Multi-Modality Medical Image Fusion Using Convolutional Neural Network and Contrast Pyramid

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Medical image fusion is crucial for enhancing diagnostic reliability and accuracy across various clinical applications.
• Existing fusion methods often struggle to preserve fine structural details and achieve optimal visual quality.

Purpose of the Study:

• To develop a novel convolutional neural network (CNN) based algorithm for high-quality medical image fusion.
• To improve the preservation of detailed structure information and human visual effects in fused medical images.

Main Methods:

• A Siamese convolutional network was trained to generate weight maps by fusing pixel activity information from source images.
• A contrast pyramid was employed to decompose source images into different spatial frequency bands.
• A weighted fusion operator integrated source images across various frequency bands.

Main Results:

• The proposed CNN-based fusion algorithm demonstrated effective preservation of detailed structure information from source images.
• Comparative experiments confirmed superior human visual effects compared to existing methods.
• The algorithm achieved high visual quality in the fused medical images.

Conclusions:

• The developed CNN-based medical image fusion algorithm offers a significant advancement in the field.
• This technique enhances the reliability and accuracy of medical diagnoses through improved image fusion.
• The algorithm provides a promising tool for clinical applications requiring high-fidelity fused medical images.