Optic Disc Segmentation in Human Retina Images Using a Meta Heuristic Optimization Method and Disease Diagnosis with Deep Learning
Abstract
:1. Introduction
- i.
- We provide a metaheuristic approach for optic disc segmentation in fundus images that uses the Grey Wolf Optimization (GWO) algorithm, which has not previously been deployed in literature for this purpose. Thanks to this approach, we achieved a high performance of 99.39% accuracy in optical disc segmentation;
- ii.
- Vision transformers, including the Swin Transformer, show robustness to variations in image quality and noise, which are common challenges in medical imaging. Because of this, we used a new application of vision transformers, specifically the state-of-the-art Swin Transformer, for glaucoma image classification, which has received little attention in previous research. Thanks to this method, we achieved high classification performance;
- iii.
- We provide a thorough evaluation and comparison of two-stage and one-stage approaches for glaucoma classification. This evaluation provides valuable guidance for researchers working on glaucoma classification. These two approaches represent different strategies for glaucoma classification. The one-stage approach considers the entire image, potentially capturing broader contextual information but also including irrelevant features. On the other hand, the two-stage approach focuses specifically on the optic disc region, which is directly relevant to glaucoma diagnosis;
- iv.
- By effectively addressing the problem of class imbalance in ORIGA and Drishti-GS, our experimental results were able to achieve good results in glaucoma detection and classification.
2. Related Work
2.1. Optic Disc Localization
2.2. Classification of Glaucoma
3. Materials and Methods
3.1. Datasets
3.2. Methods
3.2.1. GWO Meta-Heuristic Optimization Algorithm
- Search, approach, and track the prey (exploitation);
- Pursuing, harassing, and encircling the prey until it stops moving;
- Hunting;
- Attacking the prey when it is exhausted (exploration).
3.2.2. YOLO
3.2.3. Vision Transformer
3.2.4. Performance Metrics
4. Experiments
4.1. Hyperparameter Selection
4.2. OD Localization Using GWO
4.3. YOLO Object Detection
4.4. Glaucoma Classification
5. Results and Discussion
5.1. Segmentation of Optic Disc Area Using GWO
5.2. Glaucoma Classification Results
5.2.1. Classification Results for DRISHTI-GS Dataset
5.2.2. Classification Result for ORIGA Dataset
5.3. Comparison and Discussion with Previous Studies
6. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
References
- Sivaprasad, S.; Gupta, B.; Crosby-Nwaobi, R.; Evans, J. Prevalence of diabetic retinopathy in various ethnic groups: A worldwide Perspective. Surv. Ophthalmol. 2012, 57, 347–370. [Google Scholar] [CrossRef] [PubMed]
- Pruthi, J.; Arora, S.; Khanna, K. Metaheuristic techniques for detection of optic disc in retinal fundus images. 3D Res. 2018, 9, 47. [Google Scholar] [CrossRef]
- Senjam, S.S. Glaucoma blindness—A rapidly emerging non-communicable ocular disease in India: Addressing the issue with advocacy. J. Fam. Med. Prim. Care 2020, 9, 2200–2206. [Google Scholar] [CrossRef]
- Yan, M.; Lin, Y.; Peng, X.; Zeng, Z. mixDA: Mixup domain adaptation for glaucoma detection on fundus images. Neural Comput. Appl. 2023, 32, 1–20. [Google Scholar] [CrossRef]
- Kumar, B.V.; Zhang, S.; Wu, T.; Prakash, J.; Zhou, L.; Li, K. A novel JAYA algorithm for optic disc localization in eye fundus images. Int. J. Comput. Vis. Robot. 2022, 12, 324–342. [Google Scholar] [CrossRef]
- Abed, S.; Al-Oraifan, D.; Safar, A. Optic disc detection using fish school search algorithm based on FPGA. J. Eng. 2019, 7, 161–177. [Google Scholar]
- Abed, S.; Al-Roomi, S.A.; Al-Shayeji, M. Effective optic disc detection method based on swarm intelligence techniques and novel pre-processing steps. Appl. Soft Comput. 2016, 49, 146–163. [Google Scholar] [CrossRef]
