In vivo Adaptive Focusing for Clinical Contrast - Enhanced Transcranial Ultrasound Imaging in Human

2025-05-06 0 0 2.12MB 16 页 10玖币

侵权投诉

In vivo Adaptive Focusing for Clinical Contrast-

Enhanced Transcranial Ultrasound Imaging in

Human

Justine Robin*1,2, Charlie Demené*1,2, Baptiste Heiles1, Victor Blanvillain1, Liene Puke2,

Fabienne Perren-Landis†2, Mickael Tanter†1

1 Physics for Medicine Paris, Inserm, ESPCI Paris, PSL Research University, CNRS

2 Neurocenter of Geneva, LUNIC Laboratory, University of Geneva, Switzerland

*Contributed equally to this work

† Contributed equally to this work

Email:

justine.robin@espci.fr

Abstract

Imaging the human brain vasculature with high spatial and temporal resolution remains

challenging in the clinic today. Transcranial ultrasound is scarcely used for cerebrovascular

imaging, due to low sensitivity and strong phase aberrations induced by the skull bone that

only enable major brain vessel imaging, even with ultrasound contrast agent injection

(microbubbles). Here, we propose an adaptive aberration correction technique for skull bone

aberrations based on the backscattered signals coming from intravenously injected

microbubbles. Our aberration correction technique was implemented to image brain

vasculature in adult humans through temporal and occipital bone windows. For each patient,

an effective speed of sound, as well as a phase aberration profile, were determined in several

isoplanatic patches spread across the image. This information was then used in the

beamforming process. It improved image quality both for ultrafast Doppler imaging and

Ultrasound Localization Microscopy (ULM), especially in cases of thick bone windows. For

ultrafast Doppler images, the contrast was increased by 4dB on average, and for ULM, the

number of detected microbubble tracks was increased by 38%. This technique is thus

promising for better diagnosis and follow-up of brain pathologies such as aneurysms or stroke

and could make transcranial ultrasound imaging possible even in particularly difficult-to-

image patients.

Keywords: Transcranial, super resolution, adaptive focusing, phase aberration correction, clinical brain imaging,

microbubbles

1. Introduction

Imaging the human brain vasculature with high enough spatial

and temporal resolutions to detect small vessels and monitor

their hemodynamics is still a challenge today in the clinic.

Accessing this information would yet be of great interest for

the diagnostic, follow-up, and better understanding of various

cerebral pathologies such as aneurysms, or stroke-related

events, but also in several degenerative pathologies such as

Alzheimer’s disease [1]–[3]. The gold standard imaging

methods generally require contrast injections, and ionizing

(CT) or expensive (MRI) imaging devices. They provide

resolutions in the 0.4-0.34 mm range [4], [5], when the small

vessel diameters are in the 10 µm range. Ultrasound is

traditionally scarcely used in neuroimaging due to its very

limited sensitivity and resolution at frequencies passing

through the skull bone. Transcranial Color Doppler (TCCD)

Ultrasound thus provides low-resolution images, where only

the proximal segments of the cerebral basal arteries (circle of

Willis and afferent arteries) can be depicted [6].

Increasing the sensitivity of ultrasound to blood flow has been

an extensive field of research over the last 20 years [7],

entailing ultrasonic probe developments, modification of the

ultrasonic emission schemes [8], and refining of the signal

processing chain [9]. An interesting setting for ultrasensitive

blood flow imaging combines Ultrafast imaging and

spatiotemporal Singular Value Decomposition (SVD) [10],

with proven efficiency for cerebrovascular imaging [11].

More recently, ultrafast imaging combined with a contrast

agent was proved able to beat by almost two orders of

magnitude the resolution limit of conventional Ultrasound, a

technique named Ultrasound Localization Microscopy (ULM)

[12]–[15]. Inspired by super resolution optical microscopy

techniques [16], the idea is to use microbubbles, a widely used

ultrasound contrast agent, as strong point-like scatterers

distributed in the vasculature [17], [18]. These microbubbles

indeed have a typical 2-3 µm mean diameter but are yet very

echogenic thanks to the large acoustic impedance mismatch

between gas and liquid. Imaged by ultrasound at thousands of

frames per second, they can thus be individually localized and

followed in time from one frame to the next, providing

vasculature maps resolving structures as small as 9 µm, highly

surpassing the ultrasound diffraction limit.

