ResNet Repeat Vector Optimized Output Layer based Brain Stroke Prediction

A cerebral infarction is an illness in the nervous system that occurs when the arterial blood circulation to the central nervous system is interrupted, resulting in harm to the cerebral cortex. There might be negative effects if the blood and other resources to the brain are restricted. Immediate recognition of numerous stroke indicators can help to lessen the risk of mortality. This research presents a model Repeat Vector Optimized ResNet (RVO-ResNet) to predict the brain stroke with optimal accuracy. The brain stroke Dataset from Kaggle have been utilized for this research that has 2500 images. The brain stroke Dataset consists of 2500 images, 1550 of which are healthy brain images and 950 of them are brain stroke images. The proposed RVO-ResNet model initiates with labelling the brain images as ‘0’ for healthy and ‘1’ for brain stroke images. The labelled images are fitted with several CNN models to select the best model that has high efficiency and the analysis reveals that the ResNet model is showcasing high accuracy with 84% comparing to all other existing models. The ResNet is choosen to improvize the accuracy by finetuning the optimizers in the output layer. The proposed RVO-ResNet model is designed with single convolution coupled with max pooling layer followed by three Residual network blocks ending with global average pooling optimized with Repeat Vector optimizer for the output layer. The labelled images are fitted various optimizers of ResNet and proposed RVO-ResNet model. The evaluation results demonstrate that the recommended RVO-ResNet with Repeat Vector optimizer exhibits highest Accuracy of 97.51 when analyzed against other optimizers.