A study on the identification of lesions in fundus images based on lightweight convolutional neural networks_Chinese Journal of Ocular Fundus Diseases

Authors：

Cheng Hongzhuang ^1,2 , Li Chunxiu ³ , Shen Yadan ² , Kang Shuangwen ³ , Xu Bingjie ² , Ning Xinru ⁴ , Lang Zhuo ¹ ,  Zhong Jie ¹

1. Department of Ophthalmology, Sichuan Academy of Medical Sciences & Sichuan Provincial People's Hospital (Affiliated Hospital of University of Electronic Science and Technology of China), Chengdu 610072, China;
2. Chengdu University of Traditional Chinese Medicine, Chengdu 610075, China;
3. Medical School, University of Electronic Science and Technology of China, Chengdu 610054, China;
4. School of Clinical Medicine, Southwest Medical University, Luzhou 646000, China;

Corresponding?author：

Zhong Jie, Email: zhongjie@med.uestc.edu.cn

Keywords：

Artificial intelligence; Deep learning; Lesion identification; Fundus photography; Medical image analysis; Computer-aided diagnosis

DOI：

10.3760/cma.j.cn511434-20250609-00252

Video：

Export PDF Favorites Scan Get Citation

Abstract Full text Figures/Tables Video References Cited by

Objectives To develop a fundus photography (FP) image lesion recognition model based on the EfficientNet lightweight convolutional neural network architecture, and to preliminary evaluate its recognition performance. Methods A diagnostic test. The data was collected in the Department of Ophthalmology at Sichuan Provincial People's Hospital from June 2023 to June 2025. A lightweight 16-category lesion recognition model was constructed based on deep learning and 610 072 FP images. The FP images were sourced from Sichuan Provincial People's Hospital as well as the APTOS, Diabetic Retinopathy_2015, Diabetic Retinopathy_2019, and Retinal Disease datasets. Model performance was evaluated as follows: first, testing was performed on four independent external validation sets using metrics such as accuracy, F1 score (the harmonic mean of precision and recall), and the area under the receiver operating characteristic curve (AUC) to measure the model's generalizability and accuracy. Second, the classification results of the model were compared with those of junior and mid-level ophthalmologists (two each) using the overlapping confidence interval (CI) comparison method to assess the clinical experience level corresponding to the model's medical proficiency. Results The model achieved an accuracy of 96.78% (59 039/61 003), an F1 score of 82.51% (50 334/61 003), and an AUC of 99.93% (60 960/61 003) on the validation set. On the four external validation sets, it achieved an average accuracy of 87.77% (57 358/65 350), an average precision of 87.06% (56 894/65 350), and an average Kappa value of 82.28%. The average accuracy of FP image lesion identification for junior and mid-level ophthalmologists was 79.00% (79/100) (95%CI 67.71-90.29) and 87.00% (87/100) (95%CI 77.68-96.32), respectively. Conclusions A 16-category FP image lesion recognition model is successfully constructed based on the EfficientNet lightweight convolutional neural network architecture. Its clinical performance preliminarily reaches the level of mid-level ophthalmologists.

Citation： Cheng Hongzhuang, Li Chunxiu, Shen Yadan, Kang Shuangwen, Xu Bingjie, Ning Xinru, Lang Zhuo, Zhong Jie. A study on the identification of lesions in fundus images based on lightweight convolutional neural networks. Chinese Journal of Ocular Fundus Diseases, 2026, 42(2): 119-126. doi: 10.3760/cma.j.cn511434-20250609-00252 Copy

