Using Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) methods, this study aims to create accurate models that can predict the growth of brain tumours. We used a complete method that included steps like picture cropping, normalization, resampling, and improvement to prepare the BraTS dataset from the University of Pennsylvania. After that, we used the U-net model to do photo segmentation, which was very important for appropriately defining the rims of the tumour. We use LSTM to predict how a tumour will develop through the years and then enhance it with a blended LSTM-GRU model to make it greater correct and use less computing electricity. We checked how properly our models anticipated the future by using looking on the dice coefficient and Intersection over Union (IOU) measures for segmentation consequences. next, we checked the accuracy of our predictions the usage of suggest Absolute mistakes (MAE) evaluation The comparison of performance showed that the blend model not only minimizes losses better, but it also makes much better predictions about how tumours will grow, which suggests that it could be used in personalized treatment plans and keeping track of progress.