Diffusion-driven multi-modality medical image fusion.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Multi-modality medical image fusion (MMIF) enhances clinical diagnosis by integrating complementary information from different imaging techniques.
• Current deep learning methods for MMIF often neglect inter-modality information distribution, leading to suboptimal fusion of details and color.

Purpose of the Study:

• To develop an advanced MMIF method that effectively leverages the correlation of information distribution across modalities.
• To improve the fusion of image details and color information for more comprehensive diagnostic insights.

Main Methods:

• A diffusion-driven MMIF approach is proposed, utilizing latent space relationships among multi-modality images.
• A local and global network (LAGN) is introduced to preserve complementary information.
• A specialized loss strategy is designed to enforce constraints between original, generated, and fused images, preventing information loss.

Main Results:

• The proposed diffusion-driven MMIF method outperformed state-of-the-art techniques on MRI/CT, MRI/PET, and MRI/SPECT datasets based on unsupervised metrics.
• The method demonstrated superior capability in capturing rich image details and color information.
• Clinical evaluation by medical professionals indicated the method's effectiveness in assisting diagnosis and treatment planning.

Conclusions:

• The diffusion-driven MMIF method offers a significant advancement in medical image fusion.
• This approach effectively addresses limitations of existing deep learning models by preserving critical image details and color.
• The method shows strong potential for improving clinical decision-making through enhanced diagnostic imaging.