Research on Medical Image Fusion Based on Multi-Scale Dual-Branch Pyramid Attention Network

Medical image fusion technology integrates complementary information from multimodal medical images. This technology plays a significant role in clinical applications. The article introduces a novel convolutional neural network, termed the Multi-Scale Dual-Branch Pyramid Attention (MSPA) network, for fusing multimodal medical images. The network comprises a feature extraction module, a feature fusion module, and an image reconstruction module. The MSPA block comprises two distinct branches. The first branch incorporates a pyramid mechanism, which employs six convolution kernels of identical size to collectively extract features. In contrast, the second branch utilizes four 3x3 convolution kernels. Additionally, the article adopts a feature L1-norm fusion strategy to merge the features from the input images. Compared to other algorithms, it achieves superior results across multiple fusion evaluation metrics.