Deep Neural Networks for Ultrasound Beamforming

Abstract

We investigate the use of deep neural networks (DNNs) for suppressing off-axis scattering in ultrasound channel data. Our implementation operates in the frequency domain via the short-time Fourier transform. The inputs to the DNN consisted of the separated real and imaginary components (i.e. in-phase and quadrature components) observed across the aperture of the array, at a single frequency and for a single depth. Different networks were trained for different frequencies. The output had the same structure as the input and the real and imaginary components were combined as complex data before an inverse short-time Fourier transform was used to reconstruct channel data. Using simulation, physical phantom experiment, and in vivo scans from a human liver, we compared this DNN approach to standard delay-and-sum (DAS) beamforming and an adaptive imaging technique that uses the coherence factor (CF). For a simulated point target, the side lobes when using the DNN approach were about 60 dB below those of standard DAS. For a simulated anechoic cyst, the DNN approach improved contrast ratio (CR) and contrast-to-noise (CNR) ratio by 8.8 dB and 0.3 dB, respectively, compared to DAS. For an anechoic cyst in a physical phantom, the DNN approach improved CR and CNR by 17.1 dB and 0.7 dB, respectively. For two in vivo scans, the DNN approach improved CR and CNR by 13.8 dB and 9.7 dB, respectively. We also explored methods for examining how the networks in this work function.

Adaptive and Compressive Beamforming Using Deep Learning for Medical Ultrasound

“…Recently, inspired by the tremendous success of deep learning, many researchers have investigated deep learning approaches for various inverse problems [10]- [21]. In US literature, the works in [22], [23] were among the first to apply deep learning approaches to US image reconstruction. In particular, Allman et al [22] proposed a machine learning method to identify and remove reflection artifacts in photoacoustic channel data.…”

“…In particular, Allman et al [22] proposed a machine learning method to identify and remove reflection artifacts in photoacoustic channel data. Luchies and Byram [23] proposed a frequency domain deep learning method for suppressing offaxis scattering in ultrasound channel data. In [24], a deep neural network is designed to estimate the attenuation characteristics of sound in human body.…”

Image Reconstruction: From Sparsity to Data-Adaptive Methods and Machine Learning

“…One of the most important recent developments in the field of image reconstruction is the introduction of deep learning approaches [38]. Motivated by the tremendous success of deep learning for image classification [153], [154], image segmentation [155], denoising [156], etc, many groups have recently successfully applied deep learning approaches to various image reconstruction problems such as in X-ray CT [157]- [163], MRI [33], [162], [164]- [168], PET [169], [170] ultrasound [171], [172], and optics [173]- [175]. As of mid 2019, two commercial CT vendors received FDA approval for deep learning image reconstruction [176] [177].…”

CAI4CAI: The Rise of Contextual Artificial Intelligence in Computer-Assisted Interventions

“…This raises an interesting question: If it is known what information is desired or desirable at any given point during surgery, is it possible to prospectively acquire an image that is most informative in that particular context? First steps in this direction have recently been reported, exploiting ultrasound image formation to suppress scatter [89] or beamforming a B-mode image [90], [91] together with producing its segmentation [69]. Zaech et al [92] use an AI-based algorithm to recommend task-optimal and patient specific C-arm X-ray trajectories during conebeam CT of spinal fusion surgery, and similar ideas arise for ultrasound transducer positioning [93].…”