3D Convolutional Neural Networks for Brain Tumor Analysis in Multimodal MRI: A Systematic Review

Convolutional neural networks (CNNs) have emerged as a preferred approach for medical image analysis. The dimensionality of images is an important factor in CNN models, as they are designed to interpret 2D or 3D image data. This study provides a comprehensive assessment of 3D CNN-based models for brain tumor segmentation and classification of magnetic resonance imaging (MRI) scans, focusing on their performance in terms of accuracy and the Dice Similarity Coefficient (DSC), along with an analysis of self-reported limitations. The integration and processing of diverse data from different MRI sequences, such as resolution, contrast, and noise, pose challenges in the construction of robust models. Consequently, we focused on multimodal MRIs to provide a comprehensive overview of peer-reviewed literature on 3D CNN models. Therefore, we performed a systematic literature review (SLR) of articles published between 2019 and 2024 in PubMed, Scopus, and IEEE Xplore databases. The SLR initially identified 554 potentially eligible studies screened for relevance and quality, resulting in the inclusion of 32 studies. Based on these studies, we conducted a systematic and quantitative analysis from technical and task perspectives. Our findings indicate that the technical aspects of 3D CNN-based models for brain tumor segmentation and classification can be further improved. We also discussed the limitations of implementing 3D models for brain tumor analysis. Furthermore, we explore the challenges of translating deep learning (DL) techniques into clinical settings and offer insights into future research trends and advancements.