If these methods have successfully been used in rodents,

their translation to the clinic remains challenging, due to the

much higher imaging depth required, and the huge obstacle of

the skull bone. Bones indeed exhibit much higher density

( ) and sound speed

( ) than soft tissues such as skin,

muscle, or brain ( ,

 ), which leads to a high acoustic

impedance mismatch and poor acoustic transmission at the

skin/bone and bone/dura mater interfaces. Furthermore, the

acoustic attenuation coefficient inside the skull bone is one of

the highest in the human body (~30 dB/cm at 2 MHz), which

means that imaging is only feasible through the thinnest point,

the temporal window. Finally, the skull biases the image

reconstruction by modifying the speed of sound on a section

of the propagation medium, and the spatial heterogeneities of

the skull bone width and sound speed lead to phase and

amplitude distortions in the transmitted and received

wavefronts, heavily affecting image quality. To overcome this

problem, a large number of phase aberration correction

techniques have been developed [19], [20]. They usually

model the skull as a thin phase – or a phase and amplitude –

screen located in the near field of the transducer. The relative

temporal delays to apply to each transducer element are

calculated to correct the aberrations. Most techniques obtain

these delays using the correlation between signals received on

different elements. They then iteratively repeat the process to

increase the spatial coherence of the signals based on an

indicator such as the focus criterion [21] or speckle brightness

[22]. They can take advantage of a point-like scatterer if one

is present in the medium [23], [24], or use diffuse scatterers if

not. In particular, in the case of blood flow imaging, several

methods using moving scatterers have been developed [25],

[26]. Their implementation in the clinic however remains

challenging, and in the case of transcranial Doppler, is still

limited to the major brain vessels [27], [28].

In the particular case of transcranial ULM however, the

microbubbles not only compensate for the high attenuation of

the skull bone but also enable the correction of the strong skull

aberrations. Each micro-bubble can indeed be considered as a

point-like source conveniently placed behind the skull. This

microbubble can be used as a beacon for the recovery of the

skull bone aberration by providing an experimental estimation

of the Green’s function relating the microbubble position to

the piezoelectric elements of the ultrasonic array. The skull-

induced phase aberration profile, as well as the effective speed

of sound of the medium, can thus be derived by studying the

distortions in the wavefront coming from this beacon. This

technique, in a very preliminary form, was briefly mentioned

in the first proof of concept of transcranial ULM in human

[15]. In the present paper, we take this technique to an

accomplished level, giving extensive details about its

implementation, showing results both on contrast transcranial

Doppler and transcranial ULM images, and quantifying the

improvement on the images, for application in cerebro-

vascular imaging in human adults.

2. Materials and Methods

2..1 Clinical Protocol

All experiments strictly comply with the ethical principles

for medical research involving human subjects of the World

Medical Association Declaration of Helsinki. Healthy

volunteers and patients were recruited under the protocol

accepted by the CCER of Geneva (n° 2017-00353) and gave

informed and written consent. The dose of injected contrast

agent as well as the amplitude and duration of ultrasound

exposures were kept to the minimum enabling the ultrasound

localization microscopy to follow the ALARA (“As Low As

Reasonably Achievable”) principle. Ultrasound parameters

were well below the FDA recommendations (AUIM/NEMA

2004, Track 3) for ultrasound imaging, with a maximum

Mechanical Index (MI) of 0.46 (maximum FDA

recommended value is 1.9), a maximum derated Spatial Peak

Temporal Average Intensity (ISPTA) of 64.3mW/cm²

(maximum FDA recommended value 720mW/cm²), a

maximum Thermal Cranial Index (TIC) of 1.99 (FDA

regulations ask for explanations for values above 6). A widely

clinically used echo-contrast agent consisting of Sulphur

hexafluoride micro-bubbles with a mean diameter of 2.5 μm

and a mean terminal half-life of 12 min (SonoVue®, Bracco,

Italy) was injected intravenously via the cubital vein. Up to

three 0.1 mL boli of contrast agent were injected successively,

corresponding to a maximum bubble concentration of  to

 in the blood. The brain vasculature

of adult healthy volunteers (27 to 82 years old, age median:

79) was imaged transcranially through the temporal or

occipital bone windows.

2.2 Ultrasound acquisitions

A phased array ultrasonic probe (XP 5-1, pitch 0.2 mm, 96

elements, central frequency 2.93 MHz, 90 % bandwidth at -

6dB) (Vermon, Tours, France) and an ultrafast programmable

ultrasound scanner (Aixplorer® Supersonic Imagine, Aix-en-

Provence, France) were used for ultrafast ultrasound imaging.

