Deep Null Space Learning Improves Dataset Recovery for High Frame Rate Synthetic Transmit Aperture Imaging

Area of Science:

• Medical Imaging
• Ultrasound Technology
• Signal Processing

Background:

• Synthetic transmit aperture (STA) imaging offers high resolution but suffers from low frame rates and signal-to-noise ratio (SNR).
• Previous compressed sensing (CS-STA) and least squares (LS-STA) methods improved frame rate but introduced artifacts, especially in shallow regions, due to neglecting the null space component.
• Understanding the null space component is crucial for accurate STA dataset recovery.

Purpose of the Study:

• To propose a novel convolutional neural network under the null space learning framework (CNN-Null) for accurate recovery of STA datasets.
• To estimate the missing null space component of the STA dataset from fewer Hadamard-encoded (HE) plane wave (PW) transmissions.
• To effectively suppress artifacts in beamformed images, particularly in shallow regions.

Main Methods:

• Trained a convolutional neural network (CNN-Null) to learn the mapping between low-quality STA datasets (range space component) and the missing null space component.
• Utilized high-quality STA datasets from full HE-STA imaging as training labels.
• Compared CNN-Null performance against LS-STA, conventional STA, and HE-STA using metrics like NRMSE, gCNR, and FWHM.

Main Results:

• CNN-Null significantly improved STA dataset recovery accuracy, evidenced by lower Normalized Root Mean Square Error (NRMSE).
• The method effectively suppressed artifacts in beamformed images, particularly in shallow regions, achieving a 0.4 generalized contrast-to-noise ratio (gCNR) improvement in carotid artery images.
• High lateral resolution was maintained with as few as 16 PW transmissions, similar to LS-STA.

Conclusions:

• The proposed CNN-Null method accurately recovers STA datasets by estimating the null space component, overcoming limitations of previous methods.
• CNN-Null offers a significant advancement in ultrasound imaging, reducing artifacts and improving image quality without compromising resolution.
• This technique enables high-quality STA imaging with reduced data acquisition, paving the way for faster and more reliable ultrasound diagnostics.