Diagnosis of Diabetic Retinopathy through Retinal Fundus Images and 3D Convolutional Neural Networks with Limited Number of Samples
Table 5
Overall accuracy, RCI, and class-wise CEN values for the multiclass classification tasks without augmentation, with random weak Gaussian blurred augmentation, with random shift augmentation, and with combined augmentations.
Method
Overall accuracy
RCI
CEN (mild)
CEN (moderate)
CEN (no)
CEN (proliferate)
CEN (severe)
Without augmentation
36.64%
0.0867
0.7752
0.8557
0.7734
0.5673
0.8015
With random weak Gaussian blurred augmentation
31.04%
0.038
0.8086
0.8458
0.8391
0.7008
0.8272
With random shifted augmentation
30.56%
0.0546
0.8259
0.8834
0.8242
0.6418
0.8165
With combined augmentations
33.12%
0.0647
0.793
0.8352
0.8085
0.6155
0.8341
Test set validated on the model trained without augmentation