Full-waveform inversion imaging of the human brain

Guasch, Lluís; Agudo, Òscar Calderón; Tang, Meng‐Xing; Nachev, Parashkev; Warner, Michael

doi:10.1101/809707

Cited by 35 publications

(48 citation statements)

References 40 publications

(90 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…FWI is similar to other inverse problems like brain-imaging (Guasch et al, 2020), shape optimization (Jameson et al, 1998), and even training a neural network. When training a neural network, the activations calculated when propagating forward along the network need to be stored in memory and used later during backpropagation.…”

Section: Fwi and Other Similar Problemsmentioning

confidence: 99%

Lossy Checkpoint Compression in Full Waveform Inversion

Kukreja¹,

Hückelheim²,

Louboutin³

et al. 2020

Preprint

View full text Add to dashboard Cite

Abstract. This paper proposes a new method that combines check- pointing methods with error-controlled lossy compression for large-scale high-performance Full-Waveform Inversion (FWI), an inverse problem commonly used in geophysical exploration. This combination can signif- icantly reduce data movement, allowing a reduction in run time as well as peak memory. In the Exascale computing era, frequent data transfer (e.g., memory bandwidth, PCIe bandwidth for GPUs, or network) is the performance bottleneck rather than the peak FLOPS of the processing unit. Like many other adjoint-based optimization problems, FWI is costly in terms of the number of floating-point operations, large memory foot- print during backpropagation, and data transfer overheads. Past work for adjoint methods has developed checkpointing methods that reduce the peak memory requirements during backpropagation at the cost of additional floating-point computations. Combining this traditional checkpointing with error-controlled lossy compression, we explore the three-way tradeoff between memory, precision, and time to solution. We investigate how approximation errors introduced by lossy compression of the forward solution impact the objective function gradient and final inverted solution. Empirical results from these numerical experiments indicate that high lossy-compression rates (compression factors ranging up to 100) have a relatively minor impact on convergence rates and the quality of the final solution.

show abstract

Section: Fwi and Other Similar Problemsmentioning

confidence: 99%

Lossy Checkpoint Compression in Full Waveform Inversion

Kukreja¹,

Hückelheim²,

Louboutin³

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Transcranial ultrasound (US) as a modality to image the brain is gaining interest as an alternative for expensive MRI scans or radiation based CT scans. Although started in the early forties by Dussik [1], it only revived recently by the application of full-waveform inversion (FWI) methods to image the soft tissue in the brain [2]. FWI methods are commonly used for seismic applications [3].…”

Section: Introductionmentioning

confidence: 99%

“…obtained from a CT scan) or the application of an Adaptive Waveform Inversion (AWI) method. With AWI, a starting model from the data is constructed and successively used as input for their conventional FWI method [2]. Here we test the applicability of Contrast Source Inversion (CSI) as a FWI method to reconstruct speedof-sound profiles of the brain and the skull [7], [8].…”

Section: Introductionmentioning

confidence: 99%

Ultrasound Imaging of the Brain using Full-Waveform Inversion

Taskin

Eikrem

Nævdal

et al. 2020

2020 IEEE International Ultrasonics Symposium (IUS)

View full text Add to dashboard Cite

Transcranial ultrasound has been used to image the brain since 1942. Currently, it is regaining interest and full-waveform inversion (FWI) methods are now employed to reconstruct speed-of-sound profiles of the brain. Many of these methods require a good starting model. Here, we test the applicability of contrast source inversion (CSI) as a FWI method to reconstruct two-dimensional speed-of-sound profiles of the soft brain tissue enclosed by the skull. The advantage of CSI is that it can handle large acoustic contrasts without the need for a good starting model. To test the performance of CSI, we first compute synthetic data. The resulting pressure field clearly shows a significant amount of multiple scattering caused by the skull that acts as a hard acoustic contrast. Next we invert the resulting synthetic data within the Born approximation as well as by applying CSI as a FWI method. The results clearly show that Born inversion can only image the soft brain tissue in the absence of the skull whereas it generates erroneous results when the skull is present. On the other hand, with CSI it is feasible to reconstruct both the skull and the soft brain tissue accurately. Importantly, as compared to other methods CSI does not require any a priori information about the contrast, a mask or a heterogeneous starting model to reconstruct the soft tissue enclosed by the skull.

show abstract

“…They build a forward model of a USCT based on a full-wave equation in a frequency or space domain, then iteratively solve the forward model and reconstruct the SSI. Waveform-based methods are commercially available, and can provide high quality image, but their computational costs are high [24][25][26], because they need to solve the fullwave equation. On the other hand, ray-based methods are faster and more stable [27], and the reconstructed SSIs can be used as the initial condition of iterations for waveform-based methods.…”

Section: Introductionmentioning

confidence: 99%

A Combined Regularization Method Using Prior Structural Information for Sound-Speed Image Reconstruction of Ultrasound Computed Tomography

et al. 2020

View full text Add to dashboard Cite

Ultrasound computed tomography (USCT) is a promising technique for breast imaging. It provides three modalities: echo image, sound speed image (SSI), and attenuation image. The echo image reveals structural details of the breast with high resolution but is not quantitative. The SSI is quantitative, but its resolution is generally lower than that of the echo image. The SSI reconstruction is an ill-posed problem, thus the Tikhonov or total variation (TV) regularization methods are generally used. Tikhonov regularization is stable, but it tends to smooth the SSI and results in low resolution. TV regularization can preserve the sharp edges and provide higher resolution, but it may result in staircase artifacts. To combine the advantages of Tikhonov and TV regularizations, this article proposes a combined regularization method using prior structural information from the echo image. The proposed method mainly includes three steps. First, the echo image is reconstructed using the synthetic aperture (SA) technique and then segmented into breast and water regions. The segmentation result is used as prior structural information. Second, the USCT sound propagation forward model was built. Finally, with a penalty term according to the prior structural information, the combined regularization method is used to reconstruct the SSI. Both simulation and ex vivo experiments were conducted for evaluation. In the simulation, the proposed method has a low root-mean-square-error (13.78 m/s), a high correlation coefficient (0.812), a high structural similarity (0.755), and a low standard deviation of water sound speed (2.33 m/s). In the ex vivo experiment, the method has a low standard deviation of water sound speed (1.44 and 3.13 m/s for small and large objects, respectively), a low contrast to noise ratio (3.09 and 5.08), and a high Dice coefficient (0.92 and 0.97). Thus, the proposed combined regularization method using prior structural information can improve reconstruction quality.

show abstract

Full-waveform inversion imaging of the human brain

Cited by 35 publications

References 40 publications

Lossy Checkpoint Compression in Full Waveform Inversion

Lossy Checkpoint Compression in Full Waveform Inversion

Ultrasound Imaging of the Brain using Full-Waveform Inversion

A Combined Regularization Method Using Prior Structural Information for Sound-Speed Image Reconstruction of Ultrasound Computed Tomography

Contact Info

Product

Resources

About