A Medical Image Fusion Method Based on SIFT and Deep Convolutional Neural Network in the SIST Domain

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Traditional medical image fusion methods struggle with sparse feature representation and information transfer.
• Existing techniques like multi-scale decomposition often yield suboptimal fusion results.

Purpose of the Study:

• To propose an advanced medical image fusion method.
• To overcome limitations of traditional fusion techniques by improving feature representation and fusion rule efficacy.

Main Methods:

• A novel fusion method utilizing Scale Invariant Feature Transform (SIFT) and deep Convolutional Neural Networks (CNN) in the Shift-Invariant Shearlet Transform (SIST) domain.
• Decomposition into high-pass and low-pass coefficients.
• Fusion of high-pass components using a pre-trained CNN model (feature detection, segmentation, verification, fusion).
• Fusion of low-pass subbands based on SIFT descriptor matching degree.
• Reconstruction via SIST inversion.

Main Results:

• The proposed method effectively preserves detailed information without artifacts or distortions.
• Objective measurements (standard deviation, Q^AB/F, entropy, mutual information) show superior quantitative performance.
• Enhanced fusion quality compared to traditional methods.

Conclusions:

• The SIFT and CNN-based fusion method in the SIST domain offers significant improvements in medical image fusion.
• The approach successfully addresses the challenges of sparse representation and information transfer in feature fusion.
• This method provides a robust and accurate solution for medical image fusion applications.