Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Jul 14;13(14):2375.
doi: 10.3390/diagnostics13142375.

Enhancement of Diabetic Retinopathy Prognostication Using Deep Learning, CLAHE, and ESRGAN

Ghadah Alwakid  1 , Walaa Gouda  2 , Mamoona Humayun  3
Affiliations

Enhancement of Diabetic Retinopathy Prognostication Using Deep Learning, CLAHE, and ESRGAN

Ghadah Alwakid et al. Diagnostics (Basel). .

Abstract

One of the primary causes of blindness in the diabetic population is diabetic retinopathy (DR). Many people could have their sight saved if only DR were detected and treated in time. Numerous Deep Learning (DL)-based methods have been presented to improve human analysis. Using a DL model with three scenarios, this research classified DR and its severity stages from fundus images using the "APTOS 2019 Blindness Detection" dataset. Following the adoption of the DL model, augmentation methods were implemented to generate a balanced dataset with consistent input parameters across all test scenarios. As a last step in the categorization process, the DenseNet-121 model was employed. Several methods, including Enhanced Super-resolution Generative Adversarial Networks (ESRGAN), Histogram Equalization (HIST), and Contrast Limited Adaptive HIST (CLAHE), have been used to enhance image quality in a variety of contexts. The suggested model detected the DR across all five APTOS 2019 grading process phases with the highest test accuracy of 98.36%, top-2 accuracy of 100%, and top-3 accuracy of 100%. Further evaluation criteria (precision, recall, and F1-score) for gauging the efficacy of the proposed model were established with the help of APTOS 2019. Furthermore, comparing CLAHE + ESRGAN against both state-of-the-art technology and other recommended methods, it was found that its use was more effective in DR classification.

Keywords: APTOS; Deep Learning; blindness; diabetic retinopathy; transfer learning.

PubMed Disclaimer

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Figure 1
Figure 1
Listed in order of increasing severity, the five stages of DR.
Figure 2
Figure 2
The process of DR classification.
Figure 3
Figure 3
Class-Wide Image Distribution of the APTOS dataset.
Figure 4
Figure 4
Various examples of the image-improvement methods that have been proposed (a) An unaltered version of the image; (b) a CLAHE version of the same image; and (c) an ESRGAN-enhanced version of the same image.
Figure 5
Figure 5
Some examples of the image-improvement methods that have been proposed. The four images shown here are: (a) the raw, unedited original; (b) the image after CLAHE; (c) the image utilizing HIST; and (d) the image after ESRGAN has been applied to it.
Figure 6
Figure 6
Some examples of image-improvement methods that have been proposed. The four images shown here are: (a) the raw, unedited original; (b) the image utilizing HIST; (c) the image after CLAHE; and (d) the image after ESRGAN has been applied to it.
Figure 7
Figure 7
Total number of training images after augmentation techniques have been employed.
Figure 8
Figure 8
Examples of augmenting the same image with different methods (CLAHE + ESRGAN).
Figure 9
Figure 9
Examples of augmenting the same image with different methods (CLAHE + HIST + ESRGAN).
Figure 10
Figure 10
Examples of augmenting the same image with different methods (HIST + CLAHE + ESRGAN).
Figure 11
Figure 11
Scenario I-specific workflow depiction of the DR detection system.
Figure 12
Figure 12
The finest DenseNet-121 confusion matrix with enhancement (CLAHE + ESRGAN).
Figure 13
Figure 13
Scenario II-specific workflow depiction of the DR detection system.
Figure 14
Figure 14
The finest DenseNet-121 confusion matrix with enhancement (CLAHE + HIST + ESRGAN).
Figure 15
Figure 15
Scenario III-specific workflow depiction of the DR detection system’.
Figure 16
Figure 16
The finest DenseNet-121 confusion with enhancement (HIST + CLAHE + ESRGAN).
Figure 17
Figure 17
Best results for the three scenarios.
Figure 18
Figure 18
ROC curve for the three scenarios.
Figure 19
Figure 19
Original and enhanced image samples.
Figure 20
Figure 20
Original and Enhanced Images + Histogram.
Figure 21
Figure 21
Superior Confusion Matrix for the EyePACS dataset.
Figure 22
Figure 22
Superior Confusion Matrix for the retrained APTOS model using the EyePACS dataset.
See this image and copyright information in PMC

References

    1. Association A.D. Diagnosis and classification of diabetes mellitus. Diabetes Care. 2014;37:S81–S90. doi: 10.2337/dc14-S081. - DOI - PubMed
    1. Al-Antary M.T., Arafa Y. Multi-scale attention network for diabetic retinopathy classification. IEEE Access. 2021;9:54190–54200. doi: 10.1109/ACCESS.2021.3070685. - DOI
    1. Hayati M., Muchtar K., Maulina N., Syamsuddin I., Elwirehardja G.N., Pardamean B. Impact of CLAHE-based image enhancement for diabetic retinopathy classification through deep learning. Procedia Comput. Sci. 2023;216:57–66. doi: 10.1016/j.procs.2022.12.111. - DOI
    1. Taylor R., Batey D. Handbook of Retinal Screening in Diabetes: Diagnosis and Management. John Wiley & Sons; Hoboken, NJ, USA: 2012.
    1. Atwany M.Z., Sahyoun A.H., Yaqub M. Deep learning techniques for diabetic retinopathy classification: A survey. IEEE Access. 2022;10:28642–28655. doi: 10.1109/ACCESS.2022.3157632. - DOI