A Comprehensive Analysis of Recent Deep Learning Based Methodologies for Brain Tumor Diagnosis

Brain tumors are a serious global health challenge because their occurrence is associated with high mortality and difficulties in accurate diagnosis.. Timely and efficient diagnosis is essential to successful treatment and better patient outcomes. This paper is a systematic comparison-based review of the recent deep learning methods of brain tumor detection and classification based on magnetic resonance imaging (MRI). A structured screening process was applied to 523 candidate papers retrieved from IEEE Xplore, Scopus, Google Scholar, and ScienceDirect (2022–2025). Studies were evaluated against four quality criteria: (i) reproducibility of the experimental setup, (ii) completeness of reported metrics, (iii) clarity of architectural description, and (iv) dataset transparency. Of 523 papers, 136 duplicates were removed, leaving 387 for screening. After title-and-abstract screening, 94 full-text papers remained, from which 83 were further excluded for insufficient methodological detail, non-reproducibility, or risk of bias, yielding 11 studies for final inclusion. This paper summarizes the research outcomes of six deep learning models such as ResNet50, Efficient Net, and YOLOv7, U-Net, VGG16 and a CNN-RNN hybrid model in terms of reported metrics, such as accuracy, precision, recall and F1-score. These models were evaluated on different tumor datasets and task types, and results are interpreted accordingly within their respective domains. The reported findings reveal that segmentation-based and hybrid deep learning models are more effective. U-Net has the most accuracy of 99.9%, then Yolov7 with 99.5%. ResNet50 has an accuracy of 98.78%, whereas Efficient Net has 97.8%. The hybrid CNN-RNN model is reported to record 94.7% accuracy and VGG16 is reported to be significantly low at 93.35% accuracy. The paper concludes by identifying future research directions, including privacy-preserving federated learning, integration of multimodal imaging, and the development of deep learning systems suitable for clinical settings.