Eye disease has evolved into a global health issue and has become widespread. Nowadays, ocular diseases spread all over the world and cause dangerous effects in humans. A wide range of eye diseases can significantly affect visual clarity, which can lead to permanent vision loss. Among these, glaucoma affects the optic nerve due to the increased eye pressure, often leading to vision loss. Thus, it requires an effective strategy that includes healthcare providers, public health officials and community education to prevent these diseases. One of the most effective measures for reducing the consequences of ocular disease among the population is periodic check-ups and early recognition of particular diseases. By extracting enriched features in the fundus and ocular computed tomography images, the multiple layers in the deep learning architectures help in accurate image classification and segmentation of specific areas in images. In this research, ocular disease detection with a severity classification model is proposed to prevent early vision loss. The process is started by gathering input images needed for the detection task. Then, the segmentation process is done on the input images using the developed Trans-EfficientUnetFormula: see text (TEUNetFormula: see text model. This segmentation process is helpful for faster analysis of the affected regions. Then, the ocular disease classification into the normal case and abnormal case will be performed via Adaptive and Region Attention-based Pyramid Dilated EfficientNetB7 (ARA-PDEB7). Here, an Enhanced Magnificent Frigatebird Optimization with Random Number Amendment (EMFO-RNA) was introduced for optimizing the performance of the classification model to obtain accurate outcomes. If the disease is detected, then severity assessments are done to take appropriate treatments. The features in segmented images are retrieved using the Pyramid Dilated EfficientNetB7 (PDEB7). The ocular disease classification performance of the presented model was analyzed among existing models to verify its efficiency.
Murikipudi et al. (Tue,) studied this question.