Prematüre Bebeklerde Otomatik Retina Görüntüsü Doğrulama, ROP Taraması ve Plus Hastalığı Tespiti için Üç Aşamalı YOLOv11 Çerçevesi: Gerçek dünya çalışması

Abstract

Prematüre Retinopatisi (ROP), çocukluk çağı körlüğünün önde gelen nedenlerinden biri olup, Artı (Plus) hastalık bu durumun tedavi gerektiren en şiddetli formunu teşkil etmektedir. Erken ve doğru tanı kritik öneme sahip olmakla birlikte, gözlemciler arası değişkenlik ve yenidoğan retina görüntülerinin karmaşıklığı nedeniyle zorluklar içermektedir. Bu çalışma, gerçek dünya yenidoğan retina görüntülerini kullanarak ROP ve Artı hastalığın tespiti için sağlam ve otomatik bir derin öğrenme çatısı geliştirmeyi ve değerlendirmeyi amaçlamaktadır. Bu kapsamda, bağımsız olarak eğitilmiş YOLOv11 modellerini kullanan üç aşamalı bir tanısal yol önerilmiştir. Birinci aşama, girdi görüntülerinin fundus veya retina dışı olarak doğrulanmasını sağlamaktadır. İkinci aşama, fundus görüntülerinde ROP varlığını sınıflandırmaktadır. Üçüncü aşama ise ROP pozitif görüntülerde Artı hastalığı tespit etmektedir. Modeller, 188 prematüre bebekten alınan 6.004 fundus görüntüsünden oluşan halka açık bir veri seti üzerinde eğitilmiş ve test edilmiştir. Ek olarak, birinci aşama için COCO veritabanından 6.000 retina dışı görüntü kullanılmıştır. Ön işleme adımları, yeniden boyutlandırma, normalizasyon ve damar segmentasyonunu içermiştir. Performans, doğruluk (accuracy), kesinlik (precision), duyarlılık (recall), F1-skoru ve Eğri Altında Kalan Alan (AUC) metrikleri kullanılarak değerlendirilmiştir. Birinci aşama modeli, retina görüntüsü doğrulamasında %99,95 doğruluk (AUC 1.00) elde etmiştir. İkinci aşama modeli, ROP'u %99,83 doğruluk (AUC 1.00) ve %100 duyarlılık ile tespit etmiştir. Kritik olarak, üçüncü aşama modeli test setinde Artı hastalığın saptanmasında mükemmel bir performans (%100 doğruluk, kesinlik, duyarlılık, F1-skoru ve AUC 1.00) göstermiştir. Eğitim dinamikleri, aşırı öğrenme (overfitting) olmaksızın kararlı bir yakınsama göstermiştir. Önerilen üç aşamalı YOLOv11 çatısı, otomatik ROP ve Artı hastalığı tespiti için olağanüstü yüksek doğruluk ve güvenilirlik sergilemektedir. Bu sistem, özellikle kaynakları kısıtlı ortamlarda, prematüre bebeklerde körlüğün önlenmesi ve zamanında müdahale edilmesine yardımcı olan bir klinik karar destek aracı olarak önemli bir potansiyele sahiptir.

Keywords

• Prematüre Retinopatisi
• Plus Hastalığı
• Derin Öğrenme
• Otomatik Tanı
• Tıbbi Görüntü Sınıflandırması
• Prematüre Bebekler

A Three-Stage YOLOv11 Framework for Automated Retinal Image Validation, ROP Screening, and Plus Disease Detection in Premature Infants: A real-world study

Abstract

Retinopathy of Prematurity (ROP) is a major cause of blindness in children, and Plus disease is the most serious form of the disease that needs treatment. It is very important to make an early and correct diagnosis, but this is hard because different observers see things differently and neonatal retinal images are hard to understand. The goal of this study is to create and test a strong, automated deep learning system for finding ROP and Plus disease in real-world neonatal retinal images. A three-stage diagnostic pathway employing independently trained YOLOv11 models was proposed. In Stage 1, the input images are checked to see if they are fundus or not. Stage 2 sorts fundus images to see if ROP is there. Stage 3 finds Plus disease in images that show ROP. The models were trained and tested on a publicly accessible dataset comprising 6,004 fundus images from 188 premature infants, supplemented by 6,000 non-retinal images from the COCO database for Stage 1. Resizing, normalizing, and segmenting vessels were all part of the preprocessing. We used accuracy, precision, recall, F1-score, and AUC to measure performance. The Stage 1 model was 99.95% accurate (AUC 1.00) when it came to validating retinal images. The Stage 2 model found ROP with 99.83% accuracy (AUC 1.00) and 100% sensitivity. The Stage 3 model was very important because it was able to perfectly classify Plus disease on the internal test set (100% accuracy, precision, recall, F1-score, and AUC 1.00). The training dynamics showed that the model was converging steadily without overfitting. The suggested three-stage YOLOv11 framework shows very high accuracy and dependability for finding ROP and Plus disease automatically. This system has a lot of potential as a clinical decision support tool, especially in places with few resources. It can help doctors act quickly and stop premature babies from going blind.

