Deep-Learning-Based Electrical Noise Removal Enables High Spectral Optoacoustic Contrast in Deep Tissue
IEEE Transactions on Medical Imaging
|June 3, 2022
Summary
A new deep learning method effectively removes electrical noise from multispectral optoacoustic tomography (MSOT) signals. This technique enhances image contrast and spectral information, improving diagnostic capabilities for real-time clinical applications.
Area of Science:
- Medical Imaging
- Biomedical Engineering
- Artificial Intelligence
Background:
- Electrical noise significantly degrades image contrast in multispectral optoacoustic tomography (MSOT).
- Existing signal processing methods are inadequate for complex noise patterns and real-time applications.
- Advanced noise reduction is crucial for accurate MSOT image interpretation.
Purpose of the Study:
- To develop a deep learning approach for effective electrical noise removal in MSOT.
- To improve image quality and diagnostic potential of MSOT through enhanced signal processing.
- To enable real-time noise reduction for clinical MSOT applications.
Main Methods:
- A discriminative deep learning algorithm was developed to separate electrical noise from optoacoustic signals.
- The algorithm utilizes spatiotemporal correlations within the entire optoacoustic sinogram.
- Training involved a large dataset of experimental noise and synthetic optoacoustic signals.
Main Results:
- The deep learning model accurately removed electrical noise from synthetic data, phantoms, and in vivo human breast images.
- Significant improvements in morphological and spectral MSOT images were observed.
- A 19% increase in blood vessel contrast and enhanced spectral contrast at depths over 2 cm were achieved.
Conclusions:
- The proposed deep learning framework offers a robust solution for electrical noise reduction in MSOT.
- This method significantly enhances image quality and diagnostic accuracy.
- The approach is suitable for real-time clinical multispectral optoacoustic tomography.