1.	Vofo BN, Chowers I. Suppressing inflammation for the treatment of diabetic retinopathy and age-related macular degeneration: dazdotuftide as a potential new multitarget therapeutic candidate[J/OL]. Biomedicines, 2023, 11(6): 1562[2023-05-27]. https://pubmed.ncbi.nlm.nih.gov/37371657/. DOI: 10.3390/biomedicines11061562.
2.	Li J, Wei D, Mao M, et al. Ultra-widefield color fundus photography combined with high- speed ultra-widefield swept-source optical coherence tomography angiography for non-invasive detection of lesions in diabetic retinopathy[J/OL]. Front Public Health, 2022, 10: 1047608[2022-11-02]. https://pubmed.ncbi.nlm.nih.gov/36408020/. DOI: 10.3389/fpubh.2022.1047608.
3.	Niemeijer M, van Ginneken B, Russell SR, et al. Automated detection and differentiation of drusen, exudates, and cotton-wool spots in digital color fundus photographs for diabetic retinopathy diagnosis[J]. Invest Ophthalmol Vis Sci, 2007, 48(5): 2260-2267. DOI: 10.1167/iovs.06-0996.
4.	Fenner BJ, Wong RLM, Lam WC, et al. Advances in retinal imaging and applications in diabetic retinopathy screening: a review[J]. Ophthalmol Ther, 2018, 7(2): 333-346. DOI: 10.1007/s40123-018-0153-7.
5.	Dallas EG, Clement RA, Taylor DS. Diagnosis from fundus photographs[J]. Br J Ophthalmol, 2007, 91(5): 608-612. DOI: 10.1136/bjo.2006.105726.
6.	Graber ML, Wachter RM, Cassel CK. Bringing diagnosis into the quality and safety equations[J]. JAMA, 2012, 308(12): 1211-1212. DOI: 10.1001/2012.jama.11913.
7.	Arkin E, Yadikar N, Xu X, et al. A survey: object detection methods from CNN to transformer[J]. Multimedia Tools Appl, 2022, 82(14): 21353-21383. DOI: 10.1007/s11042-022-13801-3.
8.	Rodriguez D, Nayak T, Chen Y, et al. On the role of deep learning model complexity in adversarial robustness for medical images[J/OL]. BMC Med Inform Decis Mak, 2022, 22(Suppl 2): 160[2022-06-20]. https://pubmed.ncbi.nlm.nih.gov/35725429/. DOI: 10.1186/s12911-022-01891-w.
9.	Xiao P, Qin Z, Chen D, et al. FastNet: a lightweight convolutional neural network for tumors fast identification in mobile-computer-assisted devices[J]. IEEE IoT J, 2023, 10(11): 9878-9891. DOI: 10.1109/jiot.2023.3235651.
10.	Zhang P, Zhao F, Liu P, et al. Efficient lightweight attention network for face recognition[J] IEEE Access, 2022, 10: 31740-31750. DOI: 101109/ACCESS20223150862.
11.	Tan M, Le QV. EfficientNet: rethinking model scaling for convolutional neural networks[C]//Proceedings of the International Conference on Machine Learning, 2020: 6105-6114.
12.	Lu G, Zhang W, Wang Z. Optimizing depthwise separable convolution operations on GPUs[J]. IEEE Trans Parallel Distrib Syst, 2021, 30: 70-87. DOI: 10.1109/TPDS.2021.3084813.
13.	Seo Y, Kim J, Park U. Swish-T: enhancing swish activation with Tanh bias for improved neural network performance[EB/OL]. (2024-07-01)[2024-07-03]. https://doi.org/10.48550/arXiv.2407.01012 .
14.	Lawrence T, Zhang L. IoTNet: an efficient and accurate convolutional neural network for IoT devices[J]. Sensors, 2019, 19(24): 27. DOI: 10.3390/s19245541.
15.	Dang L, Pang P, Lee J. Depth-wise separable convolution neural network with residual connection for hyperspectral image classification[J/OL]. Remote Sens, 2020, 12(20): 3408[2020-10-17]. https://www.mdpi.com/2072-4292/12/20/3408. DOI: 10.3390/rs12203408.
16.	Wang Z, Xie X, Yang J, et al. RA-Net: reverse attention for generalizing residual learning[J/OL]. Sci Rep, 2024, 14(1): 12771[2024-06-04]. https://pubmed.ncbi.nlm.nih.gov/38834620/. DOI: 10.1038/s41598-024-63623-6.
