Method and apparatus for 3D ultrafast ultrasound localization microscopy imaging

Abstract

The present disclosure proposes a method and apparatus for imaging at a microscopic scale a region of an organ of a human or animal, allowing the use of a 2D subsampled ultrasound probe while providing 3D ULM images with high resolution usually obtained only with a full 2D array matrix, by sorting the detected microbubbles and keeping only the microbubbles corresponding to the spots belonging to the main lobe of the Point Spread Function (PSF) of the microbubble and removing the phantom microbubbles corresponding to the spots belonging to the side lobes. It is also possible to provide a 3D non-invasive imaging of the microvascular flow of the region of an organ, such as a brain in-vivo, with high resolution and high sensitivity.

Description

DescriptionTitle: Method and apparatus for 3D Ultrafast Ultrasound Localization Microscopy imagingTechnical FieldThe present disclosure relates to ultrasound imaging techniques, and in particular to methods and apparatuses for 3D Ultrafast Ultrasound Localization imaging with a large field view at microscopic scale in a human or animal.Background Art

[0001] Generally, ultrasounds may be generated from a probe comprising an array of a plurality of transducers able to generate individually ultrasound waves. These ultrasound waves may be directed toward a medium which may generate in response backscattered signals, which in turn may be recorded by the same array of transducers or a different array. From the backscattered signal, it may be possible to reconstruct a 2D or 3D image of the medium.

[0002] One type of imaging of high interest, particularly regarding spatial resolution and penetration depth, is Ultrasound Localization Microscopy (ULM), which has been described in particular by Errico et al. [Errico, C. et al. Ultrafast ultrasound localization microscopy for deep super-resolution vascular imaging. Nature 527, 499-+ (2015)] and Demene et al. [Demene, C. et al. Transcranial ultrafast ultrasound localization microscopy of brain vasculature in patients. Nat. Biomed. Eng 5, 219-228 (2021)]. It has already been applied, for example, on rat and human brain, to study its ability to detect or prevent stroke and to better understand brain diseases such as Alzheimer's and Parkinson’s.

[0003] The conventional ULM imaging method comprises the following main steps :- acquisition of ultrasound raw data from the region to be imaged;- beamforming of the data with delay-and-sum (DAS) to reconstruct 3D images;- application of a Singular Value Decomposition (SVD) on the 3D images to reject the tissue signal;- detection of microbubbles by detecting local maxima in each 3D image;- localization of the microbubbles in each 3D image;- tracking the trajectories of the microbubbles through successive frames to establish velocity profiles;- mapping of the accumulated localizations gathered over the series of frames to produce a super-resolved image of the vascular region far beyond the diffraction limit.

[0004] The step of the detection and localization of microbubbles is thus crucial to obtain high-quality images. Indeed, the super-resolved image is constructed by localizing each microbubble center separately and accumulating their positions to reveal microvascular structure.

[0005] By imaging isolated intravenously injected microbubbles (MB) circulating in blood vessels at ultrafast frame rates and localizing the center of their individual point spread function with a sub-resolution precision, it enables to map the microvascularization of organs with a resolution of tens of microns and blood flow dynamic velocities in 2D and 3D.

[0006] Compared to 2D, 3D imaging can offer volumetric coverage with super-resolution in the three dimensions. While 3D ultrasound imaging is becoming a powerful tool in medical field, their main drawback is the difficulty to image large 3D volume. Indeed, the capacity to image large 3D volume is directly related to the dimension of the array of transducers. In order to not lose in spatial resolution, it is necessary to use an array of transducers, wherein the size of the transducers does not exceed the wavelength of the ultrasound wave. This requirement leads to configurations of array which are not reachable or at too high cost with the current technology. For instance, a probe with an opening of 10cmx10cm (i.e. array of 10cmx10cm) and a size of transducer around 250 micrometers (corresponding to a central frequency of 6MHz) leads to an array of 160 000 transducers in order to keep a good resolution in 3D imaging and large field of view.

[0007] In order to overcome this problem of controlling very large numbers of probe elements as well as the large amount of data to be acquired, transmitted, stored and processed, instead of having small transducers filling in the grid of the 2D matrix array, the transducers are organized in rows and columns in order to reduce the number of channels. Another approach to reduce the number of channels is to reduce the number of active elements. Instead of using a complete 2D matrix array, the matrix is sparse following a specific pattern. The active elements may be distributed randomly or following a predetermined pattern. However, these configurations reduce the contrast of imaging.

