Fig 1.
The pipeline of the proposed CNN-Transformer model.
MLP denotes multi-layer perceptron.
Fig 2.
The stem module in the proposed Inception-Resnet-V2 model.
Fig 3.
The structures of the introduced vision mamba model.
SSM denotes state space model introduced in the work of [20].
Table 1.
The distribution of APTOS2019 and Messidor datasets.
Table 2.
The leveraged hyper-parameters values of the proposed approach.
Table 3.
Lesion detection comparison between the state-of-the-arts and the proposed approach on the Messidor dataset.
Fig 4.
The loss and accuracy curves of both training and testing processes for the proposed approach on the Messidor dataset.
Fig 5.
The confusion matrix for DR and healthy classification on the Messidor dataset.
Table 4.
DR grading comparison between the state-of-the-arts and the proposed approach.
Fig 6.
The loss and accuracy curves of both training and testing processes for the proposed approach on the APTOS2019 dataset.
Fig 7.
The confusion matrix for DR grading on the APTOS2019 dataset by using the proposed approach.
Fig 8.
The fundus images in the dataset using by the proposed approach.
(Top) DR 0; (Middle Left) DR1; (Center) DR 2; (Middle Right) DR 3; (Bottom) DR 5.
Table 5.
Outcome of the ablation study on the Messidor dataset.
Table 6.
Outcome of the ablation study on the APTOS2019 dataset.