PROJECT TITLE :

Automatic Cataract Classification Using Deep Neural Network With Discrete State Transition

ABSTRACT:

Cataract is the primary cause of blindness in the globe because of the clouding of the lens. It will be better if cataract detection and severity assessment can be done in a more accurate and convenient manner. This research proposes methods for automatically detecting and rating cataracts. Classifiers are built using multilayer perceptron with discrete state transition (DST-MLP) or exponential DST (EDST-MLP) and improved Haar features combined with visible structure characteristics based on prior knowledge. The residual neural networks with DST or EDST (ResNet-ResNet) are proposed without prior knowledge. To prevent overfitting and limit storing memory in neural networks, we've devised DST and EDST techniques that might be used with or without prior knowledge. These neural networks obtain the highest level of accuracy in cataract diagnosis and grading. Combining features yields better results than using a single type of feature alone, and approaches that use feature extraction based on prior knowledge are better suited for difficult medical picture classification problems. There are many other medical image processing applications that could benefit from these findings.