Science and Technology Innovation Platform Program","award":["Guike LT2600640001"],"award-info":[{"award-number":["Guike LT2600640001"]}]},{"name":"National Key R&D Program of China","award":["2018YFA0701700"],"award-info":[{"award-number":["2018YFA0701700"]}]},{"name":"Natural Science Foundation of Guangxi","award":["2023GXNSFAA026027"],"award-info":[{"award-number":["2023GXNSFAA026027"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["J. Biomed. Opt."],"published-print":{"date-parts":[[2026,10,8]]},"abstract":"<jats:p>SignificanceDiabetic retinopathy (DR) lesion segmentation provides pixel-level lesion localization in fundus images. However, such lesions are often tiny, sparsely distributed, low-contrast, morphologically diverse, and visually similar to vessels or surrounding retinal structures, making accurate localization difficult. Many existing methods emphasize multi-scale spatial representation learning and handle boundary ambiguity and subtle texture differences mainly through encoder\u2013decoder fusion. We therefore examine boundary and frequency representations as explicit, complementary inputs to semantic decoding.AimWe aim to develop SEFCA-Net, a semantic\u2013edge\u2013frequency cross-attention network for diabetic retinopathy lesion segmentation. The main objective is to examine whether semantic, edge, and frequency-domain representations can be fused under semantic guidance to improve pixel-level lesion localization without compromising the semantic decoding stream. The evaluation focuses on four lesion categories, including hard exudates, hemorrhages, soft exudates, and microaneurysms, using public benchmarks and a DDR-to-IDRiD cross-dataset setting.ApproachSEFCA-Net uses HRNet-W48 as the semantic backbone to extract multi-scale semantic features and explicitly constructs edge and frequency-domain representations from input fundus images. The learnable multi-scale edge branch extracts boundary-sensitive features, while the all-channel multi-scale discrete cosine transform (DCT) frequency-domain branch generates texture-sensitive responses to low-contrast lesions. The semantic\u2013edge\u2013frequency cross-attention decoder uses the semantic representation as the primary stream and injects edge and frequency-domain features through semantic-guided window-based cross-attention. Performance is evaluated using the area under the precision\u2013recall curve (AUPR), Dice, and Intersection-over-Union (IoU).ResultsOn DDR, SEFCA-Net achieves 47.85% mean AUPR (mAUPR), 46.76% mean Dice, and 31.22% mean IoU. On IDRiD, it reaches 69.74% mAUPR; when trained on DDR and tested on IDRiD, it obtains 59.83% mAUPR. Ablation studies show that either auxiliary branch improves the semantic-only baseline and that the complete configuration gives the highest mAUPR. Multi-scale Sobel modeling and DCT outperform the tested single-scale and wavelet alternatives, respectively.ConclusionsSEFCA-Net shows that semantic features can remain the main decoding stream while boundary-sensitive features and texture-sensitive responses are integrated in a controlled manner. Under the evaluated protocols, the model shows competitive in-domain performance and records the highest DDR-to-IDRiD mAUPR among the reproduced methods.<\/jats:p>","DOI":"10.1117\/1.jbo.31.10.106002","type":"journal-article","created":{"date-parts":[[2026,10,8]],"date-time":"2026-10-08T21:00:00Z","timestamp":1791493200000},"page":"1-17","source":"Crossref","is-referenced-by-count":0,"title":["SEFCA-Net: a semantic\u2013edge\u2013frequency cross-attention network for diabetic retinopathy lesion segmentation"],"prefix":"10.1117","volume":"31","author":[{"given":"Jin","family":"Xie","sequence":"first","affiliation":[{"name":"Nanning Normal University, Guangxi Key Lab of Human-Machine Interaction and Intelligent Decision, Nanning, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Duo","family":"Xu","sequence":"additional","affiliation":[{"name":"Soochow University, School of Electronics and Information Engineering, Suzhou, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yuzhong","family":"Peng","sequence":"additional","affiliation":[{"name":"Zhejiang Wanli University, College of Big Data and Software Engineering, Ningbo, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Kunlong","family":"Wu","sequence":"additional","affiliation":[{"name":"Nanning Normal University, Guangxi Key Lab of Human-Machine Interaction and Intelligent Decision, Nanning, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Fangning","family":"Gu","sequence":"additional","affiliation":[{"name":"Nanning Normal University, Guangxi Key Lab of Human-Machine Interaction and Intelligent Decision, Nanning, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xue","family":"Zhang","sequence":"additional","affiliation":[{"name":"Nanning Normal University, Guangxi Key Lab of Human-Machine Interaction and Intelligent Decision, Nanning, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhizhong","family":"Zhang","sequence":"additional","affiliation":[{"name":"Nanning Normal University, Guangxi Key Lab of Human-Machine Interaction and Intelligent Decision, Nanning, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xinjian","family":"Chen","sequence":"additional","affiliation":[{"name":"University of Electronic Science and Technology of China, School of Medicine, Chengdu, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0002-3573-3143","authenticated-orcid":false,"given":"Jianping","family":"Liao","sequence":"additional","affiliation":[{"name":"Nanning Normal University, Guangxi Key Lab of Human-Machine Interaction and Intelligent Decision, Nanning, China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"189","container-title":["Journal of Biomedical Optics"],"original-title":[],"link":[{"URL":"https:\/\/www.spiedigitallibrary.org\/journals\/journal-of-biomedical-optics\/volume-31\/issue-10\/106002\/SEFCA-Net--a-semanticedgefrequency-cross-attention-network-for-diabetic\/10.1117\/1.JBO.31.10.106002.full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,8]],"date-time":"2026-10-08T21:00:01Z","timestamp":1791493201000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.spiedigitallibrary.org\/journals\/journal-of-biomedical-optics\/volume-31\/issue-10\/106002\/SEFCA-Net--a-semanticedgefrequency-cross-attention-network-for-diabetic\/10.1117\/1.JBO.31.10.106002.full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,10,8]]},"references-count":0,"journal-issue":{"issue":"10","published-online":{"date-parts":[[2026,10,1]]}},"URL":"https:\/\/doi.org\/10.1117\/1.jbo.31.10.106002","relation":{},"ISSN":["1083-3668","1560-2281"],"issn-type":[{"value":"1083-3668","type":"print"},{"value":"1560-2281","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,10,8]]}}}