The ability to anticipate brain stroke is essential for both early intervention and sustaining a healthy life. This work explores the development of an efficient framework for the prediction of persistent stroke prevalence using CNN, ResNet, and the VGG-16 deep learning algorithms. These have improved decision support which leads to early detection of stoke due to its ability to analyze large image dataset. Assessment measures including reliability, F1-score, recall, precision, and specificity are built using the Sigmoid and ReLU activation functions. This study explores these deep learning models’ efficiencies in contrast to other approaches that rely on stacking techniques. High accuracy will be obtained by the trained models using the algorithms of CNN, VGG-16, and ResNet, which attain precision, recall, specificity, and an F1-score.