|
| Ref. | Methods | Metrics (%) | Performance analysis |
|
CNN-based approaches | [20] | Hyperparameter tuning inception V4 model | Accuracy = 99 Specificity = 98 Sensitivity = 99 Precision = 97 | Need employment of versatile classification methods |
[21] | Limited adaptive histogram equalization methods | Accuracy = 97 Specificity = 98 Sensitivity = 94 Precision = 97 F1 score = 94 G-Means = 94 | Image quality needs to be improvised |
[23] | CNN-based binocular network | Specificity = 82 Sensitivity = 7 Kappa score = 0.824 | Availability of limited data leading to biased results |
[25] | Capsule network | Accuracy = 95 | Training of limited image features |
[26] | Fuzzy C-means | Accuracy = 98 | Inclusion of limited datasets |
[33] | Feature extraction + segmentation methods | Accuracy, precision, and recall evaluated DR severity | Need for robust DL techniques to improve results |
[34] | Deep CNN model | Sensitivity = 91 Specificity = 90 | System excluded possibilities implementation in clinical environment |
[35] | Deep CNN with inception-V4 model | Accuracy = 88 Precision = 96 Recall = 94 | Information from other domains with plausible impact on results are not included |
[37] | EffectiveNet method | AUC = 0.68 Kappa score = 0.36 | Biased results are generated from unbalanced dataset |
[39] | Deep CNN with inception method | Accuracy = 92 Specificity = 94 Sensitivity = 81 Precision = 93 | Automated image prognosis method is not implemented |
|
Transfer learning approaches | [24] | Transfer learning approach | Accuracy = 97 Specificity = 95 Sensitivity = 92 | Biased classification results |
[38] | CNN with transfer learning | Accuracy = 74 | Inability to detect minor diseases |
|
Deep neural network-based approaches | [22] | PCA with firefly algorithm using DNN | Accuracy = 97 Specificity = 95 Sensitivity = 92 Precision = 96 Recall = 96 | Low-dimensional data are not considered |
[27] | DNN patch-based approach | Accuracy = 97 Specificity = 95 Sensitivity = 92 | Use of limited data leading to enhanced cost |
[30] | Deep feature extraction method | Sensitivity = 94 Specificity = 98 AUC curve = 0.97 | High computational cost |
[31] | DNN algorithms | Average precision (AP) = 0.88 IoU (intersection over union) = 0.17 | Lack of larger datasets |
[32] | Score propagation method | Sensitivity = 91 Specificity = 90 | Image pixel score identification model is not included |
|
Machine learning-based approaches | [29] | SVM and random forest techniques | Accuracy = 90 | Enhanced image classifier methods are not implemented |
[36] | PCA with linear regression | Accuracy = 92 | Larger datasets are not considered |
|