Sophisticated methods utilizing deep learning neural networks have surfaced as viable options for precise and automated brain tumor identification. This study, which employed use of MobileNetV2, EfficientNet, ResNet50 and VGG16 for the comparitive analysis and select the one which gives highest accuracy. We address the need for more advanced diagnostic techniques by automatically extracting pertinent characteristics from MRI scans and making accurate predictions using deep learning and Grad-Cam segmentation. Our approach entails preprocessing MRI data, CNN architecture design, and model training and assessment with dataset. Our methodology is successful, as seen by the accuracies of MobileNetV2 at 86%, EfficientNet at 84%, ResNet50 at 80%, VGG16 at 88%. The Grad-Cam segmentation performed at the final stage helps in identifying the presence of tumor. All things considered, this study advances the area of medical image analysis and emphasizes how critical it is to use MRI and deep learning to combat brain cancers.