AFPNet: An adaptive frequency-domain optimized progressive medical image fusion network

Dangguo Shao, Hongjuan Yang, Lei Ma, Sanli Yi

Abstract: This study presents AFPNet, a novel Progressive Medical Image Fusion Network incorporating Adaptive Frequency-Domain Optimization, designed to enhance the fusion process of multi-modal medical imaging modalities, including SPECT-MRI and PET-MRI. AFPNet exploits a progressive Attention Convolutional Neural Network(ACNN) to significantly enhance the quality of fused medical images. Key innovations encompass a dual-branch module for efficient spatial and channel feature fusion, a Global Enhancement Attention Module for the seamless integration of global information, and a multi-scale feature fusion strategy to capture diverse contextual information. Experimental results on the test datasets indicate that AFPNet generally outperforms state-of-the-art approaches in key metrics such as Mutual Information (MI) and Visual Information Fidelity (VIF), showing notable improvements in preserving both structural coherence and fine-grained details. The incorporation of adaptive frequency-based weighting mechanisms within the model optimize fusion of high-frequency and low-frequency components, rendering it well-suited for medical imaging applications demanding exceptional precision.