Impact of convolutional neural networks on brain tumor classification: A systematic analysis

Abstract

This systematic literature review assessed the impact of brain tumor classification on magnetic resonance imaging using convolutional neural networks (CNN). The PRISMA strategy was used to organize and document the study selection and evaluation process. From an initial total of 448 articles identified in Scopus and Web of Science, duplicates were eliminated, and inclusion and exclusion criteria were applied, resulting in the evaluation of 103 highly relevant and quality studies. The most used architectures, including ResNet-50, VGG16 and EfficientNet-B4, were identified and analyzed, assessing their effectiveness in terms of accuracy, sensitivity and F1-score. The results showed that these architectures can achieve accuracy levels above 98%, making them effective for potential implementation in clinical settings. In addition, key pre-processing techniques, such as normalization and denoising, were highlighted as contributing to improved image quality and reduced variability. However, important barriers were identified, such as the scarcity of high-quality data and privacy constraints, which limit the generalizability and robustness of the models. The review concludes that the creation of standardized datasets and the development of innovative methodological approaches are essential to advance the clinical applicability of CNNs in brain tumor diagnosis.