Effective brain tumor detection is vital in medical diagnostics, as early and accurate identification supports timely treatment. In our research, we introduce a deep learning approach that combines a sparse autoencoder with a Deep Neural Network (DNN) to improve brain tumor classification accuracy. Using a dataset of grayscale brain MRI images categorized by the presence or absence of tumors, we trained our model to automatically compress and learn features that distinguish tumor cases from non-tumor cases. Our proposed model, achieving a notable 98% accuracy, surpasses established architectures, indicating strong potential for practical diagnostic applications in healthcare.