Brain tumor detection and classification are critical for timely diagnosis and effective treatment. The surge in demand for automated and accurate methods is driven by the advancements in deep learning and the need for faster, more reliable diagnostic tools to assist clinicians. Despite the current literature about brain tumor classification and detection, several limitations persist. This survey reviews and contrasts the state-of-the-art deep learning diagnostic techniques that utilized brain tumor datasets such as Figshare and BraTS. This study provides a technical analysis of research papers on brain tumor diagnosis techniques, covering the period from 2020 to 2024 from well-known databases such as Scopus and Web of Science. Recent deep learning methodologies, including convolutional neural networks (CNNs), transfer learning, vision transformers (ViTs), hybrid techniques, and explainable AI, are explored regarding their performance, advantages, and limitations. We examine various architectures, preprocessing techniques, and datasets commonly used in brain tumor studies, focusing on multi-class classification, detection, and interpretability. Furthermore, the survey discusses the challenges in deep learning-based approaches related to brain tumor detection, including data scarcity and model interpretability, and outlines future directions.