[0008] While the row-column array technique or sparse matrix allows large field of view with a limited number of channels, the image quality is lowered in the case of ULM imaging. Indeed, as indicated above, the construction of ULM images is based on the position of the ultrasound contrast agents or microbubbles. These very small scatters are imaged much larger in the ultrasound image due to spreading by the imaging system, which only has a diffraction-limited resolution. Therefore, the response of an imaging system to an isolated microbubble is defined as a point spread function (PSF). The microbubbles are localized as the brightest local maxima with high correlation with the point spread function which maybe modelled as a Gaussian spot. Thus, the microbubble appears as a spot representing the center of the interpolated PSF.

[0009] An ideal PSF of an isolated MB is represented only by a single spot in an image. However, in case of using a subsampling ultrasound imaging device such as the RCA probe or the sparse matrix array probe, results show that reducing the number of channels causes distortions in the shape of the PSF and causes strong side lobes.

[0010] Figure 1A shows an example of the PSF of a microbubble imaged with a subsampling probe, in an image in the three directions. The image shows a main spot 12 surrounded by a plurality of secondary spots 11 which are less bright. The spot 12 at the center is formed by the main lobe of the PSF, corresponding to the microbubble. The secondary spots represent the side lobes of the PSF of the microbubble. Consequently, the presence of these side lobes in the PSF introduces the risk of mistaking these secondary spots for separated microbubbles and could lead to false microbubbles detection or phantom microbubbles in the image. The MB localization error may propagate along the successive steps of the ULM processing workflow and subsequently affects the ULM image quality. In particular, these incorrectly localized microbubbles, or false microbubbles may create false blood vessels 15 next to the real blood vessel 13 in the reconstruction of the ULM map as illustrated in Figure 1 B, reducing thus the accuracy of the vascular structure and hindering the velocity determination and quantification inaccurate.

[0011] Therefore, there exists a need for an improved method and device providing high accuracy in microbubbles detection and localization, allowing the use of 3D ultrasound localization microscopy with a reduced number of active elements and / or channels while keeping the quality of the 3D reconstructed vascular network image.Summary

[0012] To this end, the present disclosure proposes a method for 3D ultrasound localization microscopy (ULM) imaging at a microscopic scale a region of a vascular network of an organ, of a human or animal, said region containing microbubbles, the method comprising the following steps:(a) transmitting incident ultrasonic waves in the region by a 2D array of ultrasonic transducers and acquiring raw data from backscattered ultrasonic waves by said 2D array of transducers;(b) generating a series of successive 3D ultrasound images from said raw data;(c) filtering the 3D ultrasound images to remove tissue signal;(d) detecting and localizing a plurality of isolated spots in each filtered 3D ultrasound images, each detected and located spot being a candidate main spot formed by the main lobes ofthe PSF of a microbubble, said candidate main spot assigned to a potential microbubble (PMB) ;(e) detecting for each candidate main spot the candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot, said candidate secondary spot assigned to a potential phantom microbubble (PPMB);(f) assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the main spot corresponding to the microbubble according to a selection parameter;(g) sorting the detected and localized spots by keeping the main spots corresponding to the microbubbles and removing the secondary spots corresponding to the phantom microbubbles;(h) tracking the kept spots through successive 3D filtered ultrasound images;(i) generating a 3D ULM image by attributing for each voxel an intensity value given by the number of kept spots in each voxel.

[0013] The proposed method proposes an efficient approach to reduce the influence of the side lobes by estimating whether a detected spot in a 3D image corresponds to a real microbubble, aiming to eliminate the number of false microbubbles. The invention allows the obtention of 3D ULM images of higher quality by reducing the number of phantom vessels.

[0014] During the processing of generating ULM images, microbubbles are detected as the brightest local maxima. Among these detected microbubbles, a part of them is formed by the side lobes of the PSF of the microbubble. Before tracking the detected and localized microbubbles, they are sorted by identifying the real microbubble of belonging to the main lobe of the point spread function of the microbubble and the false or phantom microbubbles of belonging to the side lobes of the point spread function of the microbubble. The probability for each detected microbubble of belonging to the main lobe or side lobes of the Point Spread Function of the microbubble may be assessed by comparing the intensity value of the detected microbubble with the intensity value of the microbubbles at the theoretical locations of the side lobes associated to the microbubble to be assessed. This supplementary sorting step acts as a filter to filter out the phantom signal and retain microbubble signal only, allowing improving the quality of the 3D images.

[0015] The known solution uses the PSF correlation to remove noisy points from the filtered image, rather than removing phantom lobes directly from the position tables as proposed in the present disclosure. Indeed, the PSF correlation is determined around on a very small neighborhood around the main lobe which does not include the secondary lobes.

[0016] Therefore, instead of correcting the aberration in the generated images as proposed in the conventional imaging solution for improving the ultrasound image, the solution of the present disclosure proposes to improve the quality of ultrasound images by sorting the lobes before the construction of ultrasound images. The sorting step is based on the positions and amplitudes of lobes and on the image and correlation.

