Deep Neural Networks for Non-Linear Model-Based Ultrasound Reconstruction

Almansouri, Hani; Venkatakrishnan, Singanallur; Buzzard, Gregery T.; Bouman, Charles A.; Santos-Villalobos, Hector

doi:10.1109/globalsip.2018.8646704

Cited by 11 publications

(3 citation statements)

References 26 publications

(31 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Before we introduce the particular approach we follow in this work in more detail, we remark that as in many fields, there has been a recent flurry of interest into the question of whether deep neural networks can be utilized for ultrasonic imaging [89], see, e.g., [21,4,29,30,61,28,31,32,55,56,100,89] for approaches to improve the speed and accuracy of 2D UST reconstructions. While it seems likely that neural networks will find wide application in UST, there are reasons to think that, as in other imaging modalities, they will complement rather than supersede existing approaches [6].…”

Section: Image Reconstruction For Ultrasound Tomographymentioning

confidence: 99%

High resolution 3D ultrasonic breast imaging by time-domain full waveform inversion

et al. 2021

View full text Add to dashboard Cite

Ultrasound tomography (UST) scanners allow quantitative images of the human breast's acoustic properties to be derived with potential applications in screening, diagnosis and therapy planning. Time domain full waveform inversion (TD-FWI) is a promising UST image formation technique that fits the parameter fields of a wave physics model by gradient-based optimization. For high resolution 3D UST, it holds three key challenges: Firstly, its central building block, the computation of the gradient for a single US measurement, has a restrictively large memory footprint. Secondly, this building block needs to be computed for each of the 103-104 measurements, resulting in a massive parallel computation usually performed on large computational clusters for days. Lastly, the structure of the underlying optimization problem may result in slow progression of the solver and convergence to a local minimum. In this work, we design and evaluate a comprehensive computational strategy to overcome these challenges: Firstly, we exploit a gradient computation based on time reversal that dramatically reduces the memory footprint at the expense of one additional wave simulation per source. Secondly, we break the dependence on the number of measurements by using source encoding (SE) to compute stochastic gradient estimates. Also we describe a more accurate, TD-specific SE technique with a finer variance control and use a state-of-the-art stochastic LBFGS method. Lastly, we design an efficient TD multi-grid scheme together with preconditioning to speed up the convergence while avoiding local minima. All components are evaluated in extensive numerical proof-of-concept studies simulating a bowl-shaped 3D UST breast scanner prototype. Finally, we demonstrate that their combination allows us to obtain an accurate 442x442x222 voxel image with a resolution of 0.5mm using Matlab on a single GPU within 24 hours.

show abstract

Section: Image Reconstruction For Ultrasound Tomographymentioning

confidence: 99%

High resolution 3D ultrasonic breast imaging by time-domain full waveform inversion

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Before we introduce the particular approach we follow in this work in more detail, we remark that as in many fields, there has been a recent flurry of interest into the question of whether deep neural networks can be utilized for ultrasonic imaging [83], see, e.g., [19,4,27,28,59,26,29,30,53,54,93,83] for approaches to improve the speed and accuracy of 2D UST reconstructions. While it seems likely that neural networks will find wide application in UST, there are reasons to think that, as in other imaging modalities, they will complement rather than supersede existing approaches [6].…”

Section: Image Reconstruction For Ultrasound Tomographymentioning

confidence: 99%

High Resolution 3D Ultrasonic Breast Imaging by Time-Domain Full Waveform Inversion

Lucka,

Pérez-Liva,

Treeby

et al. 2021

Preprint

View full text Add to dashboard Cite

Ultrasound tomography (UST) scanners allow quantitative images of the human breast's acoustic properties to be derived with potential applications in screening, diagnosis and therapy planning. Time domain full waveform inversion (TD-FWI) is a promising UST image formation technique that fits the parameter fields of a wave physics model by gradient-based optimization. For high resolution 3D UST, it holds three key challenges: Firstly, its central building block, the computation of the gradient for a single US measurement, has a restrictively large memory footprint. Secondly, this building block needs to be computed for each of the 10 3 − 10 4 measurements, resulting in a massive parallel computation usually performed on large computational clusters for days. Lastly, the structure of the underlying optimization problem may result in slow progression of the solver and convergence to a local minimum. In this work, we design and evaluate a comprehensive computational strategy to overcome these challenges: Firstly, we introduce a novel gradient computation based on time reversal that dramatically reduces the memory footprint at the expense of one additional wave simulation per source. Secondly, we break the dependence on the number of measurements by using source encoding (SE) to compute stochastic gradient estimates. Also we describe a more accurate, TD-specific SE technique with a finer variance control and use a state-of-the-art stochastic LBFGS method. Lastly, we design an efficient TD multi-grid scheme together with preconditioning to speed up the convergence while avoiding local minima. All components are evaluated in extensive numerical proof-of-concept studies simulating a bowl-shaped 3D UST breast scanner prototype. Finally, we demonstrate that their combination allows us to obtain an accurate 442 × 442 × 222 voxel image with a resolution of 0.5 mm using Matlab on a single GPU within 24 hours.

show abstract

“…By training the network on pairs of sub-Nyquist and full Nyquist-rate data, the authorsshow that their approach enables a reduction in data-rates up to ∼ 88% without significantly compromising image quality.Model-based wavefield inversion using DLReconstruction techniques based on the inversion of a (non-)linear measurement model are often very computationally intensive, and require careful tuning of hyperparameters to ensure robust inference. Alternatively,Almansouri et al (2018b) propose a two-step approach which leverages a simple linear model to obtain an initial estimation, after which further refinement is done through a CNN. As such, the neural network can account for non-linear, and space-varying, artifacts in the measurement model.The ultrasound forward model is based on a set of differential equations, and mainly depends on three parameters: the acoustic velocity c 0 , the density ρ 0 , and the attenuation α 0 .…”

mentioning

confidence: 99%

Ultrasound Signal Processing: From Models to Deep Learning

Luijten¹,

Chennakeshava²,

Eldar³

et al. 2022

Preprint

View full text Add to dashboard Cite

Medical ultrasound imaging relies heavily on high-quality signal processing algorithms to provide reliable and interpretable image reconstructions. Hand-crafted reconstruction methods, often based on approximations of the underlying measurement model, are useful in practice, but notoriously fall behind in terms of image quality. More sophisticated solutions, based on statistical modelling, careful parameter tuning, or through increased model complexity, can be sensitive to different environments. Recently, deep learning based methods have gained popularity, which are optimized in a datadriven fashion. These model-agnostic methods often rely on generic model structures, and require vast training data to converge to a robust solution.A relatively new paradigm combines the power of the two: leveraging datadriven deep learning, as well as exploiting domain knowledge.

show abstract

Deep Neural Networks for Non-Linear Model-Based Ultrasound Reconstruction

Cited by 11 publications

References 26 publications

High resolution 3D ultrasonic breast imaging by time-domain full waveform inversion

High resolution 3D ultrasonic breast imaging by time-domain full waveform inversion

High Resolution 3D Ultrasonic Breast Imaging by Time-Domain Full Waveform Inversion

Ultrasound Signal Processing: From Models to Deep Learning

Contact Info

Product

Resources

About