A Robust Framework for Glaucoma Detection and Segmentation of Real Time Fundus Images using Hybrid ASCL Model

A rise in intraocular pressure within the retina led to the development of the neuro-degenerative eye condition known as glaucoma. Becoming the second-leading source of blindness and visual impairment, it may result in total blindness if an early diagnosis is not made. A method that really can function well without a lot of technology, highly trained medical personnel, or a lot of time is desperately needed to address this fundamental problem. In this study, an offline Computer-Aided Diagnosis (CAD) method employing retinal fundus pictures is proposed. Machine learning, deep learning, as well as image processing techniques are all used in the implementation of the software. ASCL Model comprises of AlexNet, SegNet, CNN and LGBM classifier. Region of Interest (ROI) identification is carried out using the sharpest point technique, and input picture processing is performed to use the AlexNet structure. Additionally, the SegNet design is used to divide the optic disc as well as optic cup, and LightGBM classifier along with CNN are used to classify the data. AlexNet input picture verification achieves a 99 percent accuracy rate. When the sharpest spot method is used, an efficiency of 98.67% is obtained for ROI generation. Furthermore, the classification of the optic disc as well as optic cup that used the SegNet design yielded dice coefficients of 0.93 and 0.87, correspondingly. The suggested approach used CNN, and LightGBM Models to obtain classification performance, recall, specificity, and sensitivity of 100 percent, demonstrating the device's dependability and potential.