Branch Retinal Artery Occlusion (BRAO) has become a serious disorder causing permanent loss of vision. It is mainly caused due to the rupture of blood vessels which is too small in size and thereby increasing the risk for efficient detection. Therefore an efficient detection scheme is necessary to evaluate appropriately and pave the way for systemic therapy to preserve or recover vision in the affected eye. In this paper, a fully automatic detection scheme has been developed for accurate segmentation followed by classification of BRAO volumes using the fundus images. Coming to the point this is the first automatic classification framework using an optimized neural network for classifying BRAO. It's mainly a three-step process where the first step concentrates on removing noise and thereby mproving the quality of fundus image. According to which an boosted anisotropic diffusionbased enhancement filter is used here, which effectively removes the noise. This is justified using image quality metrics. Second step here involves adaptive cluster with super pixel segmentation which clearly segments the affected area with high contrast. For the segmented image IR Analyzer is used for extracting region properties. In the final step gray wolf optimization-based convolution neural network (GWO-CNN) classifier is used. Here GWO-CNN classifier is used to differentiate the normal and abnormal images. The effectiveness of the above method is evaluated and obtained an accuracy of 98.57% when it is implemented in MATLAB 2018 software.
Read full abstract