17.	Yahya AA, Liu K, Hawbani A, et al. A novel image classification method based on residual network, inception, and proposed activation function[J/OL]. Sensors (Basel), 2023, 23(6): 2976[2023-03-09]. https://pubmed.ncbi.nlm.nih.gov/36991687/. DOI: 10.3390/s23062976.
18.	Siu C. Residual networks behave like boosting algorithms[C]//2019 IEEE International Conference on Data Science and Advanced Analytics (DSAA), IEEE, 2019: 1-8.
19.	Liu H, Brock A, Simonyan K, et al. Evolving normalization-activation layers[EB/OL]. (2020-04-06) [2024-07-17]. https://doi.org/10.48550/arXiv.2004.02967 .
20.	Xu J, Zhao Y, Xu F. RDPNet: a single-path lightweight CNN with re-parameterization for CPU-type edge devices[J]. J Cloud Comput, 2022, 11(1): 1-13. DOI: 10.1186/s13677-022-00330-5.
21.	Chen H, Lin M, Sun X, et al. MuffNet: multi-layer feature federation for mobile deep learning[C]//2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW). IEEE, 2019: 2943-2952.
22.	Chang B, Meng L, Haber E, et al. Reversible architectures for arbitrarily deep residual neural networks[C]//Proceedings of the AAAI Conference on Artificial Intelligence, AAAI, 2018.
23.	Howard A, Sandler M, Chu G, et al. Searching for MobileNetV3[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, 2019.
24.	Cen LP, Ji J, Lin JW, et al. Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks[J/OL]. Nat Commun, 12(1): 4828[2021-08-10]. https://pubmed.ncbi.nlm.nih.gov/34376678/. DOI: 10.1038/s41467-021-25138-w.
25.	Rajalakshmi R, Subashini R, Anjana RM, et al. Automated diabetic retinopathy detection in smartphone-based fundus photography using artificial intelligence[J]. Eye, 2018, 32(6): 1138-1144. DOI: 10.1038/s41433-018-0064-9.
26.	Li B, Chen H, Yu W, et al. The performance of a deep learning system in assisting junior ophthalmologists in diagnosing 13 major fundus diseases: a prospective multi-center clinical trial[J/OL]. NPJ Digit Med, 2024, 7(1): 8[2024-01-11]. https://pubmed.ncbi.nlm.nih.gov/38212607/. DOI: 10.1038/s41746-023-00991-9.
27.	Pietro?aj M, Blok M. Neural network training with limited precision and asymmetric exponent[J]. J Big Data, 2022, 9(1): 1-17. DOI: 10.1186/s40537-022-00606-2.
28.	Dubiel M, Barghouti Y, Kudryavtseva K, et al. Author correction: on-device query intent prediction with lightweight LLMs to support ubiquitous conversations[J/OL]. Sci Rep, 2024, 14(1): 16062[2024-07-11]. https://pubmed.ncbi.nlm.nih.gov/38992104/. DOI: 10.1038/s41598-024-66714-6.
29.	Chen X, Xue Y, Wu X, et al. Deep learning-based system for disease screening and pathologic region detection from optical coherence tomography images[J/OL]. Transl Vis Sci Technol, 2023, 12(1): 29[2023-01-03]. https://pubmed.ncbi.nlm.nih.gov/36716039/. DOI: 10.1167/tvst.12.1.29.
30.	Rahul M, Gupta M, Kumar R, et al. A deep learning based critical analysis of skin lesion classification[C]//2024 International Conference on Advances in Modern Age Technologies for Health and Engineering Science (AMATHE), IEEE, 2024: 1-5.
31.	Allmark P. Should research samples reflect the diversity of the population?[J]. J Med Ethics, 2004, 30(2): 185-189. DOI: 10.1136/jme.2003.004374.
32.	Kaur M, Singh D. Multi-modality medical image fusion technique using multi-objective differential evolution based deep neural networks[J]. J Ambient Intell Humaniz Comput, 2021, 12(2): 2483-2493. DOI: 10.1007/S12652-020-02386-0.

