Predicting ocular diseases using squeezenet as feature maps with Convolutional Neural Networks

Eye disorder is a major cause of vision impairments and blindness throughout the world which became necessary to have a mechanism for early diagnosis of eye disorders in order to treat them promptly and avoid eye illnesses and the visual loss they cause. This will also enhance the overall quality of life for patients. When examining the human eye and its adjacent tissues, either directly or indirectly, ophthalmologists employ pattern recognition to find out the diseases. Early fundus screening is a cheap and feasible approach to preventing blindness caused by ocular disorders in the field of ophthalmology. Manual diagnosis requires a long time in clinical settings and can exacerbate the condition due to the shortage of medical expertise. During fundus screening, ophthalmologists frequently detect multiple conditions using binocular fundus imaging. This study uses a dataset of eight eye diseases to better represent real-world medical circumstances affecting patients' both eyes. The study explores the use of Convolution neural networks (CNN)-SqueezeNet which uses up to 50 times less weights than existing CNNs. Through training on an extremely small CNN network like SqueezeNet, this work adopted an innovative strategy in this research to learn these image attributes. CNN and image pre-processing are used in the current research to predict the diagnosis of ocular eye disorders. When compared to previous CNN architectures like RestNet and DenseNet201, the accuracy of the result is 92.12% greater, as determined by the confusion matrix.