Efficient Multi-Scale Brain Tumor Segmentation: Integrating Adaptive Distributed Attention and Weighted Inception Module

Abstract

This paper introduces a novel approach to enhancing the U-Net architecture for image segmentation tasks, particularly in medical imaging. We propose the integration of a Weighted Inception Module (WIM) and an Adaptive Distributed Attention (ADA) Mechanism to improve segmentation accuracy and computational efficiency. The WIM optimizes multi-scale feature integration, while the ADA employs a learned Gaussian distribution for adaptive attention across input features. Extensive experiments on brain tumor segmentation demonstrate the effectiveness of the proposed Dynamic U-Net, achieving superior performance compared to traditional U-Net variants. This work offers significant contributions to the field of deep learning-based medical image segmentation.