2.5D Multimodal MRI Brain Tumor Segmentation: Beyond 2D Deep Learning Approaches

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This study proposes a 2.5D input framework for brain tumor segmentation in multi-sequence magnetic resonance imaging (MRI), aiming to balance computational efficiency and spatial context compared to conventional 2D approaches. While 2D methods are efficient but limited in capturing inter-slice information, the proposed approach incorporates adjacent slices as additional channels within a standard U-Net architecture to capture local context from multimodal MRI (T1, T2, T1ce, and FLAIR). The method is evaluated on a dataset of 119 glioblastoma patients. Results demonstrate that the 2.5D approach improves segmentation performance over 2D methods, achieving a significant improvement in Dice and a near-significant trend in IoU while maintaining computational efficiency. It further enhances tumor boundary delineation and better preserves fine structural details. These findings highlight the effectiveness of 2.5D representations as a practical and balanced solution for brain tumor.