Brain tumors (BT) are a very common, deadly condition with a very poor prognosis at the most aggressive grade. BT represent a prevalent and fatal condition with particularly poor prognosis at advanced stages. Accurate diagnosis and classification are critical for effective treatment planning and patient care. This study introduces a novel deep learning-based algorithm for early BT detection using CT and MRI images. The proposed model enhances image quality using Adaptive Trilateral filtering, extracts feature via the MobileNet model, selects relevant features through the Tyrannosaurus optimization algorithm, and classifies brain tumors with a Deep Belief Network (DBN). The model categorizes tumors into three classes: pituitary tumor, no tumor, and glioma tumor. In comparison to conventional deep learning networks, the model performs better in experiments using the BRATS2020 dataset, reaching a 99.3% accuracy rate and great dependability. This paper offers significant improvements in automated brain tumor detection, promising better patient outcomes through early and precise diagnosis.