{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,9,11]],"date-time":"2026-09-11T14:09:22Z","timestamp":1789135762994,"version":"build-2803163510"},"reference-count":31,"publisher":"MDPI AG","issue":"18","license":[{"start":{"date-parts":[[2026,9,11]],"date-time":"2026-09-11T00:00:00Z","timestamp":1789084800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Diagnostics"],"abstract":"<jats:p>Background\/Objectives: Class imbalance and probabilistic prediction quality are important considerations in lightweight retinal classification. We evaluated whether knowledge distillation (KD) from a ConvNeXtV2-Base teacher improved an EfficientNet-B0 student in nine-class MultiEYE fundus classification. Methods: Exact-content screening quarantined 131 of 58,036 records because of exact-content duplication or label conflicts. Six prespecified seeds were evaluated in no-KD, label-smoothing, and KD arms (18 runs) under matched training conditions. Within the controlled experiment, TEST data were not used for training, hyperparameter tuning, checkpoint or model selection, or protocol modification. The primary endpoint was the KD\u2212no-KD macro-F1 difference on decontaminated TEST (n = 11,573), assessed by paired image-level bootstrap (10,000 replicates). Grad-CAM and Grad-CAM++ were examined qualitatively using one deterministically selected case per class and six-seed consensus maps. Results: Mean macro-F1 was 0.593, 0.597, and 0.633 for no-KD, label smoothing, and KD. KD exceeded no-KD by 0.039 (95% CI, 0.025\u20130.053; p &lt; 0.001), with positive paired differences in all six seeds, and label smoothing by 0.036 (Holm-adjusted p &lt; 0.001); label smoothing did not differ from no-KD (Holm-adjusted p = 0.554). Class-wise F1 differences were non-negative across all nine classes, with six remaining significant after correction for multiple comparisons. The Brier score difference was \u22120.077 (95% CI, \u22120.081 to \u22120.072; Holm-adjusted p &lt; 0.001), while ECE decreased descriptively from 0.171 to 0.084. Nine-class Grad-CAM and direct Grad-CAM++ consensus maps enabled qualitative no-KD versus KD comparison without establishing lesion-localization superiority. Conclusions: KD improved macro-F1 and lowered the Brier score while retaining the same EfficientNet-B0 student at inference. The evaluated label-smoothing configuration did not reproduce the macro-F1 gain; clinical superiority and external generalizability remain unestablished.<\/jats:p>","DOI":"10.3390\/diagnostics16182945","type":"journal-article","created":{"date-parts":[[2026,9,11]],"date-time":"2026-09-11T13:29:29Z","timestamp":1789133369000},"page":"2945","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":0,"title":["Knowledge Distillation and Explainability Analysis for Lightweight Retinal Disease Classification Using MultiEYE Fundus Images"],"prefix":"10.3390","volume":"16","author":[{"ORCID":"https:\/\/orcid.org\/0000-0003-3210-5061","authenticated-orcid":false,"given":"G\u00fcl\u015fen","family":"T\u00fcrker","sequence":"first","affiliation":[{"name":"Department of Computer Engineering, Faculty of Engineering and Natural Sciences, Suleyman Demirel University, 32260 Isparta, Turkey"}],"role":[{"vocabulary":"crossref","role":"author"},{"role":"conceptualization","vocabulary":"credit"},{"role":"methodology","vocabulary":"credit"},{"role":"software","vocabulary":"credit"},{"role":"validation","vocabulary":"credit"},{"role":"formal-analysis","vocabulary":"credit"},{"role":"data-curation","vocabulary":"credit"},{"role":"investigation","vocabulary":"credit"},{"role":"visualization","vocabulary":"credit"},{"role":"writing-original-draft","vocabulary":"credit"},{"role":"writing-review-editing","vocabulary":"credit"},{"role":"project-administration","vocabulary":"credit"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6701-363X","authenticated-orcid":false,"given":"Arif","family":"Koyun","sequence":"additional","affiliation":[{"name":"Department of Computer Engineering, Faculty of Engineering and Natural Sciences, Suleyman Demirel University, 32260 Isparta, Turkey"}],"role":[{"vocabulary":"crossref","role":"author"},{"role":"conceptualization","vocabulary":"credit"},{"role":"methodology","vocabulary":"credit"},{"role":"software","vocabulary":"credit"},{"role":"validation","vocabulary":"credit"},{"role":"formal-analysis","vocabulary":"credit"},{"role":"investigation","vocabulary":"credit"},{"role":"data-curation","vocabulary":"credit"},{"role":"writing-original-draft","vocabulary":"credit"},{"role":"writing-review-editing","vocabulary":"credit"},{"role":"supervision","vocabulary":"credit"},{"role":"project-administration","vocabulary":"credit"},{"role":"visualization","vocabulary":"credit"}]}],"member":"1968","published-online":{"date-parts":[[2026,9,11]]},"reference":[{"key":"#cr-split#-ref_1.1","unstructured":"(2021). GBD 2019 Blindness and Vision Impairment Collaborators"},{"key":"#cr-split#-ref_1.2","unstructured":"Vision Loss Expert Group of the Global Burden of Disease Study. Causes of blindness and vision impairment in 2020 and trends over 30 years, and prevalence of avoidable blindness in relation to VISION 2020: The Right to Sight: An analysis for the Global Burden of Disease Study. Lancet Glob. Health, 9, e144-e160."