[0017] The following features, can be optionally implemented, separately or in combination one with the others:

[0018] In one or more embodiments, the selection parameter being the intensity value of spots, the step of assigning (f) may include:- comparing for each candidate main spot its intensity value with the intensity value of the identified candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot;- assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the real main spot corresponding to the microbubble if the intensity value of the candidate main spot is higher than that of the candidate secondary spots

[0019] In one or more embodiments, the selection parameter being the number of candidate secondary spots detected associated to the candidate main spot, the step of assigning (f) may include :- counting for each candidate main spot the number of candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot;- assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the real main spot corresponding to the microbubble if the number of candidate secondary spots detected corresponds to the number of side lobes of the point spread function (PSF) of the microbubble associated to said candidate main spot

[0020] In one or more embodiments, before the step of sorting (g), the method may further comprise:- storing the location of the main spots corresponding to the microbubbles and the location of the secondary spots corresponding to the phantom microbubbles;- correcting the location of the main spots by fitting the location of secondary spots to the theoretical secondary lobe location of a microbubble model.

[0021] In one or more embodiments, the step of detecting of the candidate secondary spots for each candidate main spot may comprise:- calculating for each candidate main spot the side lobes of the point spread function (PSF) of the microbubble associated to said candidate main spot;- detecting the spots at the location of the side lobes, said detected spots being the candidate secondary spots.

[0022] In one or more embodiments, the step of detecting of candidate secondary spots for each candidate main spot may comprise:- detecting for each candidate main spot the systematic presence of a set of spots surrounding said candidate spot over the successive 3D ultrasound images, said detected set of spots corresponding to the candidate secondary spots generated by the side lobes of the point spread function (PSF) of the microbubble associated to said candidate main spot.

[0023] In one or more embodiments, the point spread function (PSF) of the candidate main spot may be varied spatially.

[0024] In one or more embodiments, detecting and localizing a plurality of isolated spots in each 3D ultrasound image comprises:- detecting local maxima in each 3D ultrasound image;- determining the center position of each local maxima by fitting the voxel around the local maxima with a 3D paraboloid on the voxels around the local maxima.

[0025] In one or more embodiments, the method may further comprise a step of computing of the speed velocity components Vx, Vyand Vzalong respectively the three directions X, Y, Z for each kept spot to generate a 3D velocity map.

[0026] In one or more embodiments, the 2D array of ultrasonic transducers may be a rowcolumn array (RCA).

[0027] In one or more embodiments, the 2D array of ultrasonic transducers may be a sparse array having a reduced number of active elements being distributed randomly or according to a predetermined pattern.

[0028] In one or more embodiments, the least one vascular parameter may be extracted from the 3D ULM image, said at least one vascular parameter being chosen in the group comprising: vessel diameter, blood flow, blood velocity, blood volume, blood pressure, and any combination thereof.

[0029] The present disclosure also concerns an apparatus for 3D ultrasound localization microscopy (ULM) imaging at a microscopic scale in a region of a vascular network of an organ, of a human or animal, said apparatus including at least an ultrasound probe having a 2D array of ultrasonic transducers and a control system configured to:(a) transmit incident ultrasonic waves in the region though said 2D array of ultrasonic transducers and acquire raw data from backscattered ultrasonic waves through said 2D array of ultrasonic transducers;(b) generate a series of successive 3D ultrasound images from said raw data;(c) filter the 3D ultrasound images to remove tissue signal;(d) detect and localize a plurality of isolated spots in each filtered 3D ultrasound images, each detected and localized spot being a candidate main spot formed by the main lobes of the PSF of a microbubble, said candidate main spot assigned to a potential microbubble (PMB);(e) detect for each candidate main spot the candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot, said candidate secondary spot assigned to a potential phantom microbubble (PPMB);(f) assign the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the main spot corresponding to the microbubble according to a selection parameter;(g) sort the detected and localized spots by keeping the main spots corresponding to the microbubbles and removing the secondary spots corresponding to the phantom microbubbles;(h) track the kept spots through successive 3D filtered ultrasound images;(i) generate a 3D ULM image by attributing for each voxel an intensity value given by the number of kept spot in each voxel

[0030] In another aspect, it is proposed a computer software comprising instructions to implement at least a part of a method as defined here when the software is executed by a processor. In another aspect, it is proposed a computer-readable non-transient recording medium on which a software is registered to implement the method as defined here when the software is executed by a processor.Brief Description of Drawings

[0031] Other features, details and advantages will be shown in the following detailed description and on the figures, on which:Fig. 1

