Multi-path fusion network based global attention for brain tumor segmentation
Dongyuan Wu, Shilin Qiu, Jing Qin, Pengbiao Zhao · 2023
In many clinical applications, medical picture segmentation is essential for the precise delineation of anatomical features and diseased areas. However, due to the complexity and variability of medical images, traditional segmentation methods often face challenges in achieving robust and accurate results. This paper proposes a novel method called Multi-Path Fusion Network (MPFN) for segmenting medical image to overcome these drawbacks. The MPFN leverages the power of deep learning and incorporates multiple pathways to gather contextual information on both local and global scale. The network architecture consists of parallel pathways. Each pathways is designed to extract various levels of spatial features from the input MRI image. These pathways are then fused using an attention mechanism, allowing the network to selectively emphasize relevant features while suppressing noise and irrelevant information. Furthermore, the MPFN introduces a multi-scale strategy to deal with variations in object sizes and shapes within medical images. By incorporating feature maps at different resolutions, the network can effectively capture both fine-grained details and high-level context, improving segmentation accuracy. The experiment show that the proposed method achieves competitive result in Brat 2015 segmentation tasks, outperforming existing traditional neural network.