Keywords

• Retinopathy of Prematurity
• Plus Disease
• Deep Learning
• Automated Diagnosis
• Medical Image Classification
• Premature Infants

References

1. Yildiz VM, Tian P, Yildiz I, Brown JM, Kalpathy-Cramer J, Dy J, et al. Plus disease in retinopathy of prematurity: Convolutional neural network performance using a combined neural network and feature extraction approach. Translational Vision Science & Technology 2020;9(2):10. doi:10.1167/tvst.9.2.10.
2. Jemshi KM, Sreelekha G, Sathidevi PS, Mohanachandran PA. Plus disease classification in retinopathy of prematurity using transform based features. Multimedia Tools and Applications 2024;83:861-891. doi:10.1007/s11042-023-15430-w.
3. Vinekar A. IT-enabled innovation to prevent infant blindness in rural India: The KIDROP experience. Journal of Indian Business Research 2011;3(2):98-102. doi:10.1108/17554191111132215.
4. Vinekar A, Mangalesh S, Jayadev C, Gilbert C, Dogra M, Shetty B. Impact of expansion of telemedicine screening for retinopathy of prematurity in India. Indian Journal of Ophthalmology 2017;65(5):390-395. doi:10.4103/ijo.IJO_211_17.
5. Chiang MF, Quinn GE, Fielder AR, et al. International Classification of Retinopathy of Prematurity, Third Edition. Ophthalmology 2021;128(10):e51-e68. doi:10.1016/j.ophtha.2021.05.008.
6. Sanghi G, Gangwe A, Das P. Evidence based management of retinopathy of prematurity: More than meets the eye. Clinical Epidemiology and Global Health 2024;26:101530. doi:10.1016/j.cegh.2024.101530.
7. Deepthi K, Josephine MS, Jayabala Raja V. Automated diagnosis of plus disease in retinopathy of prematurity based on transformer-based unsupervised curriculum learning. Biomedical Signal Processing and Control 2025;104:107521. doi:10.1016/j.bspc.2025.107521.
8. Vidivelli S, Padmakumari P, Parthiban C, DharunBalaji A, Manikandan R, Gandomi AH. Optimising deep learning models for ophthalmological disorder classification. Scientific Reports 2025;15:3115. doi:10.1038/s41598-024-75867-3.

