Tumors in brain can be either benign or cancerous. In order to enhance patient outcomes and therapy, it is important to detect or diagnose nodules early on. It is common practice to use PET, MRI, and CT scans for spotting the brain tumor. Recent developments in accuracy of brain tumor diagnoses using AI-based approaches, such as ML and DL-based approaches, have provided hope to radiologists and other members of the medical community. Deep learning methods, especially convolutional neural networks, showed promising results in complicated medical datasets. CNN’s ability to accurately diagnose and categorize brain cancers using large datasets makes it an indispensable tool in contemporary medical imaging. Despite these advancements, additional research is necessary to guarantee the safe and successful clinical use of these technologies while reducing computational cost and maintaining high accuracy. Brain tumor analysis utilizing YOLOv5 and U-Net architecture is the main emphasis of this work. Finally, to evaluate which model is more suitable, we conducted a comparison based on several key parameters, including batch size, number of epochs, and the type of optimizers used. The models were validated using publicly available MRI datasets, and U-Net achieved the highest performance with an accuracy of 96.17%. This research provides valuable insights for designing more efficient CNN-based models aimed at the early detection of brain tumors.