Switchable Deep Beamformer for High-quality and Real-time Passive Acoustic Mapping.

Area of Science:

• Medical Imaging
• Acoustics
• Machine Learning

Background:

• Passive acoustic mapping (PAM) is crucial for monitoring ultrasound therapy.
• Data-adaptive beamformers offer superior image quality but are computationally intensive.
• Existing methods like time exposure acoustics (TEA) have limitations in image quality and computational efficiency.

Purpose of the Study:

• To develop a computationally efficient deep beamformer for PAM.
• To reconstruct high-quality PAM images from radiofrequency ultrasound signals.
• To enable real-time monitoring of acoustic cavitation activities in ultrasound therapy.

Main Methods:

• A deep beamformer based on a generative adversarial network was developed.
• The model was trained on simulated and experimental cavitation signals from various transducer arrays (1-15 MHz).
• Performance was evaluated against TEA and data-adaptive beamformers using test datasets.

Main Results:

• The deep beamformer reduced energy spread by 27.3%-77.8% and improved SNR by 13.9-25.1 dB compared to TEA.
• It achieved a 3-fold reduction in computational cost, with an image reconstruction speed of 10.5 ms.
• Image quality was comparable to data-adaptive beamformers, with significantly lower computational demands.

Conclusions:

• The deep beamformer offers a high-quality, computationally efficient solution for PAM.
• This technology has the potential for high-resolution monitoring of microbubble cavitation in ultrasound therapy.
• The developed method can adapt to different transducer arrays, enhancing its versatility.