[0032] [Fig. 1] Figure 1A illustrates a B-Mode image showing a main spot representing a detected microbubble and a clout of secondary spots surrounding the main spot and Figure 1 B illustrates an example of a 3D ULM image showing a real vessel reconstructed from the real microbubbles and phantom vessels reconstructed from the phantom microbubbles.Fig. 2

[0033] [Fig. 2] Figure 2 is a schematic view illustrating an embodiment of an apparatus for 3D ULM imaging according to the present disclosure.Fig. 3

[0034] [Fig. 3] Figure 3 is a block diagram illustrating an embodiment of a method of obtaining a 3D ULM image with the apparatus of Fig 2.Fig. 4

[0035] [Fig. 4] Figure 4 illustrates the main steps of the raw data acquisition sequence and the filtered 3D compounded images processing with the apparatus of Figure 2.Fig. 5

[0036] [Fig. 5] Figure 5 illustrates the main steps of the 3D Ultrasound localization Microscopy processing from the filtered 3D compounded images with the apparatus of Figure 2.Fig. 6

[0037] [Fig. 6] Figure 6 illustrates a schematic view of a main spot (in dark) formed by a real microbubble and seven secondary spots (in grey) located at the theoretical position of the side lobes of the PSF of the microbubble.Fig. 7

[0038] [Fig. 7] Figure 7 illustrates a schematic view of the main spot (in dark) and its seven secondary spots (in grey) of figure 6 at two times t and t+1 in two successive frames.Fig. 8

[0039] [Fig. 8)] Figure 8 illustrates the result of the vessel reconstructed of figure 1 B after removing the secondary spots, showing the presence of one vessel.Description of Embodiments

[0040] In the Figures, the same references denote identical or similar elements.

[0041] Ultrafast Ultrasound Localization Microscopy (ULM) has already been applied in- vivo for imaging a vascular network in 2D, and in particular a cerebral network. Nevertheless, cerebral vessels are oriented in all three dimensions (3D) and 2D velocity estimates can be underestimated when the vessels are not perfectly aligned to the 2D slices. The present disclosure proposes a method and apparatus for imaging at a microscopic scale a region of an organ of a human or animal, by performing 3D ultrafast ultrasound localization microscopy (ULM). The 3D ULM imaging method of the present disclosure allows the use of a 2D subsampled ultrasound probe to provide 3D ULM images with a higher qualityusually obtained with a 2D array matrix, by sorting the detected microbubbles and keeping only the microbubbles belonging to the main lobe of the Point Spread Function (PSF) of the microbubble. It is also possible to provide a 3D non-invasive imaging of the microvascular flow of the region of an organ, such as a brain in-vivo.

[0042] 3D ULM images may be obtained based on methods already known in the art and explained in the above articles of Errico et al., 2015 and Demene et al., 2021.

[0043] An example of apparatus 10 configured to ultrafast 3D ultrasound imaging usable in performing the method according to the present disclosure, is shown in Figure 2. The apparatus may be adapted to image a brain or a part of the brain of a living being or may be a heart or a part of the heart of the living being.

[0044] The apparatus may include for instance at least a 2D array ultrasonic probe 2 and a control system.

[0045] The control system may for instance include a control unit 3 and a computer 4. In the example of Figure 2, the control unit 3 is configured to control a 2D array ultrasonic probe 2 to transmit et acquire signals therefrom, and an ultrasound contrast agent injection device 5. The computer 4 is configured to control the control unit 3 et to generate 3D ULM images from the signals acquired by the control unit 3 and determine quantification parameters from the 3D ULM images.

[0046] The ultrasound contrast agent injection device 5 may comprise ultrasound contrast agents. The ultrasound contrast agents may be microbubbles, as for instance described by Dayton et al., [Dayton, PA et al. Molecular ultrasound imaging using microbubble contrast agent. Frontiers in Bioscience 12, 5124-5142 (2007)], or equivalent ultrasound contrast agents. In some embodiments, the ultrasound contrast agents may be based on SonoVue®.

[0047] For the purpose of the present disclosure, the term “microbubble(s)” as used, is intended to refer to an ultrasound contrast agent or any particle acting as intravascular acoustic source.

[0048] The ultrasound agent injection device 5 may be a push syringe in the example considered here.

[0049] The ultrasound agent injection device 5 may comprise a magnet in order to mix a solution comprising the ultrasound contrast agents.

[0050] The 2D array ultrasonic probe 2 may be a sparse 2D array of transducers or a rowcolumn array (RCA) as known in the art.

[0051] A conventional full 2D matrix array ultrasonic probe may have n*n transducers disposed as a matrix along two perpendicular axes X, Y, transmitting ultrasound wavesalong an axis Z which is perpendicular to the XY plane. In one specific example, the 2D array ultrasonic probe 2 may have 1024 transducers (32 x 32), requiring 1024 scanner channels.

