Classification of Eye Disease from Fundus Images Using EfficientNet

Abstract

Studies show that at least 2.2 billion people in the world have some kind of visual impairment or blindness. The prevalence of conditions progressing into preventable blindness is quite high. As more and more public data sets are available, the training of deep learning in the medical field is a possible choice, but the practical application of deep learning in clinical practice is still an open issue. We work for solving this problem and continue developing clinical data sets and models to create a practically usable model that will identify “referrable” retinal disorders that can be treated or are at the stage sufficiently progressed to start treatment as opposed to the “non-referrable” disorders with too early stage that doesn’t require treatment, or disorders having no known treatment methods. Important difference between the two is: diagnosing a “non-referrable” disorder will result in unnecessary visit to a retina specialist, while missing the “referrable” disorders might result in permanent blindness or vision loss. In this study, we explored the use of deep convolutional neural network methodology for the automatic classification of eye diseases using color fundus images. More than 10 retinal disorders have been effectively classified using the proposed model. The proposed method is tested using the public datasets and the EyeCheckup dataset we created. Our deep learning model achieved sensitivity of 0.9439, specificity of 0.8604, and an Accuracy of 0.86 with the test data set.

Keywords

• eye diseases
• fundus data set
• deep learning
• EfficientNet
• CNN

References

1. [1] World Health Organization, “World Report on Vision,” Oct. 2019. Accessed: May 05, 2020. [Online]. Available: https://www.who.int/publications/i/item/world-report-on-vision.
2. [2] A. Grzybowski et al., “Artificial intelligence for diabetic retinopathy screening: a review,” Eye, vol. 34, no. 3, pp. 451–460, Mar. 2020, doi: 10.1038/s41433-019-0566-0.
3. [3] N. Yalçin, S. Alver, and N. Uluhatun, “Classification of retinal images with deep learning for early detection of diabetic retinopathy disease,” in 2018 26th Signal Processing and Communications Applications Conference (SIU), May 2018, pp. 1–4, doi: 10.1109/SIU.2018.8404369.
4. [4] E. Şatir, F. Azboy, A. Aydin, H. Arslan, and Ş. Haciefendı̇oğlu, “Veri İndirgeme ve Sınıflandırma Teknikleri ile Glokom Hastalığı Teşhisi,” El-Cezeri J. Sci. Eng., vol. 3, no. 3, Art. no. 3, Sep. 2016, doi: 10.31202/ecjse.258576.
5. [5] B. Dizdaroğlu and B. Çorbacioğlu, “Deep Diagnosis of Non-Proliferative Diabetic Retinopathy in a Mobile System,” in 2019 Medical Technologies Congress (TIPTEKNO), Oct. 2019, pp. 1–4, doi: 10.1109/TIPTEKNO.2019.8894946.
6. [6] Ö. Deperlıoğlu and U. Köse, “Diagnosis of Diabetic Retinopathy by Using Image Processing and Convolutional Neural Network,” in 2018 2nd International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT), Oct. 2018, pp. 1–5, doi: 10.1109/ISMSIT.2018.8567055.
7. [7] Y. Yu et al., “Detecting abnormal fundus images by employing deep transfer learning,” In Review, preprint, Feb. 2020. doi: 10.21203/rs.2.24133/v1.
8. [8] S. Gayathri, A. K. Krishna, V. P. Gopi, and P. Palanisamy, “Automated Binary and Multiclass Classification of Diabetic Retinopathy Using Haralick and Multiresolution Features,” IEEE Access, vol. 8, pp. 57497–57504, 2020, doi: 10.1109/ACCESS.2020.2979753.

