1 Introduction
2 Related Work
2.1 Ensemble Learning-Based Identification of Eye Diseases
2.2 Deep Learning-Based Identification of Eye Diseases
3 Hybrid Images Deep-Trained Feature Extraction and Ensemble Learning Algorithm for Categorizing Multiple Diseases in Fundus Images
3.1 Data Characterization and Preparation
3.2 Feature Selection
Model | Input Size |
---|---|
DenseNet201 | 224, 224, 3 |
InceptionResNetV2 | 299, 299, 3 |
ResNet152V2 | 224, 224, 3 |
MobileNetV2 | 224, 224, 3 |
VGG19 | 224, 224, 3 |
NASNetMobile | 224, 224, 3 |
VGG16 | 224, 224, 3 |
NASNetLarge | 331, 331, 3 |
Model | Selected Features |
---|---|
VGG16 | 4096 |
VGG19 | 4096 |
DenseNet201 | 1920 |
MobileNetV2 | 1280 |
ResNet152V2 | 2048 |
InceptionResNetV2 | 1536 |
NASNetMobile | 1056 |
NASNetLarge | 4032 |
3.3 Classification
3.4 Deep Learning Models for Classification
Diabetic Retinopathy | Glaucoma | Age-Related Macular Degeneration | |
---|---|---|---|
DR | 125 (TP) | 10 (FP) | 5 (FP) |
Glaucoma | 8 (FP) | 120 (TP) | 12 (FP) |
ARMD | 7 (FP) | 9 (FP) | 116 (TP) |
4 Result and Discussion
Classes | Extra Tree | Histogram Gradient Boosting | Random Forest | ||||||
---|---|---|---|---|---|---|---|---|---|
PRE | REC | F1S | PRE | REC | F1S | PRE | REC | F1S | |
NORMAL | 98 | 98 | 98 | 97 | 96 | 97 | 97 | 98 | 98 |
TESSELLATED FUNDUS | 98 | 100 | 99 | 95 | 100 | 98 | 100 | 100 | 100 |
LARGE OPTIC CUP | 99 | 99 | 99 | 96 | 97 | 97 | 99 | 99 | 99 |
DR1 | 98 | 100 | 99 | 97 | 98 | 97 | 98 | 98 | 98 |
DR2 | 100 | 97 | 98 | 97 | 91 | 94 | 99 | 97 | 98 |
DR3 | 100 | 98 | 99 | 98 | 98 | 98 | 100 | 99 | 100 |
POSSIBLE GLAUCOMA | 100 | 100 | 100 | 98 | 100 | 99 | 98 | 100 | 99 |
OPTIC ATROPHY | 100 | 100 | 100 | 97 | 97 | 97 | 95 | 100 | 97 |
SEVERE HYPERTENSIVE RETINOPATHY | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 |
DISC SWELLING AND ELEVATION | 93 | 100 | 96 | 95 | 100 | 98 | 98 | 100 | 99 |
DRAGGED DISC | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 |
CONGENITAL DISC ABNORMALITY | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 |
RETINITIS PIGMENTOSA | 99 | 99 | 99 | 99 | 99 | 99 | 99 | 99 | 99 |
BIETTI CRYSTALLINE DYSTROPHY | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 |
PERIPHERAL RETINAL DEGENERATION AND BREAK | 100 | 100 | 100 | 96 | 100 | 98 | 96 | 100 | 98 |
MYELINATED NERVE FIBER | 100 | 100 | 100 | 97 | 97 | 97 | 100 | 100 | 100 |
VITREOUS PARTICLES | 98 | 100 | 99 | 98 | 100 | 99 | 100 | 100 | 100 |
FUNDUS NEOPLASM | 100 | 100 | 100 | 96 | 100 | 98 | 100 | 100 | 100 |
BRVO | 98 | 97 | 97 | 97 | 97 | 97 | 99 | 96 | 97 |
CRVO | 100 | 99 | 99 | 97 | 99 | 98 | 99 | 99 | 99 |
MASSIVE HARD EXUDATES | 100 | 100 | 100 | 100 | 100 | 100 | 95 | 100 | 98 |
YELLOW-WHITE SPOTS-FLECKS | 94 | 100 | 97 | 98 | 95 | 96 | 97 | 99 | 98 |
