To save content items to your account,
please confirm that you agree to abide by our usage policies.
If this is the first time you use this feature, you will be asked to authorise Cambridge Core to connect with your account.
Find out more about saving content to .
To save content items to your Kindle, first ensure no-reply@cambridge.org
is added to your Approved Personal Document E-mail List under your Personal Document Settings
on the Manage Your Content and Devices page of your Amazon account. Then enter the ‘name’ part
of your Kindle email address below.
Find out more about saving to your Kindle.
Note you can select to save to either the @free.kindle.com or @kindle.com variations.
‘@free.kindle.com’ emails are free but can only be saved to your device when it is connected to wi-fi.
‘@kindle.com’ emails can be delivered even when you are not connected to wi-fi, but note that service fees apply.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
In this chapter, we review biomedical applications and breakthroughs via leveraging algorithm unrolling, an important technique that bridges between traditional iterative algorithms and modern deep learning techniques. To provide context, we start by tracing the origin of algorithm unrolling and providing a comprehensive tutorial on how to unroll iterative algorithms into deep networks. We then extensively cover algorithm unrolling in a wide variety of biomedical imaging modalities and delve into several representative recent works in detail. Indeed, there is a rich history of iterative algorithms for biomedical image synthesis, which makes the field ripe for unrolling techniques. In addition, we put algorithm unrolling into a broad perspective, in order to understand why it is particularly effective, and discuss recent trends. Finally, we conclude the chapter by discussing open challenges and suggesting future research directions.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
From unconditional synthesis to conditional synthesis, GANs have shown that they are a perfect fit for such problems thanks to their ability to learn probability distributions. For unconditional synthesis, the objective is to stochastically generate MR images of target contrast. Conditional synthesis refers to the case where the model learns nonlinear mapping to the different MR tissue contrasts without altering the physiological information. Furthermore, by merging collaborative information of multiple contrast images, missing data imputation among many different domains is also effectively solved with GANs. Although promising results are seen, development in the area is still at its early stage. Interesting research directions are proposed from prior work, including the application of more advanced methods and rigorous validation in clinical settings. In effect, MRI image synthesis techniques should be able to reduce the burden of costly MR scans, benefiting both patients and hospitals.
The choking condition of a one-dimensional steady-state diabatic flow, with wall friction, of a perfect gas through a constant cross-section pipe has been analysed by applying a recent analytical solution. Such a choking condition can be achieved for an initially subsonic flow and a supersonic one, even when the heat flux goes from the fluid to the wall. Entropy–enthalpy diagrams have been analysed, and a universal choking condition for compressible diabatic flows with wall friction has been determined. The analytical solution of a supersonic flow, in which a normal shock occurs, has been obtained for a diabatic flow with wall friction and compared with the results of a numerical model.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
The development of deep learning reconstruction methods for accelerated MR acquisitions has been an ongoing area of research for the last several years. It has been repeatedly demonstrated that deep learning methods can outperform classic reconstruction approaches in terms of both quantitative image metrics like MSE to ground truth as well as qualitative reader studies where radiologists have been questioned in a subjective way. We present the basics and well-known approaches for MR image reconstruction via deep learning.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
In this chapter, we review largely targeted tasks in the computed tomography (CT) literature, including low-dose CT, sparse-view CT, limited angle CT, interior CT, etc. We present deep-learning-based methods which operate as image post-processing techniques or raw-to-image mapping techniques.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
This chapter provides a summary of some popular model-based deep learning methods and their extensions. Section 8.1 briefly describes classical model-based methods and their benefit as well as limitations. Section 8.2 describes how deep learning can help in overcoming some limitations of classical model-based methods. Section 8.3 discusses how to incorporate a pre-trained deep network as a regularizer using the plug-and-play approach. Section 8.4 describes end-to-end training using a model-based deep learning framework. This section also discusses some benefits and limitations of end-to-end training. Section 8.5 and 8.6 describe unsupervised model-based deep learning approaches when a clean training dataset is not available. Section 8.6 considers model mismatch issues as well as the joint design of acquisition and reconstruction frameworks.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
With the present demand for high-quality image reconstruction and signal extraction from less (e.g., unfocused or parallel) transmissions that facilitate fast imaging, and the push towards compact probes, modern ultrasound imaging leans heavily on innovations in powerful digital receive channel processing. Beamforming, the process of mapping received ultrasound echoes to the spatial image domain, naturally lies at the heart of the ultrasound image formation chain. In this chapter, we discuss why and when deep learning methods can play a compelling role in the digital beamforming pipeline, and then show how these data-driven systems can be leveraged for improved ultrasound image reconstruction.
