Detection of Pathological Myopia Using Convolutional Neural Network

A study reveals that 3% of the world population suffers from pathological myopia. Pathological myopia is an extreme case of nearsightedness which affects individuals during their most productive years, leads to vision loss which is progressive and irresistible. High myopia is defined as refractive error of at least -6.00 D or an axial length of 26.5mm or more. Of all the Pathological Myopia cases, thirty percent occur at birth. Most of the patients are diagnosed with the condition between the ages 6 and 13, and it continues to progress throughout life. It is crucial to examine children at risk, as failure to detect high myopia at an early age may lead to further vision loss from amblyopia. Conventional methods involve manual detection of Pathological Myopia which have led to inaccurate diagnosis and has resulted in complete vision loss. Deep Learning architecture have achieved state-ofthe- art performance; performed better than their human counterparts in problems of Computer Vision since 2016, so the chances of inaccurate diagnosis are minute. As mentioned above, Deep Learning network architectures have achieved high accuracy in Computer Vision problems. We approach this problem as an Image Classification task using deep convolutional neural networks such as Residual Networks (ResNet) and Dense Convolutional Network (DenseNet). We were able to achieve an accuracy of 95.34% using ResNet-50 and 98.08% using DenseNet121.