The research is being conducted for developing and investigating state of art deep learning methods for image segmentation and detecting the brain tumor with 3D brain magnetic resonance imaging (MRI). This research explored the U-Net deep learning method for medicinal image segmentation. This research uses publicly available dataset of 7680 high resonance images of 110 patients from the Kaggle. The U-Net pretrained deep learning method trained with three different learning rates (0.001, 0.01, and 0.1) to explore the impact of learning rates on the model efficiency. The dataset has been partitioned into 80%, 15%, and 5% for training, testing, and validation purpose respectively. The trained model used the Adam optimizer and for improving the segmentation performance Dice loss and cross-entropy loss method are utilized. The model performance is evaluated in the form of Dice coefficient and Intersection over Union (IoU) scores. This research is also drawn the confusion matrix for U-Net deep learning method for different learning rates to show the image classification results in the form of precision, recall and f1 score. The model is achieved highest accuracy in the form of 90.10% precision, 83.78% recall, and an F1 score of 0.9062 in tumor classification on learning rate of 0.1. The model is also achieving highest IoU score of 0.7802 and dice score of 0.8728 on the learning rate of 0.1. In conclusion this study explored and characterized a deep learning framework for classify the 3D MRI scans in efficient and effective manner which can help to advancing the medical science to identify the brain tumor effectively and better diagnosis treatment action can be taken.