- Rahebi, J.; Hardalaç, F. A new approach to optic disc detection in human retinal images using the firefly algorithm. Med. Biol. Eng. Comput. 2016, 54, 453–461. [Google Scholar] [CrossRef]
- Abdullah, A.S.; Özok, Y.E.; Rahebi, J. A novel method for retinal optic disc detection using bat meta-heuristic algorithm. Med. Biol. Eng. Comput. 2018, 56, 2015–2024. [Google Scholar] [CrossRef]
- Gui, B.; Shuai, R.J.; Chen, P. Optic disc localization algorithm based on improved corner detection. Procedia Comput. Sci. 2018, 131, 311–319. [Google Scholar] [CrossRef]
- Thakur, N.; Juneja, M. Optic disc and optic cup segmentation from retinal images using hybrid approach. Expert Syst. Appl. 2019, 127, 308–322. [Google Scholar] [CrossRef]
- Shaikha, H.K.; Sallow, A.B. Optic Disc Detection and Segmentation in Retinal Fundus Image. In Proceedings of the 2019 International Conference on Advanced Science and Engineering (ICOASE), Zakho, Iraq, 2–4 April 2019; pp. 23–28. [Google Scholar]
- Abdullah, A.S.; Rahebi, J.; Özok, Y.E.; Aljanabi, M. A new and effective method for human retina optic disc segmentation with fuzzy clustering method based on active contour model. Med. Biol. Eng. Comput. 2019, 58, 25–37. [Google Scholar] [CrossRef]
- Sreng, S.; Maneerat, N.; Hamamoto, K.; Win, K.Y. Deep Learning for Optic Disc Segmentation and Glaucoma Diagnosis on Retinal Images. Appl. Sci. 2020, 10, 4916. [Google Scholar] [CrossRef]
- Zhen, Y.; Wang, L.; Liu, H.; Zhang, J.; Pu, J. Performance assessment of the deep learning technologies in grading glaucoma severity. arXiv 2018, arXiv:1810.13376. [Google Scholar]
- Yang, G.; Li, F.; Ding, D.; Wu, J.; Xu, J. Automatic diagnosis of glaucoma on color fundus images using adaptive mask deep network. In MultiMedia Modeling: MMM 2021, International Conference on Multimedia Modeling, Prague, Czech Republic, 22–24 June 2021; Springer: Cham, Switzerland, 2021; pp. 99–110. [Google Scholar]
- Elangovan, P.; Nath, M.K. Glaucoma assessment from color fundus images using convolutional neural network. Int. J. Imaging Syst. Technol. 2021, 31, 955–971. [Google Scholar] [CrossRef]
- Gómez-Valverde, J.J.; Antón, A.; Fatti, G.; Liefers, B.; Herranz, A.; Santos, A.; Sánchez, C.I.; Ledesma-Carbayo, M.J. Automatic glaucoma classification using color fundus images based on convolutional neural networks and transfer learning. Br. J. Ophthalmol. 2019, 10, 892–913. [Google Scholar] [CrossRef] [PubMed]
- Hemelings, R.; Elen, B.; Barbosa-Breda, J.; Lemmens, S.; Meire, M.; Pourjavan, S.; Stalmans, I. Accurate prediction of glaucoma from colour fundus images with a convolutional neural network that relies on active and transfer learning. Acta Ophthalmol. 2020, 98, 94–100. [Google Scholar] [CrossRef]
- Natarajan, D.; Sankaralingam, E.; Balraj, K.; Thangaraj, V. Automated Segmentation Algorithm with Deep Learning Framework for Early Detection of Glaucoma. Concurr. Comput. Pract. Exp. 2021, 33, e6181. [Google Scholar] [CrossRef]
- Li, L.; Xu, M.; Wang, X.; Jiang, L.; Liu, H. Attention Based Glaucoma Detection: A Large-Scale Database and CNN Model. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, CA, USA, 15–20 June 2019; pp. 10563–10572. [Google Scholar]
- Bajwa, M.N.; Malik, M.I.; Siddiqui, S.A.; Dengel, A.; Shafait, F.; Neumeier, W.; Ahmed, S. Two-stage framework for optic disc localization and glaucoma classification in retinal fundus images using deep learning. BMC Med. Inform. Decis. Mak. 2019, 19, 136. [Google Scholar]
- Sonti, K.; Dhuli, R. A new convolution neural network model “KR-NET” for retinal fundus glaucoma classification. Optik 2023, 283, 170861. [Google Scholar] [CrossRef]
- Al-Bander, B.; Al-Nuaimy, W.; Al-Taee, M.A.; Zheng, Y. Automated glaucoma diagnosis using deep learning approach. In Proceedings of the 2017 14th International Multi-Conference on Systems, Signals & Devices (SSD), Marrakech, Morocco, 28–31 March 2017; pp. 207–210. [Google Scholar]
