Multi-Scale Mixed Attention Network for CT and MRI Image Fusion

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Telemedicine relies on analyzing multi-modal medical images, posing challenges in efficiency and convenience.
• Existing convolutional neural network (CNN)-based medical image fusion methods often use simple weighted averaging, leading to information loss from non-essential elements.

Purpose of the Study:

• To develop an improved method for fusing Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) images.
• To enhance the quality of fused medical images by preserving essential information and reducing the impact of irrelevant data.

Main Methods:

• A novel CNN-based medical image fusion method (MMAN) is proposed.
• A multi-scale mixed attention block is designed for effective feature extraction at channel and spatial levels.
• A visual saliency-based fusion strategy is employed to prioritize important image information.

Main Results:

• The proposed MMAN method preserves more textual details compared to state-of-the-art techniques.
• Enhanced clarity of edge information and higher contrast in the fused images were observed.
• The visual saliency-based approach effectively mitigates the weakening effect of inessential information.

Conclusions:

• The MMAN method offers a superior approach to CT and MRI image fusion.
• This technique improves the quality and diagnostic utility of fused medical images for telemedicine applications.
• The integration of multi-scale attention and visual saliency marks a significant advancement in medical image fusion.