DrusNet –A CONVOLUTIONAL NEURAL NETWORK FOR THE CLASSIFICATION OF DRUSEN IN OCT AND COLOUR FUNDUS IMAGES

Abstract: Background: Drusen are yellow deposits that are present in the retina. They are made up of proteins and lipids which are present as exudates. Drusen don’t particularly cause Age–Related Macular Degeneration (AMD), but the prolonged presence may cause the development of DRY AMD which can lead to WET AMD. Hence, it’s important to detect these deposits in the early stages. Deep learning-based artificial intelligence is currently creating an impact on healthcare. The application of Deep Learning in Ophthalmology speeds up the diagnosis process and gives accurate results in the early stages. Aim: To develop a deep learning-based algorithm for the diagnosis of drusen in Optical Coherence Tomography (OCT) and Colour Fundus Images. Method: In this study, we proposed a novel deep learning algorithm DrusNet for the diagnosis of drusen in OCT and Colour Fundus Images. DrusNet is an ensemble neural network developed from the pre-trained models EfficientNetB3 and VGG16 for accurate diagnosis. The proposed model DrusNet performed well in both OCT and fundus images. Result: For the classification of OCT images this model achieved an accuracy of 90.15%, an AUC of 0.95, a precision of 86.89%, and a recall of 91.38%. For the classification of Fundus images, this model achieved an accuracy of 95.9an 9%, AUC of 0.99 and, the precision of 98.91%, and recall of 91.00%. Conclusion: The proposed architecture may be used by ophthalmologists as a pre-screening tool to identify drusen deposits at an early stage to prevent AMD.