Medical diagnosis and treatment planning significantly depend on brain tumor classification processes to detect tumors early and develop appropriate therapies. The presented approach utilizes Convolutional Neural Networks (CNN) together with Long Short-Term Memory (LSTM) networks for classifying four categories of MRI brain scans which include glioma and meningioma as well as pituitary tumor and no tumor. The trained model works on a preprocessed set which includes grayscale MRI images that received resizing and normalization procedures to reach better learning outcomes. Images produce spatial features through CNN evaluation which pairs effectively with LSTM analysis that detects sequential patterns for better classification. The proposed network produces performance results matching or exceeding those of typical networks VGG16, ResNet50, and EfficientNet when evaluated with accuracy measurements along with confusion matrices and classification report metrics. The robustness is enhanced through data augmentation that includes Gaussian and salt-and-pepper noise application as well as noise reduction techniques to achieve better image quality. The model generates effective tumor classifications through high accuracy which indicates its usefulness in automated brain tumor diagnosis.