Automatic Detection of AMD and DME Retinal Pathologies Using Deep Learning

Table 1

Relevant works.

Reference	Pathology	Dataset	Preprocessing	Used CNN	Classifier	Result Val_accuracy
Bhowmik et al. [12]	DME, drusen, CNV,; and normal	Kaggle dataset: retinal OCT images	Resizing	InceptionV3 and VGG16	Softmax	VGG16: 91.6%; InceptionV3: 92%
Yang et al. [13]	Normal, dry and wet AMD	Local datasets	—	CNN (EG)	Softmax	96.05%
Wang D. and; Wang L. [14]	AMD, DME, and normal	Dataset 1: SERI; Dataset 2: Duke	Filtering: FastNIMeans+bilateralFilter	VGG16; VGG19; InceptionV3; CliqueNet; DPN; DenseNet; ResNet; ResNext		Dataset 1: VGG16: 91.6%; VGG19: 98.2%; InceptionV3: 92.7%; CliqueNet: 99%; DPN: 99.6%; DenseNet: 98.7%; ResNet: 98.7%; ResNext: 97.3%; Dataset 2: VGG16: 86.3%; VGG19: 95.1%; InceptionV3: 85.3%; CliqueNet: 98.6%; DPN: 95.8%; DenseNet: 94.2%; ResNet: 95.2%; ResNext: 94.8%
Vaghefi et al. [15]	Normal_Young, Normal_Old,; and dry AMD	75 participants	Resizing	Inception-ResnetV2	Softmax	99.8%
Karri et al. [16]	MD, DME, and normal	Duke	Denoising using BM3D+(flattening)	GoogLeNet	SVM	96%