},{"key":"ref_2","doi-asserted-by":"crossref","unstructured":"Biswas, S., Khan, M.I.A., Hossain, M.T., Biswas, A., Nakai, T., and Rohdin, J. (2022). Which color channel is better for diagnosing retinal diseases automatically in color fundus photographs?. Life, 12.","DOI":"10.3390\/life12070973"},{"key":"ref_3","doi-asserted-by":"crossref","first-page":"2211","DOI":"10.1001\/jama.2017.18152","article-title":"Development and validation of a deep learning system for diabetic retinopathy and related eye diseases using retinal images from multiethnic populations with diabetes","volume":"318","author":"Ting","year":"2017","journal-title":"JAMA"},{"key":"ref_4","doi-asserted-by":"crossref","first-page":"39","DOI":"10.1038\/s41746-018-0040-6","article-title":"Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices","volume":"1","author":"Lavin","year":"2018","journal-title":"npj Digit. Med."},{"key":"ref_5","doi-asserted-by":"crossref","first-page":"2739","DOI":"10.1109\/JBHI.2022.3214086","article-title":"Multi-label retinal disease classification using transformers","volume":"27","author":"AlMarzouqi","year":"2023","journal-title":"IEEE J. Biomed. Health Inform."},{"key":"ref_6","doi-asserted-by":"crossref","first-page":"22","DOI":"10.1167\/tvst.12.1.22","article-title":"Automatic multilabel classification of multiple fundus diseases based on convolutional neural network with squeeze-and-excitation attention","volume":"12","author":"Lu","year":"2023","journal-title":"Transl. Vis. Sci. Technol."},{"key":"ref_7","doi-asserted-by":"crossref","first-page":"18898","DOI":"10.1038\/s41598-026-47858-z","article-title":"Long-tailed multi-label retinal disease classification using alternate group training and gradient-based re-weighting","volume":"16","author":"Jian","year":"2026","journal-title":"Sci. Rep."},{"key":"ref_8","doi-asserted-by":"crossref","first-page":"42361","DOI":"10.1109\/ACCESS.2023.3272228","article-title":"A lightweight robust deep learning model gained high accuracy in classifying a wide range of diabetic retinopathy images","volume":"11","author":"Raiaan","year":"2023","journal-title":"IEEE Access"},{"key":"ref_9","doi-asserted-by":"crossref","unstructured":"Arnob, A.K.B., Chayon, M.H.R., Al Farid, F., Husen, M.N., and Ahmed, F. (2025). A lightweight CNN for multiclass retinal disease screening with explainable AI. J. Imaging, 11.","DOI":"10.3390\/jimaging11080275"},{"key":"ref_10","doi-asserted-by":"crossref","unstructured":"Rafi, U., Nawaz, Q., Latif, M.A., and Khatoon, A. (2026). LiteFeatNet: A parameter-efficient and performance-centric deep learning model for multi-ocular disease identification using intermediate feature reduction from fundus images. PLoS ONE, 21.","DOI":"10.1371\/journal.pone.0347867"},{"key":"ref_11","doi-asserted-by":"crossref","unstructured":"Wang, H., Ke, T., and Lv, H. (2026). A CNN-ViT hybrid architecture Res101-MViT-Ens for accurate and lightweight automated ocular disease diagnosis. Appl. Sci., 16.","DOI":"10.3390\/app16062905"},{"key":"ref_12","unstructured":"Hinton, G., Vinyals, O., and Dean, J. (2015). Distilling the knowledge in a neural network. arXiv."},{"key":"ref_13","doi-asserted-by":"crossref","unstructured":"Meng, H., Lin, Z., Yang, F., Zhang, J., He, W., Xu, Y., and Cui, L. (2021). Knowledge distillation in medical data mining: A survey. Proceedings of the 5th International Conference on Crowd Science and Engineering (ICCSE \u201921), Jinan, China, 16\u201318 October 2021, Association for Computing Machinery.","DOI":"10.1145\/3503181.3503211"},{"key":"ref_14","doi-asserted-by":"crossref","unstructured":"Woo, S., Debnath, S., Hu, R., Chen, X., Liu, Z., Kweon, I.S., and Xie, S. (2023, January 18\u201322). ConvNeXt V2: Co-designing and scaling ConvNets with masked autoencoders. Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Vancouver, BC, Canada.","DOI":"10.1109\/CVPR52729.2023.01548"},{"key":"ref_15","first-page":"6105","article-title":"EfficientNet: Rethinking model scaling for convolutional neural networks","volume":"Volume 97","author":"Tan","year":"2019","journal-title":"Proceedings of the 36th International Conference on Machine Learning (ICML), Long Beach, CA, USA, 9\u201315 June 2019"},{"key":"ref_16","doi-asserted-by":"crossref","first-page":"1711","DOI":"10.1109\/TMI.2024.3518067","article-title":"MultiEYE: Dataset and benchmark for OCT-enhanced retinal disease recognition from fundus images","volume":"44","author":"Wang","year":"2025","journal-title":"IEEE Trans. Med. Imaging"},{"key":"ref_17","first-page":"639","article-title":"Fundus-enhanced disease-aware distillation model for retinal disease classification from OCT images","volume":"Volume 14226","author":"Greenspan","year":"2023","journal-title":"Medical Image Computing and Computer Assisted Intervention\u2014MICCAI 2023, Proceedings, Part VII"},{"key":"ref_18","first-page":"279","article-title":"QD-RetNet: Efficient retinal disease classification via quantized knowledge distillation","volume":"Volume 15918","author":"Ali","year":"2026","journal-title":"Medical Image Understanding and Analysis\u2014MIUA 2025, Proceedings, Part III"},{"key":"ref_19","doi-asserted-by":"crossref","unstructured":"Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., and Batra, D. (2017, January 22\u201329). Grad-CAM: Visual explanations from deep networks via gradient-based localization. Proceedings of the IEEE International Conference on Computer Vision (ICCV), Venice, Italy.","DOI":"10.1109\/ICCV.2017.74"},{"key":"ref_20","doi-asserted-by":"crossref","unstructured":"Chattopadhyay, A., Sarkar, A., Howlader, P., and Balasubramanian, V.N. (2018, January 12\u201315). Grad-CAM++: Generalized gradient-based visual explanations for deep convolutional networks. Proceedings of the 2018 IEEE Winter Conference on Applications of Computer Vision (WACV), Lake Tahoe, NV, USA.","DOI":"10.1109\/WACV.2018.00097"},{"key":"ref_21","doi-asserted-by":"crossref","unstructured":"Yuan, L., Tay, F.E.H., Li, G., Wang, T., and Feng, J. (2020, January 13\u201319). Revisiting knowledge distillation via label smoothing regularization. Proceedings of the IEEE\/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA.","DOI":"10.1109\/CVPR42600.2020.00396"},{"key":"ref_22","unstructured":"Loshchilov, I., and Hutter, F. (2019, January 6\u20139). Decoupled weight decay regularization. Proceedings of the 7th International Conference on Learning Representations (ICLR), New Orleans, LA, USA."},{"key":"ref_23","unstructured":"Loshchilov, I., and Hutter, F. (2017, January 24\u201326). SGDR: Stochastic gradient descent with warm restarts. Proceedings of the 5th International Conference on Learning Representations (ICLR), Toulon, France."},{"key":"ref_24","unstructured":"Micikevicius, P., Narang, S., Alben, J., Diamos, G., Elsen, E., Garc\u00eda, D., Ginsburg, B., Houston, M., Kuchaiev, O., and Venkatesh, G. (May, January 30). Mixed precision training. Proceedings of the 6th International Conference on Learning Representations (ICLR), Vancouver, BC, Canada."},{"key":"ref_25","unstructured":"Menon, A.K., Jayasumana, S., Rawat, A.S., Jain, H., Veit, A., and Kumar, S. (2021, January 3\u20137). Long-tail learning via logit adjustment. Proceedings of the 9th International Conference on Learning Representations (ICLR), Virtual Conference."},{"key":"ref_26","doi-asserted-by":"crossref","unstructured":"Davison, A.C., and Hinkley, D.V. (1997). Bootstrap Methods and Their Application, Cambridge University Press.","DOI":"10.1017\/CBO9780511802843"},{"key":"ref_27","doi-asserted-by":"crossref","first-page":"439","DOI":"10.1086\/341527","article-title":"A note on the calculation of empirical P values from Monte Carlo procedures","volume":"71","author":"North","year":"2002","journal-title":"Am. J. Hum. Genet."},{"key":"ref_28","first-page":"65","article-title":"A simple sequentially rejective multiple test procedure","volume":"6","author":"Holm","year":"1979","journal-title":"Scand. J. Stat."},{"key":"ref_29","doi-asserted-by":"crossref","first-page":"289","DOI":"10.1111\/j.2517-6161.1995.tb02031.x","article-title":"Controlling the false discovery rate: A practical and powerful approach to multiple testing","volume":"57","author":"Benjamini","year":"1995","journal-title":"J. R. Stat. Soc. Ser. B Methodol."},{"key":"ref_30","first-page":"1321","article-title":"On calibration of modern neural networks","volume":"Volume 70","author":"Precup","year":"2017","journal-title":"Proceedings of the 34th International Conference on Machine Learning (ICML 2017), Sydney, Australia, 6\u201311 August 2017"}],"container-title":["Diagnostics"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/2075-4418\/16\/18\/2945\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,11]],"date-time":"2026-09-11T13:31:59Z","timestamp":1789133519000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/2075-4418\/16\/18\/2945"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,9,11]]},"references-count":31,"journal-issue":{"issue":"18","published-online":{"date-parts":[[2026,9]]}},"alternative-id":["diagnostics16182945"],"URL":"https:\/\/doi.org\/10.3390\/diagnostics16182945","relation":{},"ISSN":["2075-4418"],"issn-type":[{"value":"2075-4418","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,9,11]]}}}