CT and MRI Medical Image Fusion Using Noise-Removal and Contrast Enhancement Scheme with Convolutional Neural Network

Area of Science:

• Medical Imaging
• Image Processing
• Computer Vision

Background:

• Medical Image Fusion (MIF) is crucial for accurate clinical diagnosis.
• Existing MIF methods struggle with poor contrast, noise, and information loss.
• A robust MIF technique is needed to overcome these limitations.

Purpose of the Study:

• To develop an improved medical image fusion method.
• To address challenges like non-uniform illumination and noise.
• To enhance the representation of significant features in fused medical images.

Main Methods:

• Morphological preprocessing (bottom-hat-top-hat) for noise and illumination correction.
• Grey-Principal Component Analysis (grey-PCA) for RGB to grayscale conversion preserving details.
• Local Shift-Invariant Shearlet Transform (LSIST) for multi-scale and multi-directional decomposition.
• Siamese Convolutional Neural Network (CNN) for high-pass sub-band analysis.
• Local energy fusion for low-pass sub-band merging.

Main Results:

• Subjective validation by twelve specialists confirmed superior detail, contrast, and noise reduction.
• Objective metrics showed significant gains in QFAB (0.6836-0.8794), CRR (0.5234-0.6710), and AG (3.8501-8.7937).
• Demonstrated substantial noise reduction from 0.3397 to 0.1209 compared to other methods.

Conclusions:

• The proposed MIF method effectively enhances image quality and diagnostic utility.
• It successfully mitigates common issues like noise and non-uniform illumination.
• The technique offers a superior approach for medical image fusion, validated by expert and objective assessments.