[0052] The sparse 2D array probe uses only a part of the transducers of a conventional full 2D array ultrasonic probe to reduce the number of channels while having a large imaging view. In one specific example, the 2D sparse array probe may select 256 elements out of a 32*32 array probe. The selection may be based an optimized pattern which may be either deterministic or stochastic.

[0053] The row-column addressing array (RCA) probe 2 may have n+n rectangle transducers disposed in rows and columns extending a X Y plane, transmitting ultrasound waves along an axis Z which is perpendicular to the XY plane.

[0054] The transducers may be configured to work as a receiver, or emitter, or emitter and receiver.

[0055] The transducers may be adapted to transmit and receive ultrasound waves having a central frequency comprised for instance between 0.5 and 100 MHz, for instance between 1 and 20 MHz. One example of usable central frequency is 15 MHz.

[0056] An example of method of conventional 3D Ultrasound Localization Microscopy imaging, will now be explained with regards to Figure 3 and Figure 4.

[0057] Injection of ultrasound contrast agents or microbubbles

[0058] The ultrasound contrast agent injection device 5 may be controlled by the control unit 3 to inject S1 (MBJNJ) intravenously microbubbles to a region 1 of the vascular network.

[0059] The ultrasound contrast agent injection may be a bolus injection with a maximum injection volume of 40 mL / kg for rats (corresponding to 12 mL for a 300g rat and approximately 3 mL for a 70g mice) and a typical 6 mL injection volume for rats and 2 mL injection volume for mice. For a human, a maximum volume of 10 mL may be injected. For instance, two bolus of 2,4 mL may be injected in the human.

[0060] The ultrasound contrast agent injection may be a continuous injection. For instance, a flow rate of the continuous injection may be comprised between 3 mL / h / kg and 60 mL / h / kg, and preferably approximatively 10 mL / h / kg (or typically 3.5 mL / h for a 300 g rat and 1.0 mL / h for a 70 g mice). Continuous injections enable a stable number of microbubbles for more than 20 minutes with approximately 30 microbubbles per ultrasound frame which correspond to a compound image.

[0061] Acquisition of ultrasound data over time from the region by the 2D array ultrasonic probe

[0062] The 2D array of transducers 2 may be controlled by the control unit 3 to acquire S2 (DATA_ACQ) 3D compounded images of the region of the vascular network during the ultrasound contrast agent injection or after having injected the ultrasound contrast agent.

[0063] The 2D array ultrasonic probe may be controlled by the unit control 3 to transmit a series of successive 2D tilted plane waves in the region of the vascular network to be imaged. For instance, 256 transducers transmit simultaneously successively 256 radiofrequency signals in the region to be imaged.

[0064] In order to enhance image resolution and contrast, for each 3D image of the region, a number Npof 2D tilted ultrasonic plane waves may be successively transmitted with different propagation directions. Each series may include for instance 5 to 20 successive ultrasonic plane waves of different directions, for instance 10 to 20 successive ultrasonic plane waves of different directions.

[0065] For instance, in the example illustrated in Figure 4, a combination of 10 tilted plane waves is transmitted with x and y directions parallel to the 2D array probe. The 10 plane waves are defined by a pair of angles and may be transmitted at a PRF of 10 kHz, the rate of the 3D compounded images of the region (framerate) is thus 1 KHz. Npmay be different than 10, in which case the framerate of 3D compounded images is different. For instance, Np=16 may be used. The 2D array of transducers may be placed on a skull of a human or animal for transcranial experiments, or directly on a brain of the human or animal for experiments involving a craniotomy. For instance, the probe may be placed perpendicularly to the surface of the head. To image the brain entirely, the correct placement of the probe may be ensured by a four-axis motor. The transmitted ultrasonic waves may be diverging waves, planar waves, or unfocused waves.

[0066] The 2D array ultrasonic probe 2 may be controlled by the unit control 3 to acquire the backscattered ultrasonic waves or backscattered echoes. The received signals may be registered as a set of raw data (also usually called RF data or radiofrequency data) for each transmitted ultrasonic wave. For each transmitted signal by the 2D array ultrasonic probe, backscattered signals may be generated by the region in response to the transmitted signal. The generated backscattered signals contain information about scatters formed by the microbubbles in the region.

[0067] Beamforming of the received ultrasonic signals and filtering

[0068] The computer 4 comprises a computing module configured to process S3 (DATA_PROC) the acquired signals from the region by the 2D array ultrasonic probe to obtain 3D compounded images and filtered 3D compounded images of the region.

[0069] A 3D delay-and-sum algorithm is performed on the received Npsets of raw data to reconstruct an ultrasound volume of the region, which is thus a 3D compounded image or 3D IQ volume. In the example of figure 4, 10 plane waves are coherently compounded to form a final, high quality 3D image. The framerate, i.e. the rate of 3D compounded images may be of several thousand 3D images per second, for instance 1000 to 5000 3D images per second.

