Classification of Diabetic Retinopathy using Deep Learning

Abstract

Diabetic retinopathy, a potentially fatal retinal disease, occurs in individuals with diabetes, causing vision damage and possible visual impairment traditionally, screening for diabetic retinopathy has been the norm hard by ophthalmologists manually. To facilitate this, we have DR. Our pre-trained model is trained on a large dataset of about 3662 training images, enabling it to automatically detect the DR platform in high-resolution fundus images The dataset used for this is publicly available on kaggle. The DR phases are categorized as 0, 1, 2, 3, and 4. The fundus eye images serve as the input parameters of the model. The pre-trained model then identifies the point of interest in bank eye images, then provides activation function insights By calculating weights, for example patients’ visual intensity levels, it helps classify diabetic retinopathy images into their intensity groups appropriately within the possibility of test- methods Provides variation.

Introduction

I. INTRODUCTION

Diabetic retinopathy (DR) occurs as a result of structural changes in the micro vascular system of the retina, leading to visual deterioration and potential blindness affecting individuals of all ages worldwide. Early detection of DR is crucial, with research suggesting that 90% of cases can be successfully treated if identified promptly [1]. DR primarily affects various eye structures, including hemorrhages, exudates, and microaneurysms. Exudates, characterized by fluid leakage from retinal blood vessels, are a common early symptom and play a pivotal role in accurate DR screenings shown in figure1. Detecting these exudates not only aids in disease classification but also enhances automated screening methods. Analyzing these structural changes facilitates the identification of DR severity.

Diabetic macular edema (DME), stemming from fluid leakage in the macular region, further complicates the scenario. There exists a complex interplay between DR and DME, necessitating their joint assessment. The risk of DR-DME escalation is heightened when exudates encroach upon the macula [2]. However, conventional methods such as laser photocoagulation for identifying and treating severe DR-DME stages are time-consuming. Early detection of DR-DME, on the other hand, holds promise for effectively managing the conditions and preventing vision loss.

Diabetes mellitus, characterized by elevated blood glucose levels due to insulin deficiency, contributes to various complications, including retinopathy. Both type 1 and type 2 diabetes patients are susceptible to DR, wherein increased glucose levels adversely affect retinal blood vessels. Diabetic retinopathy (DR) presents in two different forms: non- proliferative diabetic retinopathy (NPDR) and proliferative diabetic retinopathy (PDR). NPDR is marked by vessel damage and fluid leakage, presenting clinical signs like microaneurysms, hemorrhages [3], and exudates. PDR, conversely, involves abnormal blood vessel growth on the retinal surface. These clinical manifestations are readily observable through fundus imaging, though their severity and progression rates may vary among patients. NPDR is further categorized into four grades, each exhibiting distinct symptoms and implications as in the table 1.

[4]In the study "Detection of Diabetic Retinopathy and its Classification from Fundus Images" utilizes artificial intelligence for diabetic retinopathy (DR) detection. Employing a binary classification model, the methodology achieves an accuracy rate of 83.12% in identifying DR cases. This approach showcases the potential of AI-based models in accurately diagnosing DR from fundus images, offering a promising avenue for early detection and intervention. By leveraging advanced computational techniques, such as machine learning, the study contributes to improving the efficiency and accuracy of DR screening, ultimately enhancing patient outcomes and reducing the risk of vision-threatening complications associated with diabetes.

[5]The research paper entitled "Applying Supervised Contrastive Learning for Diabetic Retinopathy Detection and Severity Level Assessment from Fundus Images" is authored by Md Robiul Islam, Lway Faisal Abdulrazak, and Md.. Utilizing the "APTOS 2019 Blindness Detection" dataset, they employ supervised contrastive learning (SCL) to identify diabetic retinopathy (DR) and its severity stages from fundus images. Their approach achieves an 84.36% accuracy for multiclass classification, demonstrating SCL's effectiveness in accurate DR detection and severity level classification.

II. LITERATURE REVIEW

Diabetic retinopathy (DR) is a widespread complication linked to diabetes and remains a major contributor to global blindness. Timely detection and precise diagnosis of DR are imperative for mitigating vision loss among diabetic individuals. Traditional DR diagnostic methods involve manual scrutiny of retinal images by skilled ophthalmologists, a process prone to human error and time-consuming nature.

The advent of deep learning has sparked interest in employing machine learning algorithms, notably Convolutional Neural Networks (CNNs), for automating DR diagnosis. CNN models, such as VGG16 and ResNet50[6], have showcased exceptional performance across diverse image classification tasks, including medical image analysis like retinal image classification.

