DCL-A: An Unsupervised Ultrasound Beamforming Framework with Adaptive Deep Coherence Loss for Single Plane Wave...

Area of Science:

• Ultrasound imaging
• Medical imaging physics
• Signal processing

Background:

• Single plane wave imaging (SPWI) enables ultrafast ultrasound acquisition but suffers from beamforming artifacts.
• Sidelobe and grating lobe interferences degrade SPWI image quality, limiting real-time applications.

Purpose of the Study:

• To introduce an unsupervised deep learning beamforming framework, adaptive deep coherence loss (DCL-A), for enhanced artifact suppression in SPWI.
• To improve the overall image quality of SPWI without compromising its high frame rates.

Main Methods:

• Developed an unsupervised beamforming framework utilizing adaptive deep coherence loss (DCL-A) with linear or nonlinear weighting.
• The adaptive weight (α) is determined by the angular distance between input and target plane wave (PW) frames during training.
• Validated the DCL-A framework using simulation, phantom, and in vivo ultrasound datasets.

Main Results:

• DCL-A with nonlinear weighting achieved superior peak range sidelobe level (PRSLL) reduction (7-14 dB) compared to conventional DCL and linear DCL-A in simulation and phantom studies.
• Maintained comparable full width at half maximum (FWHM) across methods.
• In vivo studies showed DCL-A with nonlinear weighting yielded the highest contrast resolution, improving generalized contrast-to-noise ratio (gCNR) by 2-3%.

Conclusions:

• The proposed deep learning-based beamforming framework significantly enhances SPWI image quality.
• The DCL-A method effectively suppresses artifacts without sacrificing frame rate.
• Demonstrates potential for high-speed, high-resolution clinical ultrasound applications like cardiac imaging and interventional guidance.