- Abbas, Q. Glaucoma-Deep: Detection of Glaucoma Eye Disease on Retinal Fundus Images Using Deep Learning. Int. J. Adv. Comput. Sci. Appl. 2017, 8, 41–45. [Google Scholar] [CrossRef]
- Khan, S.; Naseer, M.; Hayat, M.; Zamir, S.W.; Khan, F.S.; Shah, M. Transformers in Vision: A Survey. ACM Comput. Surv. 2021, 54, 1–41. [Google Scholar] [CrossRef]
- Wassel, M.; Hamdi, A.M.; Adly, N.; Torki, M. Vision Transformers Based Classification for Glaucomatous Eye Condition. In Proceedings of the 2022 26th International Conference on Pattern Recognition (ICPR), Montreal, QC, Canada, 21–25 August 2022; pp. 5082–5088. [Google Scholar]
- Hu, X.; Zhang, L.-X.; Gao, L.; Dai, W.; Han, X.; Lai, Y.-K.; Chen, Y. GLIM-Net: Chronic Glaucoma Forecast Transformer for Irregularly Sampled Sequential Fundus Images. IEEE Trans. Med. Imaging 2023, 42, 1875–1884. [Google Scholar] [CrossRef] [PubMed]
- Zhang, Z.; Yin, F.S.; Liu, J.; Wong, W.K.; Tan, N.M.; Lee, B.H.; Cheng, J.; Wong, T.Y. Origa-light: An online retinal fundus image database for glaucoma analysis and research. In Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Buenos Aires, Argentina, 1–4 September 2010; pp. 3065–3068. [Google Scholar]
- Sivaswamy, J.; Krishnadas, S.; Joshi, G.D.; Jain, M.; Tabish, A.U.S. Drishti-gs: Retinal image dataset for optic nerve head (onh) segmentation. In Proceedings of the IEEE International Symposium on Biomedical Imaging, Beijing, China, 29 April–2 May 2014; pp. 53–56. [Google Scholar]
- Komaki, G.M.; Kayvanfar, V. Grey wolf optimizer algorithm for the two-stage assembly flow shop scheduling problem with release time. J. Comput. Sci. 2015, 8, 109–120. [Google Scholar] [CrossRef]
- Redmon, J.; Divvala, S.; Girshick, R.; Farhadi, A. You only look once: Unified, real-time object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 27–30 June 2016; pp. 779–788. [Google Scholar]
- Wu, S.; Yang, J.; Yu, H.; Gou, L.; Li, X. Gaussian Guided IoU: A Better Metric for Balanced Learning on Object Detection. arXiv 2021, arXiv:2103.13613. [Google Scholar]
- Singh, P.B.; Singh, P.; Dev, H.; Tiwari, A.; Batra, D.; Chaurasia, B.K. Glaucoma Classification using Light Vision Transformer. In Proceedings of the International Conference on Intelligent Systems and Machine Learning (ICISML), Odisha, India, 27–28 July 2023; pp. 1–12. [Google Scholar]
- Lin, T.; Wang, Y.; Liu, X.; Qiu, X. A survey of transformers. AI Open 2022, 3, 111–132. [Google Scholar] [CrossRef]
- An, W.; Wang, H.; Sun, Q.; Xu, J.; Dai, Q.; Zhang, L. A PID controller approach for stochastic optimization of deep networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, 18–22 June 2018; pp. 8522–8531. [Google Scholar]
- Maas, A.L.; Qi, P.; Xie, Z.; Hannun, A.Y.; Lengerich, C.T.; Jurafsky, D.; Ng, A.Y. Building dnn acoustic models for large vocabulary speech recognition. Comput. Speech Lang. 2017, 41, 195–213. [Google Scholar] [CrossRef]
- Gao, J.; Jiang, Y.; Zhang, H.; Wang, F. Joint disc and cup segmentation based on recurrent fully convolutional network. PLoS ONE 2020, 15, e0238983. [Google Scholar] [CrossRef] [PubMed]
- Samawi, H.J.; Al-Sultan, A.Y.; Al-Saadi, E.H. Optic disc segmentation in retinal fundus images using morphological techniques and intensity thresholding. In Proceedings of the 2020 International Conference on Computer Science and Software Engineering (CSASE), Duhok, Iraq, 16–18 April 2020; pp. 302–307. [Google Scholar]
- Tadisetty, S.; Chodavarapu, R.; Jin, R.; Clements, R.J.; Yu, M. Identifying the Edges of the Optic Cup and the Optic Disc in Glaucoma Patients by Segmentation. Sensors 2023, 23, 4668. [Google Scholar] [CrossRef]
- Sevastopolsky, A. Optic disc and cup segmentation methods for glaucoma detection with modification of U-Net convolutional neural network. Pattern Recognit. Image Anal. 2017, 27, 618–624. [Google Scholar] [CrossRef]
- Al-Bander, B.; Williams, B.M.; Al-Nuaimy, W.; Al-Taee, M.A.; Pratt, H.; Zheng, Y. Dense Fully Convolutional Segmentation of the Optic Disc and Cup in Colour Fundus for Glaucoma Diagnosis. Symmetry 2018, 10, 87. [Google Scholar] [CrossRef]
