Ultrasound Phase Aberrated Point Spread Function Estimation with Convolutional Neural Network: Simulation Study.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Ultrasound
• Biomedical Engineering

Background:

• Accurate point spread function (PSF) estimation is crucial for enhancing ultrasound image quality.
• Phase aberration in biological tissues distorts the PSF, posing a significant challenge for accurate estimation.
• Existing methods struggle with the complex distortions introduced by phase aberration.

Purpose of the Study:

• To develop a deep learning framework for robust PSF estimation in the presence of phase aberration.
• To evaluate the performance of U-Net and complex U-Net architectures on RF and k-space data.
• To identify optimal loss functions for accurate PSF prediction.

Main Methods:

• Proposed a deep learning framework utilizing U-Net and complex U-Net architectures.
• Trained networks on synthetic phase aberration data generated using the near-field phase screen model.
• Evaluated various loss functions, identifying log-compressed B-mode perceptual loss as optimal.

Main Results:

• The complex U-Net architecture operating on k-space data demonstrated superior performance.
• Log-compressed B-mode perceptual loss accurately predicted both mainlobe and near sidelobe regions of the PSF.
• The proposed method effectively estimated PSFs under varying levels of phase aberration.

Conclusions:

• The deep learning framework provides accurate PSF estimation in phase-aberrated ultrasound imaging.
• Improved PSF estimation enhances performance in downstream phase aberration correction tasks.
• The method offers broader utility for ultrasound image quality enhancement.