9. Vyas R, Kanumuri T, Sheoran G, Dubey P. Efficient iris recognition through curvelet transform and polynomial fitting. Optik 2019;185:859-867. doi:10.1016/j.ijleo.2019.04.015.
10. Gschließer A, Stifter E, Neumayer T, Moser E, Papp A, Pircher N, et al. Inter-expert and intra-expert agreement on the diagnosis and treatment of retinopathy of prematurity. American Journal of Ophthalmology 2015;160(3):553-560. doi:10.1016/j.ajo.2015.05.016.
11. Solarte CE, Awad AH, Wilson CM, Ells A. Plus disease: Why is it important in retinopathy of prematurity? Middle East African Journal of Ophthalmology 2010;17(2):148-155. doi:10.4103/0974-9233.63080.
12. Rajashekar D, Srinivasa G, Vinekar A. Comprehensive retinal image analysis for aggressive posterior retinopathy of prematurity. PLoS ONE 2016;11(10):e0163923. doi:10.1371/journal.pone.0163923.
13. Oloumi F, Rangayyan RM, Casti P, Ells AL. Computer-aided diagnosis of plus disease via measurement of vessel thickness in retinal fundus images of preterm infants. Computers in Biology and Medicine 2015;66:316-329. doi:10.1016/j.compbiomed.2015.09.009.
14. Rajan RP, Kohli P, Babu N, Dakshayini C, Tandon M, Ramasamy K. Treatment of retinopathy of prematurity (ROP) outside International Classification of ROP (ICROP) guidelines. Graefe's Archive for Clinical and Experimental Ophthalmology 2020;258(6):1205-1210. doi:10.1007/s00417-020-04706-8.
15. Lei B, Zeng X, Huang S, Zhang R, Chen G, Zhao J, et al. Automated detection of retinopathy of prematurity by deep attention network. Multimedia Tools and Applications 2021;80:36341-36360. doi:10.1007/s11042-021-11208-0.
16. Peng Y, Zhu W, Chen Z, Wang M, Geng L, Yu K, et al. Automatic staging for retinopathy of prematurity with deep feature fusion and ordinal classification strategy. IEEE Transactions on Medical Imaging 2021;40(7):1750-1762. doi:10.1109/TMI.2021.3065753.
17. Ding A, Chen Q, Cao Y, Liu B. Retinopathy of prematurity stage diagnosis using object segmentation and convolutional neural networks. In: 2020 International Joint Conference on Neural Networks (IJCNN). IEEE; 2020, p. 1-6. doi:10.1109/IJCNN48605.2020.9207288.
18. Attallah O. DIAROP: Automated deep learning-based diagnostic tool for retinopathy of prematurity. Diagnostics 2021;11(11):2034. doi:10.3390/diagnostics11112034.
19. Rahim S. Using novel fundus image pre-processing to improve the classification of retinopathy of prematurity (ROP) using deep learning [PhD thesis]. Hamilton, Canada: McMaster University; 2024. Available from: https://macsphere.mcmaster.ca/handle/11375/30120.
20. Huang K, Dong W, Li J, Chen Y, Zhong J, Yi Z. GFF-net: Graph-based feature fusion network for diagnosing plus disease in retinopathy of prematurity. Applied Intelligence 2023;53(20):25259-25281. doi:10.1007/s10489-023-04766-3.
21. Li Y, Li J, Zhang X, Peng J, Li J, Zhao P. Identification of gene mutations in atypical retinopathy of prematurity cases. Journal of Ophthalmology 2020;2020:4212158. doi:10.1155/2020/4212158.
22. Govind AS, Gadad A, Garodia DJ, Gupta A, Vinekar A, Srinivasa G. Deep learning for classifying stages of retinopathy of prematurity. In: 2024 International Conference on Advances in Modern Age Technologies for Health and Engineering Science (AMATHE). IEEE; 2024, p. 1-6. doi:10.1109/AMATHE61652.2024.10582102.
23. Sankari VM, Raja U, Snekhalatha AC, Baskaran P. Automated diagnosis of retinopathy of prematurity from retinal images of preterm infants using hybrid deep learning techniques. Biomedical Signal Processing and Control 2023;85:104883. doi:10.1016/j.bspc.2023.104883.
24. Kumar V, Patel H, Paul K, Azad S. Deep learning-assisted retinopathy of prematurity (ROP) screening. ACM Transactions on Computing for Healthcare 2023;4(3):1-18. doi:10.1145/3596223.
25. Chowdary KL, Manne S, Lakshmi B. Detecting retinopathy of prematurity disease based on fundus image dataset. In: International Conference on Power Engineering and Intelligent Systems (PEIS). Springer; 2023, p. 363-377. doi:10.1007/978-981-99-7383-5_27.
26. Agrawal R, Kulkarni S, Walambe R, Kotecha K. Assistive framework for automatic detection of all the zones in retinopathy of prematurity using deep learning. Journal of Digital Imaging 2021;34(4):932-947. doi:10.1007/s10278-021-00477-8.
27. Skouta A, Elmoufdi A, Jai-Andaloussi S, Ouchetto O. Deep learning for diabetic retinopathy assessments: A literature review. Multimedia Tools and Applications 2023;82:41701-41766. doi:10.1007/s11042-023-15110-9.
28. Timkovič J, Nowaková J, Kubíček J, Hasal M, Varyšová A, Kolarčík L, et al. Retinal image dataset of infants and retinopathy of prematurity. Scientific Data 2024;11:164. doi:10.1038/s41597-024-03409-7.
29. Mao J, Luo Y, Liu L, Lao J, Shao Y, Zhang M, et al. Automated diagnosis and quantitative analysis of plus disease in retinopathy of prematurity based on deep convolutional neural networks. Acta Ophthalmologica 2020;98(3):e339-e345. doi:10.1111/aos.14264.
30. Brown JM, Campbell JP, Beers A, Chang K, Ostmo S, Chan RP, et al. Automated diagnosis of plus disease in retinopathy of prematurity using deep convolutional neural networks. JAMA Ophthalmology 2018;136(7):803-810. doi:10.1001/jamaophthalmol.2018.1934.
31. Bai A, Dai S, Hung J, Kirpalani A, Russell H, Elder J, et al. Multicenter validation of deep learning algorithm ROP.AI for the automated diagnosis of plus disease in ROP. Translational Vision Science & Technology 2023;12(8):13. doi:10.1167/tvst.12.8.13.
32. Jocher G, Qiu J. Ultralytics YOLOv11 (Version 11.0.0) [Computer software]. Ultralytics; 2024. Available from: https://github.com/ultralytics/ultralytics.