Deep Learning Methods for Retinal Disease Diagnosis: History
View Latest Version
Please note this is an old version of this entry, which may differ significantly from the current revision.
Subjects: Ophthalmology | Computer Science, Artificial Intelligence
Contributor:
Mohan Bhandari
,
Tej Bahadur Shahi
,
Arjun Neupane

The advancement of digital medical imaging has brought about a significant change in ophthalmology as it has introduced effective technologies that help in the detection of such diseases. By improving early detection through image analysis and identifying minuscule anomalies, Artificial Intelligence (AI) has considerably coped with retinal diseases. Different Machine Learning (ML) and Convolutional Neural Networks (CNNs) are efficient at analyzing images and are particularly incredible at recognizing complex patterns in medical images.

  • adversarial attacks
  • deep learning
  • health informatics
  • retinal image classification

1. Introduction

The retina, situated at the posterior aspect of the ocular globe, comprises photoreceptor cells adept at transducing luminous stimuli into intricate electrical signals, subsequently dispatched to the cerebral cortex via the optic nerve. This intricate process serves as the foundation for human visual perception, wherein the brain deciphers these electrical transmissions as coherent visual representations. Retinal diseases can seriously affect vision and in some cases, can lead to permanent blindness [1] which is a big problem for the general health of the public. Getting a prompt and accurate diagnosis with the help of automated tools is a great assist for medical specialists in making wise medical decisions. The advancement of digital medical imaging has brought about a significant change in ophthalmology as it has introduced effective technologies that help in the detection of such diseases. By improving early detection through image analysis and identifying minuscule anomalies, Artificial Intelligence (AI) has considerably coped with retinal diseases. AI has also enhanced treatment planning by analyzing patient data and enabling tailored care. Additionally, AI-driven systems help track the development of diseases, resulting in therapies that are more successful [2].
Different Machine Learning (ML) and Convolutional Neural Networks (CNNs) are efficient at analyzing images and are particularly incredible at recognizing complex patterns in medical images [3]. Their ability to diagnose complicated retinal diseases is efficient without a doubt, but in medical practice, using CNNs depends not only on how well they can diagnose the issues but also on how useful they are in places with limited computational resources. Not only CNN, but different variants of CNN like ResNet [4], VGG [5] and more have produced good accuracies statistically. These CNNs and their variants have a very high number of training parameters, and many layers which make it time-consuming in real-time predictions [6] and integration with the Internet-of-Medical-Things (IoMT) [7].
As AI technology advances, it has become essential to not only achieve better diagnostic abilities but also to understand how these AI systems make predictions and decisions [8][9][10]. As these models can be hard to understand because of their statistical nature making them black boxes [11], the addition of Explainable Artificial Intelligence (XAI) into these models can solve the problem. The combination of small and efficient CNN models in IoMT devices with XAI, as a bio-marker, helps retinal disease diagnosis to be more accurate and more accessible for medical experts, practitioners, and even ordinary people.

