Deep convolutional neural network for glaucoma detection based on image classification

Abstract

PURPOSE: Many researchers have found that the improvement in computerised medical imaging has pushed them to their limits in terms of developing automated algorithms for the identification of illness without the need for human participation. The diagnosis of glaucoma, among other eye illnesses, has continued to be one of the most difficult tasks in the area of medicine. Because there are not enough skilled specialists and there are a lot of patients seeking treatment from ophthalmologists, we have been encouraged to build efficient computer-based diagnostic methods that can assist medical professionals in early diagnosis and help reduce the amount of time and effort they spend working on healthy situations. The Optic Disc position is determined with the help of the LoG operator, and a Disc Image map is projected with the help of a U-net architecture by utilising the location and intensity profile of the optic disc. After this, a Generative adversarial network is suggested as a possible solution for segmenting the disc border. In order to verify the performance of the model, a well-defined investigation is carried out on many retinal datasets. The usage of a multi-encoder U-net framework for optic cup segmentation is the second key addition made by this proposed work. This framework greatly outperforms the state-of-the-art in this area. The suggested algorithms have been tested on public standard datasets such as Drishti-GS, Origa, and Refugee, as well as a private community camp-based difficult dataset obtained from the All-India Institute of Medical Sciences (AIIMS), Delhi. All of these datasets have been verified. In conclusion, we have shown some positive outcomes for the detection of diseases. The unique strategy for glaucoma treatment is called ensemble learning, and it combines clinically meaningful characteristics with a deep Convolutional Neural Network.