Real-time US reverberation noise suppression using permuted 2D CNN on channel data

A permuted 2D convolutional neural network suppresses diffuse reverberation noise in ultrasound channel signals to improve image quality. In 15-volunteer liver/kidney scans, the real-time model increased contrast and generalized contrast-to-noise ratio (GCNR), running at 15 fps…

• The permuted 2DCNN processes raw ultrasound channel signals, achieving a 15 fps frame rate suitable for real-time clinical use, which was not possible with prior 3DCNN implementations.
• Training used enhanced Field II and Fullwave simulation data incorporating reverberation, aberration, and attenuation to improve robustness.
• This was a proof-of-concept validation on liver and kidney scans from 15 volunteers; a key limitation is the small volunteer cohort without multi-center external validation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.