2. Deep Learning Methods for Retinal Disease Diagnosis

Researchers in health informatics are leveraging the predictive power of Deep Learning (DL) to address the automated diagnosis of various diseases such as COVID-19 [12], monkeypox [13], kidney stone [14] and so on. This section summarises the recent DL methods that have been employed for retinal disease diagnosis using various image modalities. These methods can be categorized into two broad classes: pre-trained DL models (that leverage the transfer learning strategies) and custom-designed CNN (which needs training from scratch).
Subramanian et al. [15] utilised four CNN models such as VGG16, DenseNet-201, Inception-V3, and Xception, to classify seven different retinal diseases. Moreover, Bayesian optimization was employed to fine-tune the hyperparameters of these CNN models, coupled with image augmentation techniques to enhance their ability to generalize. The use of DenseNet-201 in classifying retinal diseases on the Retinal OCT Image dataset resulted in an accuracy exceeding 99%, demonstrating superior performance compared to alternative methods. Puneet et al. [16] implemented the combination of Attention and Transfer Learning approaches into a DCNN for categorizing retinal diseases such as CNV, DME, and Drusen using OCT images. Their proposal achieved notable results, attaining accuracies of 97.79% during training and 95.6% during testing. Kayadibi et al. [17] implemented a hybrid fine-tuned CNN for retinal disease classification from OCT images. They utilized PCA to reduce the feature size and enhance the performance. The benchmarking outcomes for two OCT datasets demonstrated a high level of promise in terms of accuracy. Specifically, the UCSD dataset yielded an impressive accuracy of 99.70% according to Kermany et al.’s study [18], while the Duke dataset achieved a perfect 100% accuracy as reported by Srinivasan et al. [19]. In their research, Kim and colleagues [20] harnessed a variety of Convolutional Neural Networks (CNNs) like VGG16, ResNet50, DenseNet121, and Inception-v3 as feature extractors. Subsequently, they employed these features to develop binary OCT image classification models. A binary classifier model is developed for each category (CNV, DME, Drusen and Normal) and the VGG-16-based model for CNV vs. other classes achieved 98.6% accuracy. They achieve high accuracy using the pre-trained DL models. However, their proposal needs the training of individual models for each class which incurs high computational complexity. A pre-trained VGG-16 network was implemented by Li et al. [21] for retinal image classification on OCT images. They validated the model’s performance on 1000 independent OCT images. Their work revealed that the transfer learning with the VGG-16 model has a promising accuracy of 98.6%, sensitivity of 97.8%, and specificity of 00.4%. With such commendable performance of the model, deep learning can automate the diagnosis of retinal disease. Li et al. [22] adopted the ensemble models for retinal disease classification using OCT images. They trained four DL- models based on improved ResNet50 to build the ensemble and achieved the highest accuracy of 96.3%, sensitivity of 96.6%, and specificity of 98.7%. However, the ResNet50 model is itself the heavyweight model.
In addition to employing pre-trained deep learning models, only a limited number of researchers have created custom CNNs for the classification of retinal images. For example, a deep CNN with six convolution blocks (including the Relu, batch normalization, and pooling operation) was implemented by Sujina et al. [23]. Their proposal achieved a promising accuracy of 99.69% with a low misclassification rate. However, the generalisability of the CNN on additional datasets is not reported. Altan et al. [24] implemented a deep learning architecture to detect the macular edema on OCT images and reported an accuracy of 99.20%.
Hybrid deep learning models for retinal image classification have also been proposed recently. For instance, a hybrid deep learning model for OCT image classification was implemented by Khan et al. [25]. They extracted retinal features from OCT images using three pre-trained deep learning models (DenseNet121, InceptionV3, and ResNet50), and ant colony optimization was used for best feature selection. Finally, the SVM and KNN were employed for classification. Their proposal achieved high performance on OCT image classification. However, the approach is not applicable to end-to-end training of the model.