- Ramani, R.G.; Shanthamalar, J.J. Improved image processing techniques for optic disc segmentation in retinal fundus images. Biomed. Signal Process. Control 2020, 58, 101832. [Google Scholar] [CrossRef]
- Elangovan, P.; Nath, M.K. Performance analysis of optimizers for glaucoma diagnosis from fundus images using transfer learning. In Machine Learning, Deep Learning and Computational Intelligence for Wireless Communication: Proceedings of MDCWC; Springer: Singapore, 2021; pp. 507–518. [Google Scholar]
- Diaz-Pinto, A.; Morales, S.; Naranjo, V.; Köhler, T.; Mossi, J.M.; Navea, A. CNNs for automatic glaucoma assessment using fundus images: An extensive validation. BioMed. Eng. Online 2019, 18, 29. [Google Scholar] [CrossRef] [PubMed]
- Guo, F.; Mai, Y.; Zhao, X.; Duan, X.; Fan, Z.; Zou, B.; Xie, B. Yanbao: A mobile app using the measurement of clinical parameters for glaucoma screening. IEEE Access 2018, 6, 77414–77428. [Google Scholar] [CrossRef]
- Chen, X.-W.; Lin, X. Big data deep learning: Challenges and perspectives. IEEE Access 2014, 2, 514–525. [Google Scholar] [CrossRef]
- Alagirisamy, M. Micro statistical descriptors for glaucoma diagnosis using neural networks. Int. J. Adv. Signal Image Sci. 2021, 7, 1–10. [Google Scholar] [CrossRef]
- Pranathi, K.; Pingili, M.; Mamatha, B. Fundus Image Processing for Glaucoma Diagnosis Using Dynamic Support Vector Machine. In Advances in Cognitive Science and Communications, ICCCE 2023: Proceedings of the International Conference on Communications and Cyber Physical Engineering, Hyderabad, India, 29–30 April 2022; Springer: Singapore, 2018; pp. 551–558. [Google Scholar]
- Chaudhary, P.K.; Jain, S.; Damani, T.; Gokharu, S.; Pachori, R.B. Detection of Primary and Secondary Glaucoma Using 2D-FBSE-EWT From Different Fundus Image Modalities. Authorea Prepr. 2023, 14, 1–10. [Google Scholar]
Dataset | No of Images | Segmentation/Classification | Normal | Glaucoma | Input Size | Label Ground Truth | Source |
---|---|---|---|---|---|---|---|
DRISHTI-GS | 101 | Both | 31 | 70 | 2896 × 1944 | Labeled | Aravind Eye Hospital in Madurai (India) |
ORIGA | 650 | Classification | 482 | 168 | 3072 × 2048 | - | Singapore Eye Research Institute |
Model | Optimization Algorithm | Learning Rate | Batch Size | Epoch | Activation Function | Base Model Trainable | Loss Function |
---|---|---|---|---|---|---|---|
SwinTransformer | Rmsprop | 0.001 | 32 | 150 | |||
DenseNet201 | Rmsprop | 0.001 | 32 | 150 | Relu Softmax | False | Categorical Cross-entropy |
Resnet50 | Adam | 0.001 | 32 | 150 | |||
InceptionV3 | Sgd | 0.01 | 32 | 150 | |||
VGG19 | Adamax | 0.001 | 32 | 150 |
Training/Testing | [email protected] (%) | P | R |
---|---|---|---|
Training | 99.53 | 1 | 1 |
Test | 99.50 | 1 | 1 |
Model | Accuracy (%) | Sensitivity (%) | Specificity (%) | F1_Score (%) | Precision (%) |
---|---|---|---|---|---|
Cropped Dataset (two-stage approach) | |||||
DenseNet201 | 90.48 | 93.33 | 83.33 | 90.48 | 90.48 |
Vgg19 | 90.48 | 93.33 | 83.33 | 90.48 | 90.48 |
ResNet50 | 90.48 | 100 | 66.67 | 89.82 | 91.60 |
Swin Transformer | 76.19 | 100 | 16.67 | 69.39 | 82.14 |
Inception V3 | 80.95 | 93.33 | 50.00 | 79.64 | 80.25 |
Uncropped Dataset (one-stage approach) | |||||
Swin Transformer | 100 | 100 | 100 | 100 | 100 |
ResNet50 | 85 | 80 | 86.7 | 81.87 | 83.33 |
Inception V3 | 80 | 75 | 81.2 | 74.62 | 78.125 |
DenseNet201 | 75 | 57 | 84.6 | 70.88 | 72.62 |
Vgg19 | 75 | 66.7 | 76.5 | 71.56 | 63.09 |
Models | Fold | TP | TN | FP | FN | R | S | P | F1 | Acc |
---|---|---|---|---|---|---|---|---|---|---|
DenseNet-201 | Fold 1 | 21 | 8 | 2 | 3 | 87.50 | 80.00 | 85.84 | 85.49 | 85.29 |
Fold 2 | 21 | 5 | 6 | 2 | 91.0 | 45.45 | 75.72 | 84.00 | 76.47 | |
Fold 3 | 23 | 4 | 6 | 0 | 100 | 40.00 | 85.58 | 78.97 | 81.82 | |
Overlapped | ||||||||||
Glaucoma | 65 | 17 | 14 | 5 | 92.86 | 54.84 | 82.28 | 87.25 | ||