1. Vofo BN, Chowers I. Suppressing inflammation for the treatment of diabetic retinopathy and age-related macular degeneration: dazdotuftide as a potential new multitarget therapeutic candidate[J/OL]. Biomedicines, 2023, 11(6): 1562[2023-05-27]. https://pubmed.ncbi.nlm.nih.gov/37371657/. DOI: 10.3390/biomedicines11061562.
2. Li J, Wei D, Mao M, et al. Ultra-widefield color fundus photography combined with high- speed ultra-widefield swept-source optical coherence tomography angiography for non-invasive detection of lesions in diabetic retinopathy[J/OL]. Front Public Health, 2022, 10: 1047608[2022-11-02]. https://pubmed.ncbi.nlm.nih.gov/36408020/. DOI: 10.3389/fpubh.2022.1047608.
3. Niemeijer M, van Ginneken B, Russell SR, et al. Automated detection and differentiation of drusen, exudates, and cotton-wool spots in digital color fundus photographs for diabetic retinopathy diagnosis[J]. Invest Ophthalmol Vis Sci, 2007, 48(5): 2260-2267. DOI: 10.1167/iovs.06-0996.
4. Fenner BJ, Wong RLM, Lam WC, et al. Advances in retinal imaging and applications in diabetic retinopathy screening: a review[J]. Ophthalmol Ther, 2018, 7(2): 333-346. DOI: 10.1007/s40123-018-0153-7.
5. Dallas EG, Clement RA, Taylor DS. Diagnosis from fundus photographs[J]. Br J Ophthalmol, 2007, 91(5): 608-612. DOI: 10.1136/bjo.2006.105726.
6. Graber ML, Wachter RM, Cassel CK. Bringing diagnosis into the quality and safety equations[J]. JAMA, 2012, 308(12): 1211-1212. DOI: 10.1001/2012.jama.11913.
7. Arkin E, Yadikar N, Xu X, et al. A survey: object detection methods from CNN to transformer[J]. Multimedia Tools Appl, 2022, 82(14): 21353-21383. DOI: 10.1007/s11042-022-13801-3.
8. Rodriguez D, Nayak T, Chen Y, et al. On the role of deep learning model complexity in adversarial robustness for medical images[J/OL]. BMC Med Inform Decis Mak, 2022, 22(Suppl 2): 160[2022-06-20]. https://pubmed.ncbi.nlm.nih.gov/35725429/. DOI: 10.1186/s12911-022-01891-w.
9. Xiao P, Qin Z, Chen D, et al. FastNet: a lightweight convolutional neural network for tumors fast identification in mobile-computer-assisted devices[J]. IEEE IoT J, 2023, 10(11): 9878-9891. DOI: 10.1109/jiot.2023.3235651.
10. Zhang P, Zhao F, Liu P, et al. Efficient lightweight attention network for face recognition[J] IEEE Access, 2022, 10: 31740-31750. DOI: 101109/ACCESS20223150862.
11. Tan M, Le QV. EfficientNet: rethinking model scaling for convolutional neural networks[C]//Proceedings of the International Conference on Machine Learning, 2020: 6105-6114.
12. Lu G, Zhang W, Wang Z. Optimizing depthwise separable convolution operations on GPUs[J]. IEEE Trans Parallel Distrib Syst, 2021, 30: 70-87. DOI: 10.1109/TPDS.2021.3084813.
13. Seo Y, Kim J, Park U. Swish-T: enhancing swish activation with Tanh bias for improved neural network performance[EB/OL]. (2024-07-01)[2024-07-03]. https://doi.org/10.48550/arXiv.2407.01012 .
14. Lawrence T, Zhang L. IoTNet: an efficient and accurate convolutional neural network for IoT devices[J]. Sensors, 2019, 19(24): 27. DOI: 10.3390/s19245541.
15. Dang L, Pang P, Lee J. Depth-wise separable convolution neural network with residual connection for hyperspectral image classification[J/OL]. Remote Sens, 2020, 12(20): 3408[2020-10-17]. https://www.mdpi.com/2072-4292/12/20/3408. DOI: 10.3390/rs12203408.
16. Wang Z, Xie X, Yang J, et al. RA-Net: reverse attention for generalizing residual learning[J/OL]. Sci Rep, 2024, 14(1): 12771[2024-06-04]. https://pubmed.ncbi.nlm.nih.gov/38834620/. DOI: 10.1038/s41598-024-63623-6.
