Reverberation Noise Suppression in Ultrasound Channel Signals Using a 3D Fully Convolutional Neural Network

Abstracts for the 2021 International Symposium on Ultrasonic Imaging and Tissue Characterization

“…Isaac Brighton 1 and Nick Bottenus 1 1 Department of Mechanical Engineering, University of Colorado Boulder, Boulder, CO, nick.bottenus@colorado.edu Objective: To demonstrate the use of a controlled acoustic source to study the image formation process of a black-box ultrasound system. Methods: We use a single element transducer as a switchable acoustic point source, controlled by a microcontroller through a pulser.…”

“…Brickson 1 , Dongwoon Hyun 1 , and Jeremy Dahl 1 1 Stanford University, Stanford, California, USA Background, Motivation and Objective: Diffuse reverberation is caused by multiple reflections of a wavefront before returning to the transducer. This noise degrades both B-mode image quality and ultrasound channel signals, which decreases the performance of techniques that rely on channel signals.…”

“…Previously, we implemented a 3D convolutional neural network (3DCNN) to remove reverberation noise from the channel signals prior to beamforming [1]. In this work, we investigate architecture changes to this 3DCNN to allow reverberation noise reduction of ultrasound images in real-time.…”

“…In this work, we investigate architecture changes to this 3DCNN to allow reverberation noise reduction of ultrasound images in real-time. Statement of Contribution/Methods: The 3DCNNs were trained using the same approach as in our previous work [1]. To increase network speed, adjustments to the network architecture such as decreasing the convolution kernel size, decreasing the number of layers, combining the real and imaginary layers, and reducing the network to a series of permuted 2D convolutions were investigated.…”

“…Clutter induced by reverberation is often considered to be spatially incoherent. The nearest element correlation of backscatter signals provides a robust measure of such incoherent clutter [1]. We recently presented Lag-one Spatial Coherence Adaptive Normalization (LoSCAN) [2], an image formation technique that adaptively compensates for the SNR loss due to incoherent clutter.…”