Lightweight Adaptive Grading Model for Diabetic Retinopathy Based on Ultra-Widefield Fundus Images

Xinpeng Miao; Shuaichao Feng; Jie Zhao; Guangping Zhuo; Fei Ma

doi:10.62051/ijcsit.v5n1.17

Authors

Xinpeng Miao
Shuaichao Feng
Jie Zhao
Guangping Zhuo
Fei Ma

DOI:

https://doi.org/10.62051/ijcsit.v5n1.17

Keywords:

Kolmogorov-Arnold, Learnable nonlinear functions, Legendre polynomial, CLAHE, Lightweight

Abstract

Wide-field fundus images play a crucial role in the diagnosis of diabetic retinopathy. However, traditional convolutional neural networks face limitations when processing such images, struggling to balance local details and global structures. To address this, a novel lightweight feature extraction network, called the Adaptive Nonlinear Lightweight Retina Network (ANLR-Net), is proposed. The network consists of two key modules: the Adaptive Feature Mapping Module (AFMM) and the Lightweight Module (LEM). AFMM replaces traditional linear weight matrices with learnable nonlinear functions, enabling precise capture of both global structures and local details while reducing network parameters and improving learning efficiency. LEM performs dimensionality reduction and restoration using 1×1 convolutions, which reduces computational cost while maintaining the network's lightweight nature. ANLR-Net effectively captures multi-scale features in wide-field images, significantly reducing computational complexity and parameter count. Experimental results show that ANLR-Net achieves a classification accuracy of 95.05%, specificity of 98.64%, and an AUC of 0.9712 on the wide-field dataset, outperforming traditional models.

Downloads

Download data is not yet available.

References

[1] CHENG H-T, XU X, LIM P S, et al. Worldwide epidemiology of diabetes-related end-stage renal disease, 2000–2015 [J]. 2021, 44(1): 89-97.

[2] GUO L, XIAO X J A M. Guideline for the Management of Diabetes Mellitus in the Elderly in China (2024 Edition) [J]. 2024, 7(1): 5-51.

[3] TOMIĆ M, VRABEC R, LJUBIĆ S, et al. Patients with Type 2 Diabetes, Higher Blood Pressure, and Infrequent Fundus Examinations Have a Higher Risk of Sight-Threatening Retinopathy [J]. 2024, 13(9): 2496.

[4] ZHAO M, CHANDRA A, LIU L, et al. Investigation of the reasons for delayed presentation in proliferative diabetic retinopathy patients [J]. 2024, 19(2): e0291280.

[5] LI H, JIA W, VUJOSEVIC S, et al. Current Research and Future Strategies for the Management of Vision-Threatening Diabetic Retinopathy [J]. 2024: 100109.

[6] GREEN J, SIDDALL H, MURDOCH I J S S, et al. Learning to live with glaucoma: a qualitative study of diagnosis and the impact of sight loss [J]. 2002, 55(2): 257-67.

[7] ZHANG Z, DENG C, PAULUS Y M J B. Advances in Structural and Functional Retinal Imaging and Biomarkers for Early Detection of Diabetic Retinopathy [J]. 2024, 12(7): 1405.

[8] SORRENTINO F S, GARDINI L, FONTANA L, et al. Novel Approaches for Early Detection of Retinal Diseases Using Artificial Intelligence [J]. 2024, 14(7): 690.

[9] IKRAM A, IMRAN A, LI J, et al. A systematic review on fundus image-based diabetic retinopathy detection and grading: Current status and future directions [J]. 2024.

[10] NIGAM A, SUN J, SUBHASH V, et al. Deep Learning Approach to Identify Diabetic Retinopathy Severity and Progression Using Ultra-Wide Field Retinal Images [M]. AI for Health Equity and Fairness: Leveraging AI to Address Social Determinants of Health. Springer. 2024: 103-16.

[11] HEMANTH D J, DEPERLIOGLU O, KOSE U J N C, et al. RETRACTED ARTICLE: An enhanced diabetic retinopathy detection and classification approach using deep convolutional neural network [J]. 2020, 32(3): 707-21.

[12] KANDEL I, CASTELLI M J A S. Transfer learning with convolutional neural networks for diabetic retinopathy image classification. A review [J]. 2020, 10(6): 2021.

[13] SHANKAR K, SAIT A R W, GUPTA D, et al. Automated detection and classification of fundus diabetic retinopathy images using synergic deep learning model [J]. 2020, 133: 210-6.

[14] HUANG Y, LIN L, CHENG P, et al. Lesion-based contrastive learning for diabetic retinopathy grading from fundus images; proceedings of the Medical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part II 24, F, 2021 [C]. Springer.

[15] HOU J, XU J, XIAO F, et al. Cross-field transformer for diabetic retinopathy grading on two-field fundus images; proceedings of the 2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), F, 2022 [C]. IEEE.

[16] SOLANO A, DIETRICH K N, MARTíNEZ-SOBER M, et al. Deep learning architectures for diagnosis of diabetic retinopathy [J]. 2023, 13(7): 4445.

[17] ZHANG G, SUN B, CHEN Z, et al. Diabetic retinopathy grading by deep graph correlation network on retinal images without manual annotations [J]. 2022, 9: 872214.

[18] HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks; proceedings of the Proceedings of the IEEE conference on computer vision and pattern recognition, F, 2017 [C].

[19] SAMAD S A, GITANJALI J J I A. Augmenting DenseNet: Leveraging Multi-Scale Skip Connections for Effective Early-Layer Information Incorporation [J]. 2024.

[20] RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation; proceedings of the Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18, F, 2015 [C]. Springer.