Attention-Based Segmentation of Brain Tumors Using Feature Level Fusion and UNet Architecture

In medical image analysis, brain tumor segmentation is a crucial step in precise brain tumor diagnosis, treatment planning, and tumor monitoring. In this study, we introduce FusedD2AResNeXtUNet, a novel segmentation method for brain tumor segmentation that incorporates a fusion-attention mechanism. This approach leverages advanced fusion techniques to combine multi-level features, enhancing the model’s focus on the most relevant features and improving the accuracy and robustness of tumor segmentation. By addressing challenges such as variability in tumor shape and size, the fusion mechanism ensures more precise and reliable outcomes. We evaluate our method using the BraTS2020 dataset, a benchmark dataset in the field. Experimental results demonstrate significant improvements in segmentation performance, including better sensitivity, fewer false positives, high pixel-wise accuracy, reduced pixel-wise error, and increased Dice accuracy. The proposed method achieves a Mean Dice of $\mathbf{9 0. 0 \%}$ on test images, outperforming conventional segmentation techniques. These findings highlight the potential of our fusionenhanced method as a valuable tool for clinical applications in brain tumor diagnosis and treatment planning.