Image quality improvement in single plane-wave imaging using deep learning.

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Artificial Intelligence

Background:

• Single plane-wave imaging (SPWI) offers high temporal resolution but sacrifices spatial resolution and contrast.
• Coherent plane-wave compounding (CPWC) improves image quality but reduces temporal resolution.
• Existing methods for reconstructing high-quality images from limited plane waves do not fully utilize RF signal properties.

Purpose of the Study:

• To develop novel methods for reconstructing high-quality ultrasound images from SPWI data.
• To achieve image quality comparable to CPWC while maintaining high temporal resolution.
• To address limitations of current reconstruction techniques by incorporating RF signal characteristics.

Main Methods:

• Utilized encoder-decoder models, including 1D U-Net, 2D U-Net, and their combination.
• Minimized a loss function considering the point spread effect and RF signal frequency spectrum during training.
• Developed and utilized a large-scale public dataset of SPWI/CPWC data for training and validation.

Main Results:

• Proposed methods successfully reconstructed higher-quality ultrasound images from SPWI RF signals compared to conventional methods.
• Demonstrated the effectiveness of the deep learning approaches on both public and custom datasets.
• Achieved image quality comparable to CPWC from SPWI data.

Conclusions:

• The proposed deep learning methods effectively enhance ultrasound image quality in SPWI.
• These methods offer a promising solution for obtaining high-resolution, high-contrast ultrasound images with high temporal resolution.
• The developed dataset facilitates further research in AI-driven ultrasound imaging.