9. [9] V. Sathananthavathi, G. Indumathi, and R. Rajalakshmi, “Abnormalities detection in retinal fundus images,” in 2017 International Conference on Inventive Communication and Computational Technologies (ICICCT), Mar. 2017, pp. 89–93, doi: 10.1109/ICICCT.2017.7975165.
10. [10] A. Pak, A. Ziyaden, K. Tukeshev, A. Jaxylykova, and D. Abdullina, “Comparative analysis of deep learning methods of detection of diabetic retinopathy,” Cogent Eng., vol. 7, no. 1, p. 1805144, Jan. 2020, doi: 10.1080/23311916.2020.1805144.
11. [11] J. Wang, L. Yang, Z. Huo, W. He, and J. Luo, “Multi-Label Classification of Fundus Images With EfficientNet,” IEEE Access, vol. 8, pp. 212499–212508, 2020, doi: 10.1109/ACCESS.2020.3040275.
12. [12] B. Bulut, V. Kalın, B. B. Güneş, and R. Khazhin, “Deep Learning Approach For Detection Of Retinal Abnormalities Based On Color Fundus Images,” in 2020 Innovations in Intelligent Systems and Applications Conference (ASYU), Oct. 2020, pp. 1–6, doi: 10.1109/ASYU50717.2020.9259870.
13. [13] M. Chetoui and M. A. Akhloufi, “Explainable Diabetic Retinopathy using EfficientNET*,” in 2020 42nd Annual International Conference of the IEEE Engineering in Medicine Biology Society (EMBC), Jul. 2020, pp. 1966–1969, doi: 10.1109/EMBC44109.2020.9175664.
14. [14] EyePACS, “Diabetic Retinopathy Detection.” https://kaggle.com/c/diabetic-retinopathy-detection (accessed May 06, 2021).
15. [15] “Resized version of the Diabetic Retinopathy Kaggle competition dataset.” https://kaggle.com/tanlikesmath/diabetic-retinopathy-resized (accessed May 06, 2021).
16. [16] T. Kauppi et al., “DIARETDB0: Evaluation Database and Methodology for Diabetic Retinopathy Algorithms,” p. 17.
17. [17] P. Porwal, “Indian Diabetic Retinopathy Image Dataset (IDRiD).” IEEE, Apr. 24, 2018, Accessed:May 06, 2021. [Online]. Available: https://ieee-dataport.org/open-access/indian-diabeticretinopathy-image-dataset-idrid.
18. [18] E. Decencière et al., “Feedback on a publicly distributed database: the Messidor database,” Image Anal. Stereol., vol. 33, no. 3, pp. 231–234, Aug. 2014, doi: 10.5566/ias.1155.
19. [19] “APTOS 2019 Blindness Detection.” https://kaggle.com/c/aptos2019-blindness-detection (accessed May 06, 2021).
20. [20] “TFRecord ve tf.train.Example | TensorFlow Core,” TensorFlow. https://www.tensorflow.org/tutorials/load_data/tfrecord?hl=tr (accessed May 06, 2021).
21. [21] “Google Colab.” https://colab.research.google.com/notebooks/intro.ipynb#recent=true (accessedJul. 19, 2020).
22. [22] “scikit-optimize: sequential model-based optimization in Python — scikit-optimize 0.7.4 documentation.” https://scikit-optimize.github.io/stable/ (accessed Jul. 19, 2020).
23. [23] “TensorFlow,” TensorFlow. https://www.tensorflow.org/ (accessed Jul. 19, 2020).
24. [24] L. Liu et al., “On the Variance of the Adaptive Learning Rate and Beyond,” ArXiv190803265 CsStat, Apr. 2020, Accessed: Jul. 19, 2020. [Online]. Available: http://arxiv.org/abs/1908.03265.
25. [25] P. Ramachandran, B. Zoph, and Q. V. Le, “Swish: a Self-Gated Activation Function,” p. 12.
26. [26] W. QingJie and W. WenBin, “Research on image retrieval using deep convolutional neural network combining L1 regularization and PRelu activation function,” IOP Conf. Ser. Earth Environ. Sci., vol. 69, p. 012156, Jun. 2017, doi: 10.1088/1755-1315/69/1/012156