Normal | 17 | 65 | 5 | 14 | 54.84 | 92.86 | 77.27 | 64.15 | ||
Average | 73.85 | 73.85 | 79.78 | 75.7 | 81.19 | |||||
RESNET50 | Overlapped | |||||||||
Glaucoma | 66 | 16 | 15 | 4 | 94.29 | 51.61 | 81.48 | 87.42 | ||
Normal | 16 | 66 | 4 | 15 | 51.61 | 94.29 | 62.74 | 62.75 | ||
Average | 72.95 | 72.95 | 72.11 | 75.09 | 81.19 | |||||
Inception V3 | Overlapped | |||||||||
Glaucoma | 63 | 16 | 15 | 7 | 90.00 | 51.61 | 80.77 | 85.14 | ||
Normal | 16 | 63 | 7 | 15 | 51.61 | 90.00 | 69.57 | 59.26 | ||
Average | 70.81 | 70.81 | 75.17 | 72.20 | 78.28 | |||||
VGG19 | Overlapped | |||||||||
Glaucoma | 64 | 13 | 18 | 6 | 91.43 | 41.94 | 78.05 | 84.21 | ||
Normal | 13 | 64 | 6 | 18 | 41.94 | 91.43 | 68.42 | 52 | ||
Average | 66.69 | 66.69 | 73.24 | 68.11 | 76.20 | |||||
Swin Transformer | Overlapped | |||||||||
Glaucoma | 66 | 10 | 21 | 4 | 94.29 | 32.26 | 75.87 | 84.08 | ||
Normal | 10 | 66 | 4 | 21 | 32.26 | 94.29 | 71.43 | 44.44 | ||
Average | 63.28 | 63,28 | 73.65 | 64.26 | 75.31 |
Models | Fold | TP | TN | FP | FN | R | S | P | F1 | Acc |
---|---|---|---|---|---|---|---|---|---|---|
Swin Transformer | Fold 1 | 21 | 6 | 5 | 2 | 91.30 | 54.55 | 78.90 | 78.42 | 79.41 |
Fold 2 | 23 | 8 | 2 | 1 | 95.83 | 80 | 91.08 | 91.03 | 91.18 | |
Fold 3 | 22 | 9 | 1 | 1 | 95.65 | 90 | 93.94 | 93.94 | 93.94 | |
Overlapped | ||||||||||
Glaucoma | 66 | 23 | 8 | 4 | 94.29 | 74.19 | 89.19 | 91.67 | ||
Normal | 23 | 66 | 4 | 8 | 74.19 | 94.29 | 85.19 | 79.31 | ||
Average | 84.24 | 84.24 | 87.19 | 85.49 | 88.18 | |||||
Inception V3 | Overlapped | |||||||||
Glaucoma | 65 | 19 | 12 | 5 | 92.86 | 61.29 | 84.42 | 88.44 | ||
Normal | 19 | 65 | 5 | 12 | 61.29 | 92.86 | 79.17 | 69.09 | ||
Average | 77.08 | 77.08 | 81.80 | 78.77 | 83.18 | |||||
DenseNet-201 | Overlapped | |||||||||
Glaucoma | 67 | 14 | 17 | 3 | 95.71 | 45.16 | 79.76 | 87.01 | ||
Normal | 14 | 67 | 3 | 17 | 45.16 | 95.71 | 82.35 | 58.33 | ||
Average | 70.44 | 70.44 | 81.06 | 72.67 | 80.21 | |||||
RESNET50 | Overlapped | |||||||||
Glaucoma | 62 | 18 | 13 | 8 | 88.57 | 58.06 | 82.67 | 85.52 | ||
Normal | 18 | 62 | 8 | 13 | 58.06 | 88.57 | 69.23 | 63.16 | ||
Average | 73.32 | 73.32 | 75.95 | 74.34 | 79.26 | |||||
VGG19 | Overlapped | |||||||||
Glaucoma | 59 | 19 | 12 | 11 | 84.29 | 61.29 | 83.10 | 83.69 | ||
Normal | 19 | 59 | 11 | 12 | 61.29 | 84.29 | 63.33 | 62.30 | ||
Average | 72.79 | 72.79 | 73.22 | 72.99 | 77.21 |
Model | Accuracy (%) | Sensitivity (%) | Specificity (%) | F1_Score (%) | Precision (%) |
---|---|---|---|---|---|
Cropped Dataset (two-stage approach) | |||||
DenseNet201+RF | 95.24 | 100 | 83.33 | 95.10 | 95.54 |
Resnet50+RF | 90.48 | 100 | 66.67 | 89.82 | 91.60 |
Vgg19+RF | 90.48 | 100 | 66.67 | 89.82 | 91.60 |
InceptionV3+RF | 85.71 | 93.33 | 66.67 | 85.30 | 85.36 |
Uncropped Dataset (one-stage approach) | |||||
DenseNet201+RF | 90.48 | 100 | 66.67 | 89.82 | 91.60 |
Vgg19+RF | 90.48 | 100 | 66.67 | 89.82 | 91.60 |
Resnet50+RF | 85.71 | 100 | 50.00 | 83.98 | 88.10 |
InceptionV3+RF | 85.71 | 100 | 50.00 | 83.98 | 88.10 |
Models | Fold | TP | TN | FP | FN | R | S | P | F1 | Acc |
---|---|---|---|---|---|---|---|---|---|---|
DenseNet201+RF | Fold 1 | 24 | 5 | 5 | 0 | 100 | 50 | 87.83 | 83.54 | 85.29 |
Fold 2 | 23 | 2 | 9 | 0 | 100 | 18.18 | 80.97 | 66.53 | 73.53 | |
Fold 3 | 23 | 5 | 5 | 0 | 100 | 50 | 87.55 | 83.07 | 84.85 | |
Overlapped | ||||||||||
Glaucoma | 70 | 12 | 19 | 0 | 100 | 38.70 | 78.65 | 88.05 | ||
Normal | 12 | 70 | 0 | 19 | 38.70 | 100 | 100 | 55.80 | ||
Average | 69.35 | 69.35 | 89.33 | 71.93 | 81.22 | |||||
RESNET50+RF | Overlapped | |||||||||
Glaucoma | 70 | 12 | 19 | 0 | 100 | 38.70 | 78.65 | 88.05 | ||
Normal | 12 | 70 | 0 | 19 | 38.70 | 100 | 38.71 | 55.81 | ||
Average | 69.35 | 69.35 | 58.68 | 71.93 | 81.19 | |||||
Inception V3+RF | Overlapped | |||||||||
Glaucoma | 70 | 10 | 21 | 0 | 100 | 32.26 | 76.92 | 86.96 | ||
Normal | 10 | 70 | 0 | 21 | 32.26 | 100 | 76.92 | 48.78 | ||
Average | 66.13 | 66.13 | 76.92 | 67.87 | 79.26 | |||||