COTTON-WOOL SPOTS | 100 | 100 | 100 | 94 | 100 | 97 | 100 | 100 | 100 |
VESSEL TORTUOSITY | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 | 100 |
CHORIORETINAL ATROPHY-COLOBOMA | 98 | 100 | 99 | 98 | 98 | 98 | 98 | 100 | 99 |
PRERETINAL HEMORRHAGE | 100 | 100 | 100 | 97 | 100 | 98 | 100 | 100 | 100 |
FIBROSIS | 97 | 100 | 98 | 97 | 97 | 97 | 94 | 100 | 97 |
LASER SPOTS | 96 | 100 | 98 | 100 | 98 | 99 | 93 | 98 | 95 |
SILICON OIL IN EYE | 100 | 100 | 100 | 98 | 98 | 98 | 98 | 100 | 99 |
BLUR FUNDUS WITHOUT PDR | 99 | 100 | 99 | 99 | 98 | 98 | 99 | 99 | 99 |
BLUR FUNDUS WITH SUSPECTED PDR | 99 | 99 | 99 | 93 | 99 | 96 | 97 | 99 | 98 |
RAO | 100 | 100 | 100 | 98 | 98 | 98 | 100 | 100 | 100 |
RHEGMATOGENOUS RD | 99 | 96 | 97 | 98 | 96 | 97 | 99 | 93 | 96 |
CSCR | 96 | 100 | 98 | 96 | 100 | 98 | 98 | 100 | 99 |
VKH DISEASE | 96 | 100 | 98 | 98 | 100 | 99 | 96 | 100 | 98 |
MACULOPATHY | 100 | 96 | 98 | 96 | 97 | 97 | 98 | 96 | 97 |
ERM | 99 | 100 | 99 | 95 | 96 | 96 | 99 | 99 | 99 |
MH | 100 | 97 | 99 | 100 | 97 | 99 | 100 | 97 | 99 |
PATHOLOGICAL MYOPIA | 100 | 99 | 100 | 99 | 98 | 99 | 100 | 100 | 100 |
AVERAGE | 98.8 | 99.3 | 99 | 97.5 | 98.3 | 98 | 98.4 | 99.1 | 98.8 |
Description | ML based | EL based | DL based | Ref |
---|---|---|---|---|
Automated diabetic retinopathy classification using fundus images | Y | N | N | [6] |
Automatic diabetic retinopathy classifier | Y | N | N | [5] |
A DL Ensemble Model for classifying diabetic retinopathy | N | Y | N | [2] |
Transfer learning retinal disease classification | N | Y | N | [3] |
Neural network classification of ocular diseases in STARE database | N | N | Y | [7] |
Deep neural network for multi-label optical illness classification | N | Y | N | [4] |
Deep learning for color fundus image retinal abnormality detection | N | Y | N | [7] |
Hierarchical multilabel ANN for eye disease classification | N | N | Y | [27] |
Deep learning for retinal disease diagnosis | N | N | Y | [11] |
Optical coherence tomographical scans using CNN for retinal disease | N | N | Y | [30] |
Multi-label ocular disease detection with fundus images | N | N | Y | [31] |
DL method to analyze fundus images based on macular edema | N | N | Y | [29] |
Efficientnet’s Multi-Label Fundus Classification | N | Y | N | [9] |
Ref. | Dataset | Number of Images | Ground Truth Labels | Diagnosis Source | Both Eyes of the Same Patient | Glaucoma (or Suspect) | Glaucoma Classification |
---|---|---|---|---|---|---|---|
[37] | RIGA | 750 | — | — | ✓ | — | — |
[38] | ORIGA | 650 | 482 | 168 | ✓ | ✓ | ✓ |
[39] | RIMONE | 485 | 313 | 172 | ✓ | ✓ | Clinical |
[40] | Drishti-G | 101 | 70 | 31 | ✓ | ✓ | Image |
[41] | ACRIMA | 705 | 309 | 396 | ✓ | ✗ | Image |