This entry is adapted from the peer-reviewed paper 10.3390/jimaging9100219

References

  1. Varela, M.D.; Conti, G.M.; Malka, S.; Vaclavik, V.; Mahroo, O.A.; Webster, A.R.; Tran, V.; Michaelides, M. Coats-like vasculopathy in inherited retinal disease: Prevalence, characteristics, genetics and management. Ophthalmology 2023.
  2. Tariq, A.; Gill, A.Y.; Hussain, H.K. Evaluating the Potential of Artificial Intelligence in Orthopedic Surgery for Value-based Healthcare. Int. J. Multidiscip. Sci. Arts 2023, 2, 27–35.
  3. Mall, P.K.; Singh, P.K.; Srivastav, S.; Narayan, V.; Paprzycki, M.; Jaworska, T.; Ganzha, M. A comprehensive review of deep neural networks for medical image processing: Recent developments and future opportunities. Healthc. Anal. 2023, 4, 100216.
  4. Kumari, A.A.; Bhagat, A.; Henge, S.K.; Mandal, S.K. Automated Decision Making ResNet Feed-Forward Neural Network based Methodology for Diabetic Retinopathy Detection. Int. J. Adv. Comput. Sci. Appl. 2023, 14, 303–314.
  5. Udayaraju, P.; Jeyanthi, P.; Sekhar, B. A hybrid multilayered classification model with VGG-19 net for retinal diseases using optical coherence tomography images. Soft Comput. 2023, 27, 12559–12570.
  6. Narayan, V.; Mall, P.K.; Alkhayyat, A.; Abhishek, K.; Kumar, S.; Pandey, P. Enhance-Net: An Approach to Boost the Performance of Deep Learning Model Based on Real-Time Medical Images. J. Sensors 2023, 2023, 8276738.
  7. Sinha, A.; Garcia, D.W.; Kumar, B.; Banerjee, P. Application of Big Data Analytics and Internet of Medical Things (IoMT) in Healthcare with View of Explainable Artificial Intelligence: A Survey. In Interpretable Cognitive Internet of Things for Healthcare; Springer: Berlin/Heidelberg, Germany, 2023; pp. 129–163.
  8. Carter, S.M.; Rogers, W.; Win, K.T.; Frazer, H.; Richards, B.; Houssami, N. The ethical, legal and social implications of using artificial intelligence systems in breast cancer care. Breast 2020, 49, 25–32.
  9. Nahavandi, S. Industry 5.0—A human-centric solution. Sustainability 2019, 11, 4371.
  10. Abou El Houda, Z.; Brik, B.; Khoukhi, L. “why should i trust your ids?”: An explainable deep learning framework for intrusion detection systems in internet of things networks. IEEE Open J. Commun. Soc. 2022, 3, 1164–1176.
  11. Adadi, A.; Berrada, M. Peeking inside the black-box: A survey on explainable artificial intelligence (XAI). IEEE Access 2018, 6, 52138–52160.
  12. Bhandari, M.; Shahi, T.B.; Siku, B.; Neupane, A. Explanatory classification of CXR images into COVID-19, Pneumonia and Tuberculosis using deep learning and XAI. Comput. Biol. Med. 2022, 150, 106156.
  13. Sitaula, C.; Shahi, T.B. Monkeypox virus detection using pre-trained deep learning-based approaches. J. Med. Syst. 2022, 46, 78.
  14. Islam, M.N.; Hasan, M.; Hossain, M.; Alam, M.; Rabiul, G.; Uddin, M.Z.; Soylu, A. Vision transformer and explainable transfer learning models for auto detection of kidney cyst, stone and tumor from CT-radiography. Sci. Rep. 2022, 12, 11440.
  15. Subramanian, M.; Kumar, M.S.; Sathishkumar, V.; Prabhu, J.; Karthick, A.; Ganesh, S.S.; Meem, M.A. Diagnosis of retinal diseases based on Bayesian optimization deep learning network using optical coherence tomography images. Comput. Intell. Neurosci. 2022, 2022, 8014979.
  16. Puneet; Kumar, R.; Gupta, M. Optical coherence tomography image based eye disease detection using deep convolutional neural network. Health Inf. Sci. Syst. 2022, 10, 13.
  17. Kayadibi, İ.; Güraksın, G.E.; Köse, U. A Hybrid R-FTCNN based on principal component analysis for retinal disease detection from OCT images. Expert Syst. Appl. 2023, 230, 120617.
  18. Kermany, D.S.; Goldbaum, M.; Cai, W.; Valentim, C.C.; Liang, H.; Baxter, S.L.; McKeown, A.; Yang, G.; Wu, X.; Yan, F.; et al. Identifying medical diagnoses and treatable diseases by image-based deep learning. Cell 2018, 172, 1122–1131.
  19. Srinivasan, P.P.; Kim, L.A.; Mettu, P.S.; Cousins, S.W.; Comer, G.M.; Izatt, J.A.; Farsiu, S. Fully automated detection of diabetic macular edema and dry age-related macular degeneration from optical coherence tomography images. Biomed. Opt. Express 2014, 5, 3568–3577.
  20. Kim, J.; Tran, L. Retinal disease classification from oct images using deep learning algorithms. In Proceedings of the 2021 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB), Melbourne, Australia, 13–15 October 2021; pp. 1–6.
  21. Li, F.; Chen, H.; Liu, Z.; Zhang, X.; Wu, Z. Fully automated detection of retinal disorders by image-based deep learning. Graefes Arch. Clin. Exp. Ophthalmol. 2019, 257, 495–505.
  22. Li, F.; Chen, H.; Liu, Z.; Zhang, X.D.; Jiang, M.S.; Wu, Z.Z.; Zhou, K.Q. Deep learning-based automated detection of retinal diseases using optical coherence tomography images. Biomed. Opt. Express 2019, 10, 6204–6226.
  23. Sunija, A.; Kar, S.; Gayathri, S.; Gopi, V.P.; Palanisamy, P. Octnet: A lightweight cnn for retinal disease classification from optical coherence tomography images. Comput. Methods Programs Biomed. 2021, 200, 105877.
  24. Altan, G. DeepOCT: An explainable deep learning architecture to analyze macular edema on OCT images. Eng. Sci. Technol. Int. J. 2022, 34, 101091.
  25. Khan, A.; Pin, K.; Aziz, A.; Han, J.W.; Nam, Y. Optical coherence tomography image classification using hybrid deep learning and ant colony optimization. Sensors 2023, 23, 6706.
More
© Text is available under the terms and conditions of the Creative Commons Attribution (CC BY) license; additional terms may apply. By using this site, you agree to the Terms and Conditions and Privacy Policy.
This entry is offline, you can click here to edit this entry!
Video Production Service
Feedback