Abstract: We propose a new method for training convolutional neural networks which
integrates reinforcement learning along with supervised learning and use ti for
transfer learning for classification of glaucoma from colored fundus images.
The training method uses hill climbing techniques via two different climber
types, viz "random movment" and "random detection" integrated with supervised
learning model though stochastic gradient descent with momentum (SGDM) model.
The model was trained and tested using the Drishti GS and RIM-ONE-r2 datasets
having glaucomatous and normal fundus images. The performance metrics for
prediction was tested by transfer learning on five CNN architectures, namely
GoogLenet, DesnseNet-201, NASNet, VGG-19 and Inception-resnet-v2. A fivefold
classification was used for evaluating the perfroamnace and high sensitivities
while high maintaining high accuracies were achieved. Of the models tested, the
denseNet-201 architecture performed the best in terms of sensitivity and area
under the curve (AUC). This method of training allows transfer learning on
small datasets and can be applied for tele-ophthalmology applications including
training with local datasets.