Imaging the propagation of natural waves
Ultrafast ultrasound also allows for the imaging of shear waves propagating naturally in the body. For instance, the propagation of the pulse wave in the blood vessels has been successfully imaged (Figure 1.16) and linked to the biomechanical properties of the arterial wall (Fujikura et al., 2007; Couade et al., 2010; Konofagou et al., 2011). The electromechanical wave in the heart has also been imaged, and showed to be a promising tool for the detection of arrhythmias (Provost et al., 2011). This topic will be further discussed in chapter 5.
3D ultrafast ultrasound imaging
The developments of the technologies brought by 2D ultrafast imaging can be extended to 3D and significantly broaden their scope of application. For instance, both natural and shear waves propagation occur in 3D, but their quantitative assessment is based on 2D observations at the cost of simplifying assumptions. This is of crucial importance to detect mechanical anisotropy (Lee et al., 2012) or analysing cardiac contraction activation (Provost et al., 2013). Furthermore, extending ultrafast Doppler to 3D would allow for the complete and qualitative angiography of complex networks of blood vessels, such as brain or tumours.
Unfortunately, as discussed previously, 3D conventional imaging is limited by its low framerate. 3D ultrafast ultrasound was recently developed based on the use of 2D matrix arrays (Jensen et al., 2013; Provost et al., 2014) and promising applications have been investigated. For instance, the 3D ultrafast ultrasound scanner developed in our group was used to perform 3D ultrafast Doppler imaging of the carotid bifurcation (Figure 1.17), 3D shear wave imaging, and 3D vector flow imaging (Provost et al., 2014; Correia et al., 2016, 2018). These studies used heavy equipment, as the matrix array had 1024 channels to be sampled. Thus, promising researches are made to reduce the number of channels of 3D probes, either by using sparsely populated (Roux et al., 2017) or row-column arrays (Rasmussen et al., 2015; Flesch et al., 2017).
Presentation of the 3D ultrafast ultrasound scanners and probes
The scanner used in this work was built-in in our group to drive 2D matrix array probes up to 1024 channels (Figure 2.1). Three different probes were used in this PhD:
3MHz, 32*32=1024 elements, 0.3mm pitch (Vermon, France).
8Hz, 32*32=1024 elements, 0.3mm pitch (Vermon, France).
10MHz, 16*16=256 elements, 0.3mm pitch (Imasonic, France).
The scanner is made of 4 ultrafast ultrasound scanner combined (Aixplorer, Supersonic Imaging, France) remotely controlled by a computer. Each Aixplorer has 256 channels in reception and 128 channels in emission, and controls one fourth of the elements of the 1024 probes. Thus, to record all 1024 channels, each emission were repeated twice and combined in post-processing. All the computing is made with Matlab and CUDA languages.
As we saw in the introduction chapter, ultrafast imaging is based on the emission of tilted planes or diverging waves. To emit waves, all channels of the probe send an ultrasonic pulse (usually one or two wavelength) which delays are chosen to control the wavefront accordingly. Each wave is defined as being emitted by a virtual source located behind the probe (Figure 2.2). The position of the virtual source defines directly the wavefront: the angulations are changed by translating the source on a plane parallel to the probe, and the plane or diverging shape depend on the distance from the virtual source to the transducers: if the source is close (a few centimetres), the wavefront is diverging, whereas if it is far (a few meters), the wavefront is planar. In practice, for each individual source, the user defines the x,y,z position of the virtual source, and the delays are computed accordingly by the driving algorithm. After the reception and sampling of the echoes, the volume is reconstructed by a conventional delay-and-sum beamformer and demodulated to IQ data. These steps are implemented on GPUs, an absolute requirement to achieve reasonable computation times. For instance, in the case of plane wave imaging, the formation of a volume of 10*10*100mm with voxel size of 0.3*0.3*0.25mm takes approximately 80ms. In the case of diverging waves the beamforming calculations are based on a spherical coordinate system, thus an additional calculation step is necessary to obtain Cartesian coordinates and volumes. Because of the memory available on GPUs is limited, this conversion is computed by the CPU.
Saddle imaging: 2D imaging with a 3D probe
The first point is that the clinicians needs only 2D images to position correctly the probe. Thus, there is no need to beamform a full 3D volume but rather just one 2D slice. Moreover, once the probe correctly positioned, there is no further real-time imaging during the ultrafast acquisition. Hence, the acquisition protocol has been defined as 1) 2D real-time imaging until 2) the ultrafast 3D acquisition is launched.
To enhance the quality of the 2D-real-time images, a specific ultrasound sequence was designed to mimic a regular 2D imaging probe. Let’s define the imaging plane as the YZ slice at X=0mm, i.e. the centre of the probe (Figure 2.4). Then the transmitted beam is: focalized at X=0mm in the XZ plane, to maximize the energy sent in the imaging plane plane or diverging in the YZ plane, to perform ultrafast imaging.