[0070] Based on the successive 3D compound images of the region of the vascular network, filtered images may then be computed by the computer to remove the signal from the tissue to reveal the microbubbles. In the example of Figure 4, Nesuccessive 3D compound images are used for each filtered image. The Nc successive compound images may be temporally continuous or may partially overlap in the temporal dimension. For instance, Nc may be comprised between 10 and 1000, for instance between 200 and 600. One example of usable number of successive compound images for one filtered image is 200. Time positions may be associated to the filtered images. A time position for a given filtered image may correspond to an average of acquisition times of the Nc compound images forming the filtered image.

[0071] The filtered images may be computed for instance by a Singular Value Decomposition (SVD). More specifically, a SVD spatiotemporal clutter filter, as for instance described by Demene et al. [Demene, C. et al. Spatiotemporal Clutter Filtering of Ultrafast Ultrasound Data Highly Increases Doppler and fUltrasound Sensitivity. IEEE Transactions on Medical Imaging 34, 2271 - 2285 (2015)], may be applied. In the example of Figure 4, each filtered image may be a singular value or a sum of singular values of the SVD applied to the Nc compound images. Alternatively, owing to the SVD of the compound images, for each filtered image, the ultrasonic signature of the individual ultrasound contrast agents may be discriminated from the tissue signal of the region. By discarding some singular values of the SVD, the tissue signal may be filtered. For instance, the first 10 singular values of the SVD, mostly reflecting the tissue signal of the region, may be discarded.

[0072] 3D ultrafast Ultrasound Localization Microscopy (ULM) processing from the filtered 3D compounded images

[0073] The computer 4 may be configured to compute S4 (3D ULM_COMP) 3D ULM images of the region of the vascular network based on the filtered 3D compounded images of the region and will be described with regards to figures 5, 6 and 7.

[0074] With reference to figure 5, the filtered 3D compounded images may be processed to generate a 3D microbubble density image and velocity image according to the following steps:- detecting and localizing (DETEC_CMS) a plurality of isolated spots in each filtered 3D ultrasound images, each detected and located spot being a candidate main spot formed by the main lobes of the PSF of a microbubble, said candidate main spot assigned to a potential microbubble (PMB) ;- detecting (DETEC_CSS) for each candidate main spot the candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot, said candidate secondary spot assigned to a potential phantom microbubble (PPMB);- assigning (ASSIGN_SPOT) the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the main spot corresponding to the microbubble according to a selection parameter;- sorting (SORT_SPOT) the detected and localized spots by keeping the main spots corresponding to the microbubbles and removing the secondary spots corresponding to the phantom microbubbles;- tracking (TRACK_MB) the kept spots through successive 3D filtered ultrasound images;- generating (CONST_MAP) a 3D ULM image by attributing for each voxel an intensity value given by the number of kept spots in each voxel.

[0075] Detection and localization

[0076] After the step of filtering to remove tissue signal, ultrasound contrast agents or microbubbles are of high amplitude compared with the surrounding noise and are now visible and displayed as isolated spots in the filtered 3D images as shown schematically in Figure 4.

[0077] The 3D ULM processing comprises a step of detecting and localizing E1 (DETEC_CMS) a plurality of isolated spots in each 3D filtered ultrasound images, each detected and localized spot being a candidate main spot formed by the main lobes of the Point Spread Function “PSF” of a microbubble, said candidate main spot assigned to a potential microbubble (PMB).

[0078] Each filtered 3D compounded image is first interpolated in each dimension to give a voxel size, in lateral directions X, Y and in depth direction Z. For instance, the voxel size may be 0.1x0.1x0.04 mm3. Then, for each volume, ultrasound contrast agents or microbubbles may be detected as the brightest local maxima with high correlation with a point spread function over the entire volume. The local maxima are computed as connected components of voxels with the same intensity value, surrounded by voxels with a lower value. High correlation may be defined as a correlation superior to a threshold. The threshold may be a value comprised between 0.5 and 1. For instance, the threshold may be equal to 0.7. The point spread function is an imaging response of an isolated ultrasoundcontrast agent, which may be modelled as a Gaussian distribution extending in the depth direction Z and in the lateral directions X, Y. For instance, local maxima whose crosscorrelation with PSF exceeded a given threshold equal to 0.65 are assigned as potential microbubbles. In other words, in this step, all detected isolated bright spots are considered as potential microbubbles.

[0079] The centre position of each detected spot is provided by using a local second-order polynomial fit. A neighborhood for the local second-order polynomial fit may be a 3x3x3 voxel neighborhood. Thus, the coordinates of each detected spot are defined for each fit.