Transfer learning, a methodology wherein knowledge acquired from training one model is transferred to another, has gained traction for fine-tuning pre-trained CNNs for specialized tasks like DR diagnosis. By leveraging insights gleaned from extensive image datasets like ImageNet, transfer learning empowers models to generalize better to novel tasks with limited training data, thereby enhancing their efficacy and performance.

Cohen's Kappa metric emerges as a statistical tool assessing agreement between two raters beyond random chance. In the context of DR diagnosis, Cohen's Kappa offers a holistic evaluation of model performance by considering not just prediction accuracy but also alignment with human expert judgments. This metric proves invaluable for evaluating the reliability and consistency of automated DR diagnosis systems. In essence, the paper delineates an integrated methodology amalgamating ensemble modelling, transfer learning, and robust performance evaluation utilizing Cohen's Kappa metric for precise and dependable diagnosis of diabetic retinopathy. By harnessing cutting-edge deep learning techniques, this proposed approach holds promise in advancing early detection and management of DR, thus diminishing the risk of vision impairment in diabetic patients.

III. METHODOLOGY

Machines are advancing towards interacting with the world much like humans through the utilization of computer vision. This technology leverages pattern recognition algorithms and vast visual data training to mimic the human brain's ability to perceive visual information. Primarily powered by deep learning, convolutional neural networks (CNNs)[7] have revolutionized the field of computer vision. By employing deep learning methodologies, efforts are underway to address the challenge of developing an automated ensemble model for classifying the severity of diabetic retinopathy.

Conclusion

The integration of deep learning in diabetic retinopathy detection shows promise for early and accurate diagnosis. Challenges include model interpretability, bias in datasets, and ethical data use. Validation, regulatory approvals, and model performance comparison underscore the need for rigorous testing. ResNet50 outperforms VGG16, showing higher Cohen Kappa scores, indicating better classification agreement.

References

[1] M. T. Hagos and S. Kant, “Transfer learning-based detection of DR from small dataset,” arXiv preprint arXiv:1905.07203, 2019. [2] D.S.W. Ting, L. Carin, M.D. Abramoff, Observations and lessons learned from the artificial intelligence studies for DR screening, JAMA Ophthalmol. 06 (2019). [3] M.J. Van Grinsven, B. van Ginneken, C.B. Hoyng, T. Theelen, C.I. S´ anchez, Fast convolutional neural network training using selective data sampling: application to hemorrhage detection in color fundus images, IEEE Trans. Med. Imaging 35 (5) (2016) 1273–1284. [4] Shelar, M., Gaitonde, S., Senthilkumar, A., Mundra, M., & Sarang, A. (2021). Detection of Diabetic Retinopathy and its Classification from the Fundus Images. In 2021 International Conference on Computer Communication and Informatics (ICCCI) (pp. 1-5). IEEE. DOI:10.1109/ICCCI50826.2021.9402347 [5] Islam, M. R., Abdulrazak, L. F., Nahiduzzaman, M., Goni, M. O. F., Anower, M. S., Ahsan, M., Haider, J., & Kowalski, M. (Year). Applying supervised contrastive learning for the detection of diabetic retinopathy and its severity levels from fundus images. [6] AATILA, M., LACHGAR, M., HRIMECH, H., & KARTIT, A. (Year). Diabetic Retinopathy Classification Using ResNet50 and VGG-16 Pretrained Networks. LTI Laboratory, ENSA, Chouaib Doukkali University, El Jadida 1166, Morocco; University Hassan 1st, ENSA of Berrechid, LAMSAD, B.P 218, Morocco. [7] M. Jogin, Mohana, M. S. Madhulika, G. D. Divya, R. K. Meghana and S. Apoorva, \"Feature Extraction using Convolution Neural Networks (CNN) and Deep Learning,\" 2018 3rd IEEE International Conference on Recent Trends in Electronics, Information & Communication Technology (RTEICT), Bangalore, India, 2018, pp. 2319-2323, doi: 10.1109/RTEICT42901.2018.9012507. [8] https://towardsdatascience.com/understand-local-receptive- fields-in-convolutional-neural-networks-f26d700be16c [9] Alzubaidi, L., Zhang, J., Humaidi, A.J. et al. Review of deep learning: concepts, CNN architectures, challenges, applications, future directions. J Big Data 8, 53 (2021) [10] A. Ajit, K. Acharya and A. Samanta, \"A Review of Convolutional Neural Networks,\" 2020 International Conference on Emerging Trends in Information Technology and Engineering (ic-ETITE), Vellore, India, 2020, pp. 1-5, doi: 10.1109/ic-ETITE47903.2020.049.

Copyright

Copyright © 2024 Mr. V. Purna Chandra Reddy , Sarabu Vignya Sharma , M. Reshmi , J. Sahithi Reddy , D. Sai Priya . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.