Table 4.
Literature Review on ML and DL models for early detection of DR
| Paper Name | Images/Dataset | Features | Methodology | Classifier | Results |
|---|---|---|---|---|---|
| Lam et al. [77] | 243 | MAs, HEs, EXs, NV | CNN | CNN |
AlexNet accuracy 74% and 79% VGG16 accuracy 86% and 90% GoogLeNet accuracy 95% and 98% ResNet accuracy 92% and 95% Inception-v3 accuracy 96% and 98% |
| Buades et al. [22] | 7137 | EXs, HEs, drusen, CWSs | SURF, OR, MV and meta-SVM | SVM |
AUC 91.6% for single lesion detection (hard EXs) AUC of 88.3%multi-lesion detection |
| Mansour et al. [89] | Kaggle dataset | AlexNet, SIFT, LDA and PCA | SVM |
Accuracy of 90.15% (PCA) and 97.23% on FC6 features (LDA) Accuracy of 95.26% (PCA) and 97.28% on FC7 features (LDA) |
|
| Orlando et al. [101] |
DIARETDB1, MESSIDOR, e-ophtha |
MAs, HEs | LeNet | CNN |
AUC of 0.7912 with CNN features AUC of 0.7325 with hand crafted features AUC of 0.8932 on combination of both features |
| Pratt et al. [108] | 80,000 | MAs, EXs and HEs | CNN architecture | Sensitivity 30%, Specificity 95%, Accuracy 75% | |
| Xu et al. [142] | Kaggle |
EXs, red lesions, MAs RBVs |
CNN | GBM and CNN |
91.5% accuracy without data augmentation 94.5% accuracy with data augmentation |
| Khojasteh et al. [71] | DIARETDB1 and e-Ophtha |
EXs, MAs, HEs |
CNN | CNN |
Accuracy of 0.96 (EXs), 0.98(HEs) and 0.97 (MAs) on DIARETDB1 Accuracy of 0.88(EXs), and 3.0(MAs), on e-Ophtha dataset |
| Soniya et al. [127] | DIARETDB0 |
MA, HE, hard EX, soft EX, NVE |
Single CNN and heterogenous CNN | Multilayer perceptron network |
Single CNN accuracy: 40%–90% Heterogenous CNN accuracy: 100% |
| Alghamdi et al. [8] | PAMDI and HAPIEE | OD | Cascaded CNNs | AdaBoost ensemble algorithm |
Sensitivity 96.42%, Specificity 86%, Accuracy 86.52% on HAPIEE Sensitivity 94.54%, Specificity 98.59%, Accuracy 97.76% on PAMDI |
| Gardner et al. [44] | EyePACS-1 MESSIDOR-2 | HE and MA | Inception-V3-architecture Neural Network | Ensemble of 10 networks |
EyePACS-1: AUC of 0.991 MESSIDOR-2: AUC of 0.990 |
| Abràmoff et al. [86] | 1748 | OD, Fovea, HEs, EXs NV | AlexNet, VGGNet | RFC |
Sensitivity of 96.8% (rDR), Specificity of 87%, AUC 0.98 (rDR) Sensitivity of 100% (vtDR), Specificity of 90.8%, AUC 0.989 (vtDR) |
| Lam et al. [78] | Kaggle MESSIDOR-1 | Pretrained AlexNet and GoogLeNet | GoogLeNet 2-ary, 3-ary and 4-ary |
2-ary accuracy 74.5% 3-ary accuracy 68.75% 4-ary accuracy 51.25% |
|
| Takahashi et al. [132] |
9939 images |
HE and hard EX | GoogLeNet DCNN and ResNet |
Accuracy 96% (real prognosis) Accuracy 92% (Davis grading) |
|
| Quellec et al. [112] |
MESSIDOR Kaggle e-ophtha DiaretDB1 |
hard EXs, soft EXs, small red dots, HEs, lesions | ConvNet netB | Ensemble classifier |
Az of 0.954 in Kaggle’s dataset Az of 0.949 in e-Ophtha dataset Az of 0.9490 using ensemble classifier |
| Alban et al. [6] | 35,126 | Pre-trained AlexNet | GoogLeNet |
AUC 0.79 (GoogLeNet) AUC 0.69 (AlexNet) |
|
| Kermany et al. [70] | 207,130 OCT |
CNV, DME Drusen |
Pretrained Inception V3 |
Binary classification accuracy: >98% Multi-class classification accuracy of 96.6% |
|
| Dutta et al. [34] | Kaggle dataset | RBVs, fluid drip, EXs, HEs, MAs | NN, DNN, VGG-16 | VGG-16 |
Accuracy: 72.5% for 300 test images Accuracy: 78.3% for 600 test images |
| Grinsven et al. [136] | Kaggle MESSIDOR | VGGNet |
area under ROC of 0.894 on Kaggle dataset area under ROC of 0.972 MESSIDOR dataset |
||
| Liskowski et al. [84] |
DRIVE STARE CHASE |
DNN |
area under ROC >99% Accuracy >97% |
||
| Islam et al. [57] | EyePACS | MAs |
DCNN-18, PLAIN BALANCE, NO-POOL |
area under ROC 0.844 F1-Score 0.743 |
|
| Prentasic et al. [109] | DRiDB | OD, RBVs, EXs | DCNN | Sensitivity 78%, Positive Predictive Value (PPV) 78%, F-score 78% | |
| Mookiah et al. [96] | 156 | RBVs, EX, OD, NVE | PNN | SVM |
Sensitivity 96.27%, Specificity 96.08%, PPV 98.2%, Accuracy 96.15% |
| Rakhlin [117] |
Kaggle MESSIDOR-2 |
drusen, EXs, MAs, CWSs, HEs |
DCNN | VGGNet |
area under ROC 0.923 in Kaggle dataset area under ROC 0.967 in MESSIDOR-2 dataset |
| Gargeya et al. [45] |
EyePACS MESSIDOR-2 E-Ophtha |
HEs, hard EXs and NVE | DRL CNN |
AUC 0.94 on MESSIDOR-2 AUC 0.95 on E-Ophtha AUC 0.97 on EyePACS |
|
| Eftekhari et al. [36] |
ROC E-Ophtha-MA |
MAs | DLNN |
FROC (FAUC) of 0.660 for ROC FROC (FAUC) of 0.637 for E-Optha-MA dataset |
|
| Wang et al. [138] | 35,126 |
CNN Net-5 and Net-4 |
Kappa score 0.70 for 256-pixel images, Kappa score 0.80 for 512-pixel images Kappa score 0.81 for 768-pixel images |
||
| Dai et al. [27] | 735 | MAs | MS-CNN | Recall 87.8%, precision 99.7%, accuracy 96.1%, and F1 score of 93.4% | |
| Al-Bander et al. [7] | MESSIDOR Kaggle | Fovea and OD | multi sequential DL technique |
1R criterion: Accuracy 97% (OD) and 96.6% (foveal) in MESSIDOR test set Accuracy 96.7%(OD) and 95.6% (foveal) in Kaggle test set |
|
| Gadekallu et al. [43] | DR Debrecen | PCA, Firefly model | DNN |
Accuracy of 97%, Precision of 96%, Recall of 96%, Sensitivity of 92%, Specificity of 95%. |
|
| Chudzik et al. [26] | E-Ophtha, ROC DIARETDB1 | MAs | FCNN |
FROC score 0.193 ± 0.116 on ROC FROC score 0.392 ± 0.157 on DIARETDB1 FROC score 0.562 ± 0.233 on E-Ophtha |