By doing so, all the elements of the probe contribute to insonify the imaged area, and tilted plane or diverging waves can be combined to enhance the image. An illustration of the delays laws used to transmit the ultrasonic beam is presented Figure 2.3: as the representation looks like a saddle, this technique has been named “Saddle Imaging”. Moreover, a simulation of the pressure field obtained is visible on Figure 2.4. Finally, a comparison of the images obtained on a phantom using 3D imaging and the 2D saddle imaging using 40 diverging waves in each case is presented on Figure 2.5. Both contrast and resolution are enhanced when using saddle imaging, for the following reasons. Firstly, energy is focused in the imaging plane, rather than being spread in the whole volume. Secondly, as the transmitted beam is already focused along one dimension (here, X), all the plane/diverging waves sent are used for coherent compounding along the other dimension (Y), which improves the quality of the synthetic focusing in each pixel of the 2D image.
Magnetic Resonance Diffusion Tensor Imaging (MR-DTI)
Magnetic Resonance Diffusion Tensor Imaging (MR-DTI) is a technique based on the tracking of the water diffusion in soft tissues using an MR scanner. In fibrous tissues, the diffusion of water is more important along the fibres than perpendicularly. To estimate fibres orientation, MR-DTI computes the eigenvectors and eigenvalues of the water diffusion tensor in each voxel of the volume imaged (Hsu et al., 1998; Scollan et al., 1998). To represent the diffusion tensor, ellipsoids are often used: the direction of their axis are linked to the eigenvectors, and the radii to the eigenvalues (see figure 3.3a). However, to visualize the anisotropy and its direction in volumes containing a large number of voxels, the ellipsoidal representation is not convenient. In such cases, a solution is to encode the fibres orientation in colours (figure 3.3b). Finally, another possibility commonly used is the fibre-tracking representation, or tractography. In this case, coherent patterns of anisotropy orientation are followed voxel-wise and represented by fibres, which are also colour coded (figure 3.3c). Moreover, the degree of anisotropy of the tissues is often represented by the fractional anisotropy (FA) coefficient (Tseng et al., 1999), which is computed from the eigenvalues of the diffusion tensor. The FA is a normalized value ranging from 0 for isotropic tissues to 1 “perfectly anisotropic” tissues, i.e. the water diffusion occurring only in one direction.
Spatial coherence of ultrasonic backscattered echoes
Another way to assess the anisotropy of tissues is to analyse the spatial coherence of backscattered echoes received on the probe. Indeed, as we will show in this chapter, the spatial coherence of echoes contains information about the microstructure of scatterers in the focal zone. This concept relies greatly on the Van Cittert-Zernike (VCZ) theorem, which was introduced in optics to describe the propagation of the spatial coherence of a wave field produced by an incoherent source, and extended to ultrasounds by Mallart and Fink (Mallart and Fink, 1991). The principle is the following: if a coherent, focused ultrasound wave is emitted by a probe in a medium and echoes are backscattered, then from another point a view, one can interpret the focal zone as an incoherent ultrasound source. The statistical properties of this virtual source depends directly on those of the scattering medium: if the distribution of scatterers is random in the medium, the source is incoherent. Thus, in anisotropic tissues, the spatial coherence of backscattered echoes will be anisotropic too.
Table of contents :
Table of contents
: Introduction Chapter
1.2. Anatomy of the human heart
1.3. Clinical tools for cardiologists
1.3.2. Magnetic Resonance Imaging
1.3.4. Nuclear medicine
1.3.6. Current limits of clinical tools
1.4. Ultrafast ultrasound imaging
1.4.1. Plane waves, diverging waves and coherent compounding
1.4.3. Ultrafast Doppler
1.4.4. Imaging the propagation of natural waves
1.4.5. 3D ultrafast ultrasound imaging
1.5. Thesis objectives
1.6. Chapter bibliography
: 3D Ultrafast Ultrasound Imaging
2.2. Presentation of the 3D ultrafast ultrasound scanners and probes
2.3. Development of specific tools
2.3.1. Real-time imaging
2.3.2. Visualisation tools
2.5. Chapter bibliography
: 3D Backscatter Tensor Imaging
3.1.2. State of the art
3.2. 3D Mapping of cardiac fibres orientation
3.2.1. Material and methods
3.3. Correction of aberrations due to movement
3.3.1. Material and methods
3.5. Chapter bibliography
: 3D passive elastography of the left ventricle
4.1.1. Motivations: why myocardial stiffness in interesting
4.1.2. State of the art: how to measure (myocardial) stiffness
4.1.3. Objectives: non-invasive passive elastography
4.2. Material and methods
4.2.2. Ultrafast Ultrasound Imaging in-vivo
4.2.3. Post processing computation
4.3.2. In-vivo experiments
4.6. Chapter Bibliography
: 3D Ultrafast Imaging of Myocardial Contraction Activation
5.2. State of the art, motivation, objectives
5.2.1. Electrophysiology of the heart and electrocardiogram
5.2.2. Imaging the cardiac electrophysiology
5.2.3. Imaging the cardiac contraction
5.3. 3D mapping
5.3.1. Material and methods
5.4. Toward clinical applications
5.4.1. Material and methods
5.6. Chapter Bibliography