Fig 1.
Original capsule network (CapsNet) by Sabour et. al., 2017.
Fig 2.
Proposed CLAHE-CapsNet architecture.
Fig 3.
Comparison of original squash and power squash activation function.
Fig 4.
Sample Retina OCT images.
Table 1.
Dataset description.
Fig 5.
Training, validation accuracy and loss curve on Retina OCT images.
(a) Training and validation accuracy curves of CLAHE-CapsNet and original CapsNet., and (b) Training and validation loss curves of CLAHE-CapsNet and original CapsNet.
Fig 6.
Performance comparison of model based on individual classes (i.e., CNV, DME, DRUSEN, and NORMAL).
Fig 7.
Histogram accuracy comparison based on the overall accuracies of OA, OS, and OP.
Table 2.
Comparison of results of the proposed model and original CapsNet.
The best performance is labeled in bold.
Fig 8.
Confusion matrix comparison on the proposed model and original CapsNet.
Fig 9.
Comparison of ROC-AUC on the proposed model and original CapsNet.
(a) CLAHE-CapsNet ROC curve, and (b) Original CapsNet ROC curve.
Fig 10.
Precision-Recall curves comparison on the proposed model and original CapsNet.
(a) CLAHE-CapsNet Precision-Recall curve, and (b) Original CapsNet Precision-Recall curve.
Fig 11.
Comparison of results with state-of-the-art works based on the individual classes.
Fig 12.
Comparison of results with state-of-the-art works based on the individual classes.
Table 3.
Comparison of results of the proposed model and other state-of-the-art works.
The best performance is labeled in bold.