17. Yahya AA, Liu K, Hawbani A, et al. A novel image classification method based on residual network, inception, and proposed activation function[J/OL]. Sensors (Basel), 2023, 23(6): 2976[2023-03-09]. https://pubmed.ncbi.nlm.nih.gov/36991687/. DOI: 10.3390/s23062976.
18. Siu C. Residual networks behave like boosting algorithms[C]//2019 IEEE International Conference on Data Science and Advanced Analytics (DSAA), IEEE, 2019: 1-8.
19. Liu H, Brock A, Simonyan K, et al. Evolving normalization-activation layers[EB/OL]. (2020-04-06) [2024-07-17]. https://doi.org/10.48550/arXiv.2004.02967 .
20. Xu J, Zhao Y, Xu F. RDPNet: a single-path lightweight CNN with re-parameterization for CPU-type edge devices[J]. J Cloud Comput, 2022, 11(1): 1-13. DOI: 10.1186/s13677-022-00330-5.
21. Chen H, Lin M, Sun X, et al. MuffNet: multi-layer feature federation for mobile deep learning[C]//2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW). IEEE, 2019: 2943-2952.
22. Chang B, Meng L, Haber E, et al. Reversible architectures for arbitrarily deep residual neural networks[C]//Proceedings of the AAAI Conference on Artificial Intelligence, AAAI, 2018.
23. Howard A, Sandler M, Chu G, et al. Searching for MobileNetV3[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV), IEEE, 2019.
24. Cen LP, Ji J, Lin JW, et al. Automatic detection of 39 fundus diseases and conditions in retinal photographs using deep neural networks[J/OL]. Nat Commun, 12(1): 4828[2021-08-10]. https://pubmed.ncbi.nlm.nih.gov/34376678/. DOI: 10.1038/s41467-021-25138-w.
25. Rajalakshmi R, Subashini R, Anjana RM, et al. Automated diabetic retinopathy detection in smartphone-based fundus photography using artificial intelligence[J]. Eye, 2018, 32(6): 1138-1144. DOI: 10.1038/s41433-018-0064-9.
26. Li B, Chen H, Yu W, et al. The performance of a deep learning system in assisting junior ophthalmologists in diagnosing 13 major fundus diseases: a prospective multi-center clinical trial[J/OL]. NPJ Digit Med, 2024, 7(1): 8[2024-01-11]. https://pubmed.ncbi.nlm.nih.gov/38212607/. DOI: 10.1038/s41746-023-00991-9.
27. Pietro?aj M, Blok M. Neural network training with limited precision and asymmetric exponent[J]. J Big Data, 2022, 9(1): 1-17. DOI: 10.1186/s40537-022-00606-2.
28. Dubiel M, Barghouti Y, Kudryavtseva K, et al. Author correction: on-device query intent prediction with lightweight LLMs to support ubiquitous conversations[J/OL]. Sci Rep, 2024, 14(1): 16062[2024-07-11]. https://pubmed.ncbi.nlm.nih.gov/38992104/. DOI: 10.1038/s41598-024-66714-6.
29. Chen X, Xue Y, Wu X, et al. Deep learning-based system for disease screening and pathologic region detection from optical coherence tomography images[J/OL]. Transl Vis Sci Technol, 2023, 12(1): 29[2023-01-03]. https://pubmed.ncbi.nlm.nih.gov/36716039/. DOI: 10.1167/tvst.12.1.29.
30. Rahul M, Gupta M, Kumar R, et al. A deep learning based critical analysis of skin lesion classification[C]//2024 International Conference on Advances in Modern Age Technologies for Health and Engineering Science (AMATHE), IEEE, 2024: 1-5.
31. Allmark P. Should research samples reflect the diversity of the population?[J]. J Med Ethics, 2004, 30(2): 185-189. DOI: 10.1136/jme.2003.004374.
32. Kaur M, Singh D. Multi-modality medical image fusion technique using multi-objective differential evolution based deep neural networks[J]. J Ambient Intell Humaniz Comput, 2021, 12(2): 2483-2493. DOI: 10.1007/S12652-020-02386-0.

Chinese Journal of Ocular Fundus Diseases

A study on the identification of lesions in fundus images based on lightweight convolutional neural networks

Abstract Full text Figures/Tables Video References Cited by

Previous Article

Next Article

Format

Content