A non-sub-sampled shearlet transform-based deep learning sub band enhancement and fusion method for multi-modal...

Area of Science:

• Medical Imaging
• Image Processing
• Artificial Intelligence

Background:

• Multi-modal medical image fusion (MMMIF) integrates data from various imaging types for better diagnosis.
• Limitations in individual modalities cause information loss in fused images.

Purpose of the Study:

• To develop a novel fusion framework to overcome limitations in medical image fusion.
• To enhance the reliability of clinical decision-making systems through improved fused image quality.

Main Methods:

• A novel framework combining Non-Subsampled Shearlet Transform (NSST) and Convolutional Neural Networks (CNN) was proposed.
• Source images were decomposed into Low-Frequency Coefficients (LFC) and High-Frequency Coefficients (HFC) using NSST.
• A Concurrent Denoising and Enhancement Network (CDEN) processed sub-bands, followed by fusion using AlexNet and Pulse Coupled Neural Network (PCNN) with a Novel Sum-Modified Laplacian (NSML) metric.

Main Results:

• The proposed method significantly improved edge preservation, achieving approximately 16.5% higher QAB/F metric performance.
• Experimental results demonstrated superior performance compared to existing fusion algorithms.
• Both subjective visual assessments and objective quality indices confirmed the effectiveness of the proposed fusion technique.

Conclusions:

• The proposed NSST-CNN based fusion framework effectively addresses information degradation in multi-modal medical images.
• This advanced fusion technique enhances diagnostic accuracy and supports clinical decision-making.
• The method offers a significant improvement over conventional medical image fusion algorithms.