Optimization of array encoding for ultrasound imaging.

Area of Science:

• Ultrasound imaging
• Medical imaging physics
• Machine learning applications

Background:

• The transmit encoding model is crucial for understanding acoustic transmission effects in synthetic aperture imaging.
• Current scanning sequences represent a limited subset of possibilities for ultrasound image reconstruction.

Purpose of the Study:

• To utilize machine learning (ML) to develop optimized scanning sequences for high-quality B-mode ultrasound images.
• To explore novel encoding sequences beyond conventional methods in synthetic aperture imaging.

Main Methods:

• A custom ML model in PyTorch was developed using simulated radiofrequency (RF) data from Field II.
• The model probes encoding sequences (time delays, apodization weights) to minimize an image quality loss function.
• A novel derivative formulation for delay-and-sum beamforming enabled computational feasibility.

Main Results:

• ML-optimized encoding sequences, when used with the REFoCUS imaging framework, significantly improved resolution, field of view, and contrast.
• Experimental validation on wire targets and a tissue-mimicking phantom confirmed the enhanced image quality metrics.
• The ML approach demonstrated superior performance compared to conventional scanning sequences.

Conclusions:

• Machine learning can discover and optimize novel scanning sequences for synthetic transmit aperture imaging.
• Integrating beamforming within the ML model is valuable for synthetic transmit aperture imaging tasks.
• This work expands the understanding of available encoding schemes, moving beyond narrow, commonly used subsets.