Deep learning based early Glaucoma detection

Abstract

Glaucoma is a severe eye condition that can lead to progressive vision impairment if left untreated. Diagnosis and monitoring of glaucoma at an initial stage is critical for effective treatment of the disease. However, the diagnosis is complex with bare eyes which requires multiple checkups and tests. Image processing is important for diagnosing glaucoma by providing valuable information about the intricate structure of the eye which helps improve the accuracy of diagnosis and allows for earlier detection of the disease. This portrays the requirement for further research in this field. We aim to explore various image-processing models used for image classification and develop an efficient model that can be used in the detection of glaucoma. In this paper, we have proposed a Custom CNN model with 22 layers based on deep learning for glaucoma diagnosis where it detects from the fundus images whether the person has glaucoma or not. The model has been trained using datasets containing 4000 fundus images each with 2 categories which are Glaucoma and Non-Glaucoma. The datasets have been used on the Custom CNN model and six other pre-trained models. Our proposed model has been able to successfully classify the images with an accuracy of 98.71% which was the highest among all the models despite having a lower number of parameters compared to the other models.