AI techniques for brain tumor segmentation in MRI: a review (2019-2024)

Abstract

BackgroundBrain tumors are the tenth most common tumor to affect both adults and children, and their prevalence is increasing. Since early detection is crucial to improving treatment outcomes, extensive research is being conducted to accurately diagnose cancers using Computer Tomography (CT) and Magnetic Resonance Imaging (MRI). MRI offers comprehensive evaluations of anomalies in brain tissue and is recommended due to its superior imaging capabilities and non-invasive nature.MethodsThis review systematically analyzed 75 peer-reviewed publications from 2019 to 2024, focusing on the application of AI-based brain tumor segmentation using MRI. Studies were selected following PRISMA guidelines. Deep learning techniques, particularly convolutional neural networks (CNNs) and transformer-based models, are comprehensively analyzed across multiple benchmark datasets. Performance metrics such as Dice similarity coefficients and Hausdorff distances were used for evaluation.ResultsThe review demonstrates notable progress in brain tumor segmentation, with deep learning continuously surpassing classical techniques. Across the reviewed studies, hybrid CNN-Transformer architectures consistently outperformed standalone CNN or ViT models, often achieving Dice similarity scores exceeding 90%. Supervised learning techniques and hybrid models are also promising in tackling segmentation problems.ConclusionFuture research in computational methods and medical imaging is expected to yield improved brain tumor detection and treatment strategies. Further research is needed to enhance clinical outcomes and improve segmentation algorithms. Key challenges identified include limited generalizability across datasets, difficulties with small-sample learning, model interpretability, and dataset biases. This review offers a current and critical perspective to inform future innovations in brain tumor segmentation.