[0080] This step provides a table including the coordinates and the backscattering amplitude of each spot which can be a potential microbubble.

[0081] In case of the conventional ULM imaging, after detecting microbubbles and measuring their positions, the tracking step is performed directly on the localized microbubbles to compute a 3D ULM microbubbles density image and / or 3D velocity image as known in the prior art.

[0082] However, in case of using of a subsampling ultrasound imaging device such as the RCA probe or the sparse matrix array probe as indicated above, results show that reducing the number of channels cause distortions in the shape of the PSF and cause strong side lobes. Consequently, the presence of these side lobes in the PSF introduces the formation of secondary spots that could lead to false microbubbles detection or phantom microbubbles detection in the 3D filtered image. These incorrectly localized microbubbles, or false microbubbles may create false blood vessels in the ULM map as illustrated in figure 1 B, reducing thus the accuracy of the vascular structure and hindering the velocity determination and quantification inaccurate.

[0083] Sorting the detected and localized spots

[0084] One embodiment of the present disclosure is to provide a new 3D ULM processing that implements an additional step to reduce the influence of the presence of these strong side lobes by removing the false microbubbles from the filtered 3D images before the tracking step.

[0085] Consequently, after measuring the detected and localized microbubbles positions, instead of using all detected and localized microbubbles in the step of tracking to compute a 3D ULM microbubbles density image and / or 3D velocity image as known in the prior art, to make sure that the spots detected and localized are indeed the center of the real microbubble, the 3D ULM processing adds a second control loop to sort the detected and localized spots in order to identify the spots corresponding to the real microbubbles and the spots corresponding to the phantom microbubbles. In other words, the sorting stepproposes to determine the probability for a detected and localized microbubble belonging to the main lobe of the PSF of the microbubble or belonging to one of the side lobes of the PSF. The determined phantom microbubbles are then removed from the filtered 3D image. Consequently, the tracking step is carried out only with the kept main spots, i.e. real microbubbles, allowing the generation of a 3D ULM image with a higher quality.

[0086] In other words, this sorting step is carried out only with the table obtained after the detection and localization step and before the tracking step.

[0087] The method comprises a step E3 of detection of candidate secondary spots for each candidate main spot (DETECT_CSS). During this step, for each candidate main spot, a set of candidate secondary spots are detected generated. These candidate secondary spots are formed by the side lobes of the PSF of the microbubble associated to the candidate main spot and are assigned to a potential phantom microbubble (PPMB)

[0088] In one embodiment, the step E2 of identifying of candidate secondary spots for each candidate main spot may be based on analytical formula of the PSF and comprises:- calculating the position of each side lobe of the point spread function of the microbubble associated to the candidate main spot;- detecting the spots at the location of said side lobes (spots having coordinates corresponding to that of side lobes), said identified spots corresponding to the candidate secondary spots generated by the side lobes of the point spread function (PSF) of the microbubble.

[0089] Figure 6 illustrates a schematic view of a spot 12 in dark surrounded by seven spots in grey 13.1 , 13.2, 13.3, 13.4, 13.5, 13.6 and 13.7. Their positions are determined in the step of localization. In order to determine if the spot 12 is a real microbubble in this example, the position of each side lobe of the PSF of the microbubble is determined. The spots which are located at the location of determined side lobes correspond to the candidate secondary spots, corresponding to the potential phantom microbubbles. In the example of Figure 6, the spots in grey are identified as secondar spots.

[0090] The method comprises a step E4 of assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the main spot corresponding to the microbubble according to a selection parameter.

[0091] In one embodiment, the selection parameter may be the intensity value of spots.The step of assigning (f) includes:- comparing for each candidate main spot its intensity value with the intensity value of theidentified candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot;- assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the real main spot corresponding to the microbubble if the intensity value of the candidate secondary spots is lower than that of the candidate main spot.

[0092] The method comprises a step E5 of sorting the detected and localized spots by keeping the main spots corresponding to the microbubbles and removing the secondary spots corresponding to the phantom microbubbles. In other words, the detected and localized spots having an intensity value higher than that of the secondary spots in each 3D ultrasound images are kept and the spots having an intensity value lower than that of the secondary spots in each 3D ultrasound images are removed.

[0093] Thus, in the example of Figure 6, the intensity value of the spot 12 which is assigned as potential microbubble, is compared with the intensity value of the other spots 13.1 , 13.2, 13.3, 13.4, 13.5, 13.6, 13.7 located at the position of the side lobes.

[0094] The spot 12 will be kept in the filtered 3D image if its intensity value is higher than that of the candidate secondary spots. The identified secondary spots are removed from the image.