The imaging sequence consisted of 4 successive diverging

waves originating from different virtual sources (regularly

spaced every 3.2 mm, placed 11.44 mm behind the

transducer), emitted at a frequency of 2 MHz and a PRF of

4800 Hz (compound framerate of 800 Hz). For each emission,

backscattered echoes were recorded by the transducer array,

digitized at 200% bandwidth (meaning 4 samples per

wavelength), and stored in a so-called radiofrequency (RF)

data matrix. A 1 s emission was repeated every 2 s, for a total

acquisition time of 90 s.

2.3 Image reconstruction and bubble localization

The images were reconstructed using delay-and-sum

beamforming with and without integration of the calculated

phase aberration law. The 3D matrix (   

) of the full stack of images was then filtered to extract

signals coming from the microbubbles. First, an SVD clutter

filtering was used to remove tissue signals [29]. Depending on

the level of tissue motion, between 25 and 50 singular vectors

out of 800 were removed following the method described in

[10]. Then, a binary mask was built based on the vesselness

filtering of this stack of images [30] (available on Mathworks

file exchange, ©Dirk-Jan Kroon 2009, and © Tim Jerman,

2017). In the (    ) 3D matrix, moving

bubbles will indeed appear as tubular (vessel-like) structures

and will therefore be enhanced. Both image stack and mask

stack were interpolated (Fourier Space based interpolation

(equivalent to an exact sinc interpolation) for the image

stack, nearest-neighbor interpolation for the binary mask

stack (the goal of this interpolation being keeping consistent

dimensions for the image stack and the binary mask stack) to

obtain a radial resolution of    and an angular

resolution of   . Local maxima were then detected

within the masked area for each frame. Small regions around

these local maxima were then correlated with the typical point

spread function of our imaging system (which is the response

of an isolated microbubble), and only maxima with strong

correlation (>0.6) were kept. Localization was further refined

at the sub-pixel level using a fast local (5x5 pixel

neighborhood) 2nd order polynomial fit, and their positions

were finally converted from polar to Cartesian coordinates.

The maxima positions were then tracked using a classical

particle tracking algorithm (simpletracker.m available on

Mathworks ©Jean-Yves Tinevez, 2019, wrapping Matlab

munkres algorithm implementation of ©Yi Cao 2009) with no

gap filling and maximal distance linking of 1 mm

(corresponding to a bubble maximum speed of 80 cm/s). To

reduce further the level of false bubble detection, bubble

tracks shorter than 10 frames were removed, based on the idea

that microbubbles traveling in the bloodstream should be

observed on several consecutive frames [31].

2.4 Aberration correction method

Using isolated bubbles as individual ultrasound point-like

sources placed directly inside the patient’s brain vasculature,

an iterative aberration correction procedure was developed to

both estimate the effective sound speed of the imaged medium

(average over skull and brain tissues in the propagation path)

and to determine the phase aberration profiles introduced by

the skull. Our first goal was to validate (or invalidate) the

hypothesis modeling the skull as a thin phase screen aberrator

– in which case the aberration law can be considered the same

over the whole image. Phase aberration laws were thus first

calculated using bubbles distributed over the whole image, to

identify potential isoplanatic patches within which the phase

aberration remained constant (within a patch, the aberration

law variation across bubbles is less than T/8, T being the

period of the ultrasound wave). In each of these patches, an

iterative procedure was then used, where phase aberrations

and effective sound speed were calculated in turn, until

convergence of the effective sound speed value (see Fig. 1).

Signal pre-processing

SVD clutter filtering was performed on the raw RF signals,

to remove tissue signals. Isolated bubble wavefronts appeared

visible in the filtered RF data (see Fig 2, step 1). Filtered RF

data were beamformed (spherical delay laws,

   ) and isolated bubbles were

located in the uncorrected image (as described above). Bubble

locations -   in polar coordinates - were stored and used

in the aberration correction procedure.

Fig. 1. Description of the aberration correction process. After identification of

the different isoplanatic patches, an effectivesound speed and a phase aberration

law are recovered in each of them.