We investigate the influence of rotation on the onset and saturation of the Faraday instability in a vertically oscillating two-layer miscible fluid using a theoretical model and direct numerical simulations (DNS). Our analytical approach utilizes Floquet analysis to solve a set of the Mathieu equations obtained from the linear stability analysis. The solution of the Mathieu equations comprises stable and harmonic, and subharmonic unstable regions in a three-dimensional stability diagram. We find that the Coriolis force delays the onset of the subharmonic instability responsible for the growth of the mixing zone size at lower forcing amplitudes. However, at higher forcing amplitudes, the flow is energetic enough to mitigate the instability delaying effect of rotation, and the evolution of the mixing zone size is similar in both rotating and non-rotating environments. These results are corroborated by DNS at different Coriolis frequencies and forcing amplitudes. We also observe that for $(\,f/\omega )^2<0.25$, where $f$ is the Coriolis frequency, and $\omega$ is the forcing frequency, the instability and the turbulent mixing zone size-$L$ saturates. When $(\,f/\omega )^2\geq 0.25$, the turbulent mixing zone size-$L$ never saturates and continues to grow.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
This chapter focuses on biomedical image reconstruction methods at the intersection of MBIR and machine learning. After briefly reviewing classical MBIR methods for image reconstruction, we discuss the combination of MBIR with unsupervised learning, supervised learning, or both. Such combinations offer potential advantages for learning even with limited data.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
In this chapter, we show that image-domain deep-learning-only reconstruction methods have intrinsic limitations in reconstruction accuracy and generalizability to individual patients owing to the regressive nature of the method. The combination of deep learning methods with analytic reconstruction methods or statistical IR methods offers a promising opportunity to achieve personalized reconstruction with improved reconstruction accuracy and enhanced generalizability.
Edited by
Jong Chul Ye, Korea Advanced Institute of Science and Technology (KAIST),Yonina C. Eldar, Weizmann Institute of Science, Israel,Michael Unser, École Polytechnique Fédérale de Lausanne
We introduce a projection-based model reduction method that systematically accounts for nonlinear interactions between the resolved and unresolved scales of the flow in a low-dimensional dynamical systems model. The proposed method uses a separation of time scales between the resolved and subscale variables to derive a reduced-order model with cubic closure terms for the truncated modes, generalizing the classic Stuart–Landau equation. The leading-order cubic terms are determined by averaging out fast variables through a perturbation series approximation of the action of a stochastic Koopman operator. We show analytically that this multiscale closure model can capture both the effects of mean-flow deformation and the energy cascade before demonstrating improved stability and accuracy in models of chaotic lid-driven cavity flow and vortex pairing in a mixing layer. This approach to closure modelling establishes a general theory for the origin and role of cubic nonlinearities in low-dimensional models of incompressible flows.
Inviscid, incompressible liquid is released from rest by a sudden dam break, accelerating under gravity over a uniformly sloping impermeable plane bed. The liquid flows downhill or up a beach. A linearised model is derived from Euler's equations for the early stage of motion, of duration $2\sqrt {H/g}$, where H is the depth scale and $g$ is the acceleration due to gravity. Initial pressure and acceleration fields are calculated in closed form, first for an isosceles right-angled triangle on a slope of $45^{\circ }$. Second, the triangle belongs to a class of finite-domain solutions with a curved front face. Third, an unbounded domain is treated, with a curved face resembling a steep-fronted breaking water wave flowing up a beach. The fluid goes uphill due to a nearshore pressure gradient. In all cases the free-surface-bed contact point is the most accelerated particle, exceeding the acceleration due to gravity. Physical consequences are discussed, and the pressure approximation of shallow water theory is found poor during this early stage, near the steep free surface exposed by a dam break.
Current live-cell imaging techniques make possible the observation of live events and the acquisition of large datasets to characterize the different parameters of the visualized events. They provide new insights into the dynamics of biological processes with unprecedented spatial and temporal resolutions. Here we describe the implementation and application of a new tool called TrackAnalyzer, accessible from Fiji and ImageJ. Our tool allows running semi-automated single-particle tracking (SPT) and subsequent motion classification, as well as quantitative analysis of diffusion and intensity for selected tracks relying on the graphical user interface (GUI) for large sets of temporal images (X–Y–T or X–Y–C–T dimensions). TrackAnalyzer also allows 3D visualization of the results as overlays of either spots, cells or end-tracks over time, along with corresponding feature extraction and further classification according to user criteria. Our analysis workflow automates the following steps: (1) spot or cell detection and filtering, (2) construction of tracks, (3) track classification and analysis (diffusion and chemotaxis), and (4) detailed analysis and visualization of all the outputs along the pipeline. All these analyses are automated and can be run in batch mode for a set of similar acquisitions.