VGG19+RF | Overlapped | |||||||||
Glaucoma | 68 | 11 | 20 | 2 | 97.14 | 35.48 | 77.27 | 86.08 | ||
Normal | 11 | 68 | 2 | 20 | 35.48 | 97.14 | 84.62 | 50 | ||
Average | 66.31 | 66.31 | 80.95 | 68.04 | 78.28 |
Models | Fold | TP | TN | FP | FN | R | S | P | F1 | Acc |
---|---|---|---|---|---|---|---|---|---|---|
VGG19+RF | Fold 1 | 23 | 6 | 5 | 0 | 100 | 54.55 | 87.92 | 83.85 | 85.29 |
Fold 2 | 24 | 6 | 4 | 0 | 100 | 60 | 89.92 | 87.22 | 88.24 | |
Fold 3 | 23 | 6 | 4 | 0 | 100 | 60 | 89.67 | 86.85 | 87.88 | |
Overlapped | ||||||||||
Glaucoma | 70 | 18 | 13 | 0 | 100 | 58.06 | 84.34 | 91.50 | ||
Normal | 18 | 70 | 0 | 13 | 58.06 | 100 | 100 | 73.47 | ||
Average | 79.03 | 79.03 | 92.17 | 82.485 | 87.14 | |||||
DenseNet201+RF | Overlapped | |||||||||
Glaucoma | 70 | 10 | 21 | 0 | 100 | 32.26 | 76.92 | 86.96 | ||
Normal | 10 | 70 | 0 | 21 | 32.26 | 100 | 76.92 | 48.78 | ||
Average | 77.42 | 77.42 | 91.67 | 80.87 | 86.16 | |||||
Inception V3+RF | Overlapped | |||||||||
Glaucoma | 69 | 17 | 14 | 1 | 98.57 | 54.84 | 83.13 | 90.20 | ||
Normal | 17 | 69 | 1 | 14 | 54.84 | 98.57 | 94.44 | 69.39 | ||
Average | 76.71 | 76.71 | 88.79 | 79.80 | 85.12 | |||||
RESNET50+RF | Overlapped | |||||||||
Glaucoma | 67 | 13 | 18 | 3 | 95.71 | 41.94 | 78.82 | 86.45 | ||
Normal | 13 | 67 | 3 | 18 | 41.94 | 95.71 | 81.25 | 55.32 | ||
Average | 68.83 | 68.83 | 80.04 | 70.89 | 79.20 |
Pre-Trained CNN Model | Accuracy (%) | Recall (%) | Specificity (%) | F1_Score (%) | Precision (%) |
---|---|---|---|---|---|
cropped (two-stage approach) | |||||
DenseNet201 | 69.33 | 72.44 | 52.17 | 80 | 89.32 |
ResNet50 | 73.85 | 96 | 73.85 | 62.74 | 54.53 |
Vgg19 | 75.38 | 75.38 | N | 84.96 | 100 |
Inception V3 | 74.62 | 76.03 | 55.56 | 84.79 | 95.83 |
Swin Transformer | 96.15 | 95.05 | 100 | 97.46 | 100 |
uncropped (one-stage approach) | |||||
DenseNet201 | 78.15 | 79.17 | 74.19 | 85.20 | 92.23 |
ResNet50 | 73.85 | 96 | 73.85 | 62.74 | 54.53 |
Vgg19 | 72.31 | 96.91 | 0 | 83.93 | 74.02 |
Inception V3 | 69.23 | 100 | 0 | 81.82 | 69.23 |
Swin Transformer | 86.15 | 88.24 | 78.57 | 90.91 | 93.75 |
Models | Fold | TP | TN | FP | FN | R | S | P | F1 | Acc |
---|---|---|---|---|---|---|---|---|---|---|
Inception V3 | Fold 1 | 16 | 87 | 10 | 17 | 48.48 | 89.69 | 78.04 | 78.36 | 79.23 |
Fold 2 | 23 | 94 | 3 | 10 | 69.70 | 96.91 | 89.90 | 89.58 | 90.00 | |
Fold3 | 33 | 96 | 0 | 1 | 97.06 | 100 | 99.24 | 99.23 | 99.23 | |
Fold 4 | 34 | 95 | 1 | 0 | 100 | 98.96 | 99.25 | 99.23 | 99.23 | |
Fold 5 | 34 | 96 | 0 | 0 | 100 | 100 | 100 | 100 | 100 | |
Overlapped | ||||||||||
Glaucoma | 140 | 468 | 14 | 28 | 83.33 | 97.10 | 90.91 | 86.96 | ||
Normal | 468 | 140 | 28 | 14 | 97.10 | 83.33 | 94.35 | 95.71 | ||
Average | 90.21 | 90.21 | 92.63 | 91.33 | 93.53 | |||||
VGG19 | Overlapped | |||||||||
Glaucoma | 135 | 470 | 12 | 13 | 91.22 | 97.51 | 91.84 | 91.53 | 93.07 | |
Normal | 470 | 135 | 13 | 12 | 97.51 | 91.22 | 97.31 | 97.41 | ||
Average | 94.36 | 94.36 | 94.57 | 94.47 | ||||||
Swin Transformer | Overlapped | |||||||||
Glaucoma | 100 | 414 | 68 | 68 | 59.52 | 85.89 | 59.52 | 59.52 | ||
Normal | 414 | 100 | 68 | 68 | 85.89 | 59.52 | 85.89 | 85.89 | ||
Average | 72.71 | 72.71 | 72.71 | 72.71 | 79.07 | |||||
DenseNet-201 | Overlapped | |||||||||
Glaucoma | 127 | 368 | 114 | 41 | 75.60 | 76.35 | 52.70 | 62.10 | ||
Normal | 368 | 127 | 41 | 114 | 76.35 | 75.60 | 89.98 | 82.60 | ||
75.97 | 75.97 | 71.34 | 72.35 | 76.15 | ||||||
RESNET50 | Overlapped | |||||||||
Glaucoma | 0 | 482 | 0 | 168 | 0 | 100 | N | N | ||
Normal | 482 | 0 | 168 | 0 | 100 | 0 | 74.15 | 85.16 | ||
Average | 50 | 50 | N | N | 74.15 |
Models | Fold | TP | TN | FP | FN | R | S | P | F1 | Acc |
---|---|---|---|---|---|---|---|---|---|---|
Inception V3 | Fold 1 | 14 | 86 | 11 | 19 | 42.42 | 88.66 | 75.33 | 75.79 | 76.92 |
Fold 2 | 33 | 97 | 0 | 0 | 100 | 100 | 100 | 100 | 100 | |
Fold3 | 30 | 95 | 1 | 4 | 88.24 | 98.96 | 96.17 | 96.09 | 96.15 | |
Fold 4 | 31 | 94 | 2 | 3 | 91.18 | 97.92 | 96.13 | 96.14 | 96.15 | |