[0095] Thus, the spot 12 is stored as microbubble only if its intensity value is higher than that of the secondary spots. The spots 13.1-13.7 will be removed from the filtered 3D image. This operation will be applied for all detected and localized spots that are considered as potential microbubble.

[0096] In another embodiment, the selection parameter may be the number of candidate secondary spots detected associated to the candidate main spot, the step of assigning (f) includes :- counting for each candidate main spot the number of candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot;- assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the real main spot corresponding to the microbubble if the number of candidate secondary spots detected corresponds to the number of side lobes of the point spread function (PSF) of the microbubble associated to said candidate main spot.

[0097] Thus, in the example of Figure 6, the number of candidate secondary spots distributed around the spot 12 are counted. In the case where this number corresponds tothe number of side lobes of the PSF of the microbubble associated to the candidate main spot 12, the detected candidate secondary spots are assigned to the secondary spots. The candidate main spot 12 is assigned to the main spot and will be kept in the filtered 3D image. The secondary spots 13.1-13.7 will be removed from the filtered 3D image.

[0098] In another embodiment, the step of identifying of candidate secondary spots for each candidate main spot comprises a step of detecting the systematic presence of a set of spots surrounding said candidate main spot over the successive 3D ultrasound images, said detected set of spots corresponding to the candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to said candidate main spot.

[0099] Figure 7 illustrates the schematic view of a spot 12 in dark and seven spots 13 in grey of figure 6 in two successive frames at two times t and t+1.

[0100] For the first frame taken at time t, the spots are located at an initial location. For second frame, taken at time t+1 , the spots are moved from their initial location to another location. In the case where the spots in grey that are distributed around the spot in dark are candidate secondary spots created by the side lobes of the PSF of the microbubble, the candidate secondary spots and the candidate main spots appear necessarily and systematically simultaneously in each frame. In other words, their position relative to the candidate main spot stay the same in each frame as illustrated in the second frame. Thus, it may be possible to identify the candidate secondary spots for each candidate main spots by using a temporal co-occurrence analysis. The candidate secondary spots can be found by estimating the covariance of the temporal signals from the lobes. In the example of figure 7, two frames are processed sequentially to identify the spots located at the location of the side lobes. In another embodiment, more than two frames may be processed to identify the secondary spots.

[0101] In another embodiment, the candidate secondary spots can be found based on an experimental measurements of the PSF.

[0102] In further another embodiment, the candidate secondary spots can be found by a numerical simulation of the PSF or calculated by an analytical formula of the PSF. The simulated or calculated positions of the secondary spots are stored in a look up table.

[0103] Consequently, the sorting step (E5) permits to suppress the false microbubbles called phantom microbubbles that do not belong to the main lobe of the PSF of the microbubble. This way, the microbubble density is reduced in each filtered image facilitating the tracking step as only the kept spots are assigned to localized microbubbles. Their positions are stored over time to be tracked.

[0104] Advantageously, the location of secondary lobes can be used to refine the location of the main spot. Before the step of sorting, the method further comprises:- storing the location of the main spots corresponding to the microbubbles and the location of the secondary spots corresponding to the phantom microbubbles;- correcting the location of the main spot by fitting the location of secondary spots to the theoretical secondary lobe location of a microbubble model.

[0105] Tracking of the stored microbubbles

[0106] In the step of processing of the 3D compounded images S4, the stored microbubbles resulting from the sorting step may be tracked in the filtered images (TRACK_MAP) E6. A tracking of the stored microbubbles may be performed using a particle tracking algorithm known in the art. A track may correspond to positions of a tracked microbubble in the filtered images. Each position of a tracked microbubble in the filtered images may be associated with a time position corresponding to the time position of the filtered image in which the position is located. For instance, the Kuhn-Munkres algorithm or Hungarian algorithm may be performed to track the microbubbles. For each stored microbubble of frame at time t, the square distance to all microbubbles in frame at time t+1 is computed. The optimal pairing is given by minimizing the total square distance of matched microbubbles. This algorithm is looped in every frame.

[0107] T racks with stored microbubble in a predefined number of successive filtered images may be computed. The predefined number of successive filtered images may be comprised between 1 and 100. For instance, the predefined number of successive filtered images may be 10. A spatial interpolation may be computed for each track in order to obtain one ultrasound contrast agent in each voxel located on the path between two successive voxels of the track.

[0108] Reconstruction of microbubbles density maps and microbubbles velocity maps

[0109] As illustrated in figure 5, after the step of tracking, 3D ULM microbubble density or 3D ULM velocity images may be computed based on the tracks E7 (CONST_MAP), which are based on the filtered images.

[0110] The construction of conventional 3D ULM images is based on the (x, y, z) position of the stored microbubbles. 3D ULM images are obtained by counting