Phase aberration law calculation

To limit the influence of potentially overlapping

wavefronts, only isolated microbubbles were used in the

following algorithm. Namely, a bubble should be at least 2mm

away from any other strong scatterer to be included. For each

considered bubble, the procedure can be decomposed in 3

steps, and repeated in 5 iterations (see Fig. 2):

1) The filtered RF signals from the 4 different

transmission events were delayed by time 



,

where 

 is the distance between the bubble location

and the virtual source s used to create each diverging

wave emission. This creates a virtual emission

focusing directly on the bubble. A directional filter is

applied to the virtually focused RF signals - in k-

space – to isolate the wavefronts specifically coming

from the bubble location from potential overlapping

wavefronts. This is done by computing the 2D

Fourier transform of the compounded RF data, which

gives a k- diagram, k being the wave vector and 

the angular frequency, and by setting to 0 a certain

range of orientations before computing the invers 2D

Fourier transform. This will select the waves

propagating in a certain direction (with a certain k

vector).

2) For each receiving sensor i signal is shifted by the

delay 

 , where  is the correction for

iteration n. The phase aberration profile is

determined by finding the maximum cross-

correlation of the signals between transducer

elements (see Section 2-5).

3) These delays are added to the focusing law used in

step 1, for the next iteration.

At each iteration, the spatial coherence function is

calculated in the filtered RF at the microbubble position, to

evaluate aberration correction performance (see Section 2.5

for details). Bubbles for which the integral of the spatial

coherence does not increase are discarded for the rest of the

process. An average phase aberration law is calculated for

each isoplanatic patch and used for the next sound speed

estimation step. Finally, at the end of the process, the strength

of aberration is evaluated for each acquisition as the root-

mean-square (RMS) of the calculated aberration profile. The

whole method is detailed in Fig. 2.

Effective sound speed estimation

In this step, different effective sound speeds varying from

   to     are screened. For

each sound speed considered:

1) The bubble positions -   in polar coordinates -

obtained from the uncorrected beamformed images are

corrected to account for the tested sound speed 

  

   .

2) The filtered RF signals from the 4 different diverging

wave emissions are recombined using the delay laws

corresponding to bubble location at the considered

sound speed, including the phase aberration calculated

in a previous step. A directional filter is applied to the

focused RF signals – in the k-space – to isolate the

wavefronts specifically coming from the bubble

location from potential overlapping wavefronts.

3) The spatial coherence function is calculated at the

bubble location and stored.

The spatial coherence functions obtained for each tested

sound speed are averaged over all the considered bubbles, and

the effective sound speed exhibiting the highest area under the

spatial coherence function curve is picked and used in the next

iteration. The algorithm stops when the same sound speed is

obtained in 2 consecutive iterations.

Corrected image reconstruction

At the end of the aberration correction protocol described

above, we have identified for each isoplanatic patch of the

image a couple (effective sound speed, phase aberration law).

This information is thus used to beamform as many corrected

images as isoplanatic patches. A final composite corrected

image is then reconstructed by combining all the corrected

patches.

2.5 Aberration correction evaluation

The robustness of our aberration correction method was

evaluated in several manners, to ensure that the calculated

phase aberration laws were meaningful. First, the convergence

of the algorithm was tested in terms of the number of iterations

needed to converge towards the final delay law. To do so, the

performance of the correction algorithm with an increasing

number of iterations was quantified directly on the RF signals.

For each considered bubble, the spatial coherence function

 - defined by Van Cittert and Zernike as the average

cross-correlation between signals received at two points of

space (here the transducer elements positions) [32] – was

calculated on the filtered and flattened RF data [33], [34]:

  

     





 





where m is the distance in transducer elements, N is the

number of elements in the transducer, and  is defined as:

  

 





,

with [T1, T2] a temporal window centered on the focal time,

and  is the time-delayed (flattened by subtraction of the

parabolic time delay law in the homogeneous medium

corresponding to the effective sound speed) RF signal

received on transducer .

The Van Cittert Zernike theorem indeed states that the

coherence function is the spatial Fourier transform of the

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

InvivoAdaptiveFocusingforClinicalContrast-EnhancedTranscranialUltrasoundImaginginHumanJustineRobin*1,2,CharlieDemené*1,2,BaptisteHeiles1,VictorBlanvillain1,LienePuke2,FabiennePerren-Landis†2,MickaelTanter†11PhysicsforMedicineParis,Inserm,ESPCIParis,PSLResearchUniversity,CNRS2NeurocenterofGeneva,LUNI...

展开>> 收起<<

In vivo Adaptive Focusing for Clinical Contrast - Enhanced Transcranial Ultrasound Imaging in Human.pdf

共16页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

In vivo Adaptive Focusing for Clinical Contrast - Enhanced Transcranial Ultrasound Imaging in Human

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: