Diagnosis of Diabetic Retinopathy through Retinal Fundus Images and 3D Convolutional Neural Networks with Limited Number of Samples

Table 4

Ranking of methods for the binary classification tasks based on performance metrics without augmentation, with random weak Gaussian blurred augmentation, with random shift augmentation, and with combined augmentations.

Performance metric	Ranking
Accuracy	(1) With combined augmentations; (2) With random weak Gaussian blurred augmentation; (3) With random shift augmentation; (4) Without augmentation
-score	(1) With combined augmentations; (2) With random weak Gaussian blurred augmentation; (3) With random shift augmentation; (4) Without augmentation
MCC	(1) With combined augmentations; (2) With random weak Gaussian blurred augmentation; (3) With random shift augmentation; (4) Without augmentation
Sensitivity	(1) With combined augmentations; (2) With random weak Gaussian blurred augmentation; (3) With random shift augmentation; (4) Without augmentation
Specificity	(1) With random weak Gaussian blurred augmentation; (2) With combined augmentations; (3) With random shift augmentation; (4) Without augmentation
Precision	(1) With combined augmentations; (2) With random weak Gaussian blurred augmentation; (3) With random shift augmentation; (4) Without augmentation
Overall performance	(1) With combined augmentations; (2) With random weak Gaussian blurred augmentation; (3) With random shift augmentation; (4) Without augmentation