An equation for the evolution of mean kinetic energy, $E$, in a two-dimensional (2-D) or 3-D Rayleigh–Bénard system with domain height $L$ is derived. Assuming classical Nusselt-number scaling, $Nu \sim Ra^{1/3}$, and that mean enstrophy, in the absence of a downscale energy cascade, scales as $\sim E/L^2$, we find that the Reynolds number scales as $Re \sim Pr^{-1}Ra^{2/3}$ in the 2-D system, where $Ra$ is the Rayleigh number and $Pr$ the Prandtl number. Using the evolution equation and the Reynolds-number scaling, it is shown that $\tilde {\tau } \gtrsim Pr^{-1/2}Ra^{1/2}$, where $\tilde {\tau }$ is the non-dimensional convergence time scale. For the 3-D system, we make the estimate $\tilde {\tau } \gtrsim Ra^{1/6}$ for $Pr = 1$. It is estimated that the total computational cost of reaching the high $Ra$ limit in a simulation is comparable between two and three dimensions. The predictions are compared with data from direct numerical simulations.
This paper investigates the nonlinear evolution and acoustic radiation of coherent structures (CS) in the near-nozzle region of a subsonic turbulent circular jet. A CS is taken to be a wavepacket consisting of multiple ring/helical modes, which are considered to be inviscid instability waves supported by the mean-flow profile. As the three-dimensionality of helical modes is weak in the near-nozzle region, the ring and helical modes with the same frequency have nearly the same growth rates and critical levels. They coexist and interact with each other in their common critical layer at high Reynolds numbers. The self and mutual quadratic interactions generate a mean-flow distortion and streaks, which act back on the fundamental components through the cubic interaction. The amplitude of the CS is governed by an integro-partial-differential equation, a significant feature of which is that differentiations with respect to the azimuthal coordinate appear in the history-dependent nonlinear terms. The non-parallelism of the mean flow as well as the impact of fine-scale turbulence on CS are taken into account and found to affect the nonlinear terms. By solving the amplitude equation, the development of the constituting modes, streamwise vortices and streaks are described. For CS consisting of frequency sideband, low-frequency components are excited nonlinearly and amplify to reach a considerable level. By analysing the large-distance asymptote of the perturbation, the low-frequency acoustic waves are found to be emitted by the temporally–spatially varying mean-flow distortion and streaks generated by the nonlinear interactions of the CS, and are thereby determined on the basis of first principles. Interestingly, the energetic part of the streaky structure that contributes to the nonlinear dynamics does not radiate directly, and instead the Reynolds stresses driving the subdominant radiating components represent the true physical sources.
This paper presents an integrated power amplifier (PA) following the orthogonal load-modulated balanced amplifier (OLMBA) topology. The fixed-phase prototype in this paper is implemented with 22 nm complementary metal oxide semiconductor (CMOS) fully depleted silicon-on-insulator (FDSOI) process. The proposed PA operates at 26 GHz frequency range, where it achieves 19.5 dBm output power, 16.6 dB gain, 15.7% power added efficiency, and 18.3 dBm output 1-dB compression point ($P_{\rm 1\,dB}$). The PA is also tested with high dynamic range modulated signals, and it achieves, respectively, 11.4 dBm and 4.9 dBm average output power (Pavg) with 100 MHz and 400 MHz 64-QAM third-generation partnership project/new radio frequency range 2 signals, and 14 dBm Pavg with 0.6 Gb/s (120 MHz) single carrier 64-QAM signal, measured at 26 GHz and using −28 dBc adjacent channel leakage ratio and −21.9 dB (8%) error vector magnitude as threshold values. The proposed OLMBA is also compared to a stand-alone quadrature-balanced PA. Modulated measurements show that the stand-alone quadrature-balanced PA has better linearity in deep back-off, but the OLMBA has better efficiency.
We present a theoretical study of viscous slug motion inside a microscopically rough capillary tube, where pronounced stick–slip motion can emerge at slow displacement rates. The mathematical description of this intermittent motion can be reduced to a system of ordinary differential equations, which also describe the motion of a pendulum inside a fluid-filled rotating drum. We use this analogy to show that the stick–slip motion transitions to steady sliding at high displacement rates. We characterize this crossover with a simple scaling relation and show that the crossover is accompanied by a shift in the dominant energy dissipation mechanisms within the system.