Brain Tumor Classification Using Deep Learning: A Review

Among the neurological diseases, brain tumors are one of the most dangerous and require an early and accurate localization. It can be achieved through medical image registration and brain tumor classification, which allow performing the precise alignment of multimodal medical images and the right determination of tumor features. This review highlights the state-of-the-art of the latest advances in deep learning image registration and brain tumor classification approaches. The related literature was systematically retrieved through IEEE Xplore, PubMed, ScienceDirect, SpringerLink and Google Scholar from 2018 to 2025 using the following keywords: brain tumor classification, medical image registration, deep learning, convolutional neural network, vision transformer and magnetic resonance imaging. Journal and conference papers relevant to the topic, innovative methodological approaches and the highest performance results were selected for inclusion into the review. The paper provides a review of mono-modal and multi-modal image registration approaches based on MRI, CT, PET, ultrasound and X-ray imaging technologies. In addition, the following deep learning approaches: CNN, FCN, U-Net, GAN, ViT, GNN and hybrid learning algorithms will be discussed. The issues related to benchmark datasets, evaluation criteria, clinical applications and comparison studies are also covered in this review. Moreover, important issues such as lack of annotated data, domain adaptation, computational complexity, explainability, privacy protection and generalizability to different imaging modalities have been emphasized. At the end of this study, new research directions related to foundation models, federated learning, XAI, multimodal transformers and intelligent decision-support systems for clinical decisions have been introduced for precision neuro-oncology.