Real-Time Reverberation Suppression in Ultrasound Channel Signals Using a Permuted 2D Convolutional Neural Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Diffuse reverberation noise in ultrasound imaging obscures anatomical targets and degrades image quality.
• This noise results from multiple echo reflections, creating speckle-like artifacts difficult to distinguish from tissue.
• Existing methods struggle with real-time suppression and distinguishing noise from genuine tissue signals.

Purpose of the Study:

• To introduce a permuted 2D convolutional neural network (2DCNN) for real-time diffuse reverberation suppression in ultrasound channel signals.
• To enhance B-mode image quality and serve as a pre-processing filter for advanced ultrasound techniques.
• To achieve computational advantages for real-time implementation on existing hardware.

Main Methods:

• Developed a permuted 2D convolutional neural network (2DCNN) architecture for reverberation suppression.
• Trained the 2DCNN on a dataset combining Field II and Fullwave simulations, including various acoustic effects.
• Validated the technique on liver and kidney scans from 15 volunteers.

Main Results:

• The 2DCNN achieved real-time implementation at 15 frames per second, offering significant computational advantages over 3D CNNs.
• Demonstrated improvements in image quality metrics, including increased contrast, generalized contrast-to-noise ratio (GCNR), and lag-one coherence (LOC).
• Successfully removed diffuse reverberation noise, enhancing visualization of anatomical structures.

Conclusions:

• The proposed 2DCNN is effective for real-time diffuse reverberation suppression in ultrasound imaging.
• This technique significantly improves image quality and has potential as a pre-processing tool for various ultrasound applications.
• The method offers a computationally efficient solution for enhancing ultrasound diagnostic capabilities.