Fold 5 | 34 | 96 | 0 | 0 | 100 | 100 | 100 | 100 | 100 | |
Overlapped | ||||||||||
Glaucoma | 142 | 468 | 14 | 26 | 84.52 | 97.10 | 91.03 | 87.65 | ||
Normal | 468 | 142 | 26 | 14 | 97.10 | 84.52 | 94.74 | 95.90 | ||
Average | 90.81 | 90.81 | 92.88 | 91.77 | 93.84 | |||||
VGG19 | Overlapped | |||||||||
Glaucoma | 136 | 463 | 19 | 32 | 80.95 | 96.06 | 87.82 | 84.21 | ||
Normal | 463 | 136 | 32 | 19 | 96.06 | 80.59 | 93.54 | 94.78 | ||
Average | 88.50 | 88.32 | 90.68 | 89.49 | 92.15 | |||||
Swin transform | Overlapped | |||||||||
Glaucoma | 127 | 442 | 40 | 41 | 75.60 | 91.70 | 76.05 | 75.82 | ||
Normal | 442 | 127 | 41 | 40 | 91.70 | 75.60 | 91.51 | 91.61 | ||
Average | 83.65 | 83.65 | 83.78 | 83.71 | 87.54 | |||||
DenseNet-201 | Overlapped | |||||||||
Glaucoma | 114 | 448 | 34 | 54 | 67.86 | 92.95 | 77.03 | 72.15 | ||
Normal | 448 | 114 | 54 | 34 | 92.95 | 67.86 | 89.24 | 91.06 | ||
Average | 80.40 | 80.40 | 83.13 | 81.60 | 86.46 | |||||
RESNET50 | Overlapped | |||||||||
Glaucoma | 0 | 482 | 0 | 168 | 0 | 100 | N | N | ||
Normal | 482 | 0 | 168 | 0 | 100 | 0 | 74.15 | 85.16 | ||
Average | 50 | 50 | N | N | 74.15 |
Model | Accuracy (%) | Sensitivity (%) | Specificity (%) | F1_Score (%) | Precision (%) |
---|---|---|---|---|---|
Cropped Dataset with YOLO (two stage) | |||||
DenseNet201+RF | 77.69 | 78.15 | 72.73 | 86.51 | 96.88 |
Resnet50+RF | 73.85 | 73.85 | NaN | 84.96 | 100 |
Vgg19+RF | 76.92 | 77.50 | 70 | 86.11 | 96.88 |
InceptionV3+RF | 76.15 | 76.00 | 80 | 85.97 | 98.96 |
Uncropped Dataset (one stage) | |||||
DenseNet201+RF | 78.46 | 77.87 | 87.50 | 87.16 | 98.96 |
Resnet50+RF | 73.85 | 73.85 | NaN | 84.96 | 100 |
Vgg19+RF | 74.62 | 75.20 | 60 | 85.07 | 97.92 |
InceptionV3+RF | 76.92 | 77.05 | 75 | 86.24 | 97.92 |
Models | Fold | TP | TN | FP | FN | R | S | P | F1 | Acc |
---|---|---|---|---|---|---|---|---|---|---|
DenseNet201+RF | Fold1 | 9 | 96 | 1 | 24 | 27.27 | 98.97 | 82.54 | 76.65 | 80.77 |
Fold2 | 5 | 93 | 4 | 28 | 15.15 | 95.88 | 71.45 | 69.71 | 75.38 | |
Fold 3 | 10 | 90 | 6 | 24 | 29.41 | 93.75 | 74.65 | 73.76 | 76.92 | |
Fold 4 | 5 | 93 | 3 | 29 | 14.71 | 96.88 | 72.64 | 69.23 | 75.38 | |
Fold 5 | 9 | 90 | 6 | 25 | 26.47 | 93.75 | 73.48 | 72.60 | 76.15 | |
Overlapped | ||||||||||
Glaucoma | 38 | 462 | 20 | 130 | 22.62 | 95.85 | 65.52 | 33.63 | ||
Normal | 462 | 38 | 130 | 20 | 95.85 | 22.62 | 78.04 | 86.03 | ||
Average | 59.23 | 59.23 | 71.78 | 59.83 | 76.92 | |||||
VGG19+RF | Overlapped | |||||||||
Glaucoma | 23 | 472 | 10 | 145 | 13.69 | 97.93 | 69.69 | 22.89 | ||
Normal | 472 | 23 | 145 | 10 | 97.93 | 13.69 | 76.49 | 85.89 | ||
Average | 55.81 | 55.81 | 73.09 | 54.39 | 76.15 | |||||
Resnet50+RF | Overlapped | |||||||||
Glaucoma | 0 | 482 | 0 | 168 | 0 | 100 | N | N | ||
Normal | 482 | 0 | 168 | 0 | 100 | 0 | 74.15 | 85.15 | ||
Average | 50 | 50 | N | N | 74.15 | |||||
Inception V3+RF | Overlapped | |||||||||
Glaucoma | 10 | 471 | 11 | 158 | 0.05 | 97.72 | 47.61 | 10.58 | ||
Normal | 471 | 10 | 158 | 11 | 97.72 | 0.05 | 74.88 | 84.78 | ||
Average | 48.88 | 48.88 | 61.24 | 47.68 | 74 |
Models | Fold | TP | TN | FP | FN | R | S | P | F1 | Acc |
---|---|---|---|---|---|---|---|---|---|---|
Inception V3+RF | Fold 1 | 6 | 96 | 1 | 27 | 18.18 | 98.97 | 79.99 | 72.73 | 78.46 |
Fold 2 | 7 | 95 | 2 | 26 | 21.21 | 97.94 | 78.33 | 73.49 | 78.46 | |
Fold 3 | 4 | 89 | 7 | 30 | 11.76 | 92.71 | 64.74 | 65.79 | 71.54 | |
Fold4 | 2 | 91 | 5 | 32 | 0.05 | 94.79 | 62.11 | 63.92 | 71.54 | |
Fold 5 | 5 | 91 | 5 | 29 | 14.71 | 94.79 | 69.08 | 68.17 | 73.85 | |
Overlapped | ||||||||||
Glaucoma | 24 | 462 | 20 | 144 | 14.29 | 95.85 | 54.55 | 22.64 | ||
Normal | 462 | 24 | 144 | 20 | 95.85 | 14.29 | 76.24 | 84.93 | ||
Average | 55.07 | 55.07 | 65.39 | 53.78 | 74.76 | |||||
VGG19+RF | Overlapped | |||||||||
Glaucoma | 23 | 462 | 20 | 145 | 13.69 | 95.85 | 53.48 | 21.8 | ||
Normal | 462 | 23 | 145 | 20 | 95.85 | 13.69 | 76.11 | 84.85 | ||
Average | 54.77 | 54.77 | 64.79 | 53.33 | 74.61 | |||||
