Related Experiment Video
Updated: Sep 11, 2025
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
15.8K
Image harmonization and de-harmonization based on singular value decomposition (SVD) in medical domain
Huachao Chen1, Xinze Li1, Ka-Hou Chan1
1Faculty of Applied Sciences, Macao Polytechnic University, Macao, China.
Quantitative Imaging in Medicine and Surgery
|August 11, 2025
Summary
This study introduces novel singular value decomposition (SVD) algorithms for medical image harmonization. These methods improve image consistency and quality while preserving diagnostic details, enhancing machine learning performance.
Area of Science:
- Medical Imaging
- Machine Learning
- Image Processing
Background:
- Medical imaging variations cause inconsistencies, hindering diagnostic accuracy and AI model performance.
- Existing harmonization techniques often compromise critical image details.
- This study addresses image variability while preserving essential diagnostic information.
Purpose of the Study:
- To develop novel singular value decomposition (SVD)-based harmonization and de-harmonization algorithms.
- To ensure consistency across diverse medical imaging conditions.
- To preserve essential diagnostic information in medical images.
Main Methods:
- Utilized SVD to decompose images into frequency bands for targeted adjustments.
- Applied SVD to RGB channels for selective enhancement of relevant structures and contrast normalization.
- Integrated harmonization and de-harmonization processes, validated through homology and heterology experiments on diverse datasets (MNIST, USPS, DRIVE, CHASE_DB1, RSNAbreast, INbreast).
Main Results:
- SVD algorithms outperformed traditional methods in image quality and efficiency.
- Achieved high accuracy in digit classification (e.g., 99.21% on MNIST).
- Demonstrated strong performance in retinal vessel segmentation (AUCs up to 0.982) and breast cancer detection (AUCs up to 0.934).
Conclusions:
- SVD-based algorithms offer a robust solution for medical image variability.
- The techniques enhance visual quality and clinical utility across datasets and modalities.
- Proven potential to improve machine learning model effectiveness in medical imaging tasks.