Resnet50+RF | Overlapped | |||||||||
Glaucoma | 0 | 482 | 0 | 168 | 0 | 100 | N | N | ||
Normal | 482 | 0 | 168 | 0 | 100 | 0 | 74.15 | 85.15 | ||
Average | 50 | 50 | N | N | 74.15 | |||||
DenseNet201+RF | Overlapped | |||||||||
Glaucoma | 30 | 451 | 31 | 138 | 17.86 | 93.57 | 49.18 | 26.20 | ||
Normal | 451 | 30 | 138 | 31 | 93.57 | 17 | 76.57 | 84.22 | ||
Average | 55.71 | 55.28 | 62.87 | 55.21 | 74 |
Studies | R | S | Acc | DICE | JACCARD |
---|---|---|---|---|---|
Gao et al. [38] | 95.78 | 97.83 | 97.64 | - | - |
Samawi [39] | - | - | 97.3 | - | - |
Tadisetty [40] | - | - | - | 94.3 | 89.3 |
Sevastopolsky [41] | 85 | ||||
AL-Bander et al. [42] | - | - | 99.69 | 94.9 | 90.4 |
Ramani et al. [43] | 95.28 | 99.43 | 99.31 | 88.43 | - |
Our Study | 96.04 | 99.58 | 99.39 | 94.15 | 90.40 |
Author(s) | Methods | Acc | R | S | P | F1 | Dataset | Split | Cropped ROI |
---|---|---|---|---|---|---|---|---|---|
Elangovanetal et al. [17] | CNN with 18 layers | 86.62 | 92.31 | 48.15 | 92.38 | - | DRISHTI-GS | Hold-out | Not Applied |
78.32 | 58.06 | 92.44 | 84.36 | - | ORIGA | ||||
Elangovanetal [44] | VGG-19 | 91.50 | 98.09 | 49.17 | 92.87 | - | DRISHTI-GS | Hold-out | Not Applied |
ResNet-101 | 80.50 | 68.60 | 88.80 | 81.20 | - | ORIGA | |||
Sreng [15] | ShuffleNet(P1) | 86.67 | - | - | - | - | DRISHTI-GS | Hold-out | Segmentation using DeepLabv3+ MDCNN |
MobileNet(P1) | 81.54 | - | - | - | - | ORIGA | |||
Inception +SVM(P2) | 91.53 | - | - | - | - | DRISHTI-GS | |||
ResNet+SVM(P2) | 78.97 | - | - | - | - | ORIGA | |||
Ensemble of P1 | 85.19 | - | - | - | - | DRISHTI-GS | |||
88.86 | - | - | - | - | ORIGA | ||||
Ensemble of P2 | 92.06 | - | - | - | - | DRISHTI-GS | |||
85.26 | - | - | - | - | ORIGA | ||||
Diaz-Pinto et al. [45] | Xception architecture | 75.25 | 74.19 | 71.43 | - | - | DRISHTI-GS | 10-fold | Not applied |
Guo et al. [46] | random forest classifier + smote | 76.9 | 79.9 | 73.8 | - | - | ORIGA | Hold-out | segmented OD regions using U-net |
Juan et al., 2019 [19] | CNN a with 16 layers | 85.89 | 98.19 | 26.4 | 84.75 | - | DRISHTI GS | Hold-out | Not Applied |
71.88 | 61.62 | 79.3 | 69.44 | - | ORIGA | ||||
Chen et al. [47] | CNN a with 6 layers | 78.02 | 55.66 | 93.56 | 85.8 | - | DRISHTI-GS | Hold-out | Not Applied |
86.86 | 98.7 | 67.6 | 85.21 | - | ORIGA | ||||
Alagirisamy [48] | linear vector quantizer-artificial neural network | 95.05 | 95.71 | 93.55 | - | - | DRISHTI-GS | Hold-out | ROI extraction with micro textures feature extraction |
85.38 | 83.33 | 86.10 | - | - | ORIGA | ||||
Pranath et al. [49] | dynamic support vector machine | 85.5 | - | - | - | - | ORIGA | 10 fold | Not applied |
Chaudhary et al. [50] | image processing 2D-FBSE-EWT | 99 | 97 | 100 | - | - | DRISHTI-GS | 5-fold | Cropped histogram matching method |
92.3 | 91.9 | 94.8 | - | - | ORIGA | ||||
Present study | Swin Transformer | 100 | 100 | 100 | 100 | 100 | DRISHTI-GS | Hold-out | Uncropped |
Swin Transformer | 88.18 | 94.29 | 74.19 | 87.19 | 85.49 | 3-fold | Uncropped | ||
Swin Transformer | 96.15 | 95.05 | 100 | 100 | 97.46 | ORIGA | Hold-out | Cropped | |
InceptionV3 | 93.84 | 97.10 | 84.52 | 92.88 | 91.77 | 5-fold | Uncropped |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Almeshrky, H.; Karacı, A. Optic Disc Segmentation in Human Retina Images Using a Meta Heuristic Optimization Method and Disease Diagnosis with Deep Learning. Appl. Sci. 2024, 14, 5103. https://doi.org/10.3390/app14125103
Almeshrky H, Karacı A. Optic Disc Segmentation in Human Retina Images Using a Meta Heuristic Optimization Method and Disease Diagnosis with Deep Learning. Applied Sciences. 2024; 14(12):5103. https://doi.org/10.3390/app14125103
Chicago/Turabian StyleAlmeshrky, Hamida, and Abdulkadir Karacı. 2024. "Optic Disc Segmentation in Human Retina Images Using a Meta Heuristic Optimization Method and Disease Diagnosis with Deep Learning" Applied Sciences 14, no. 12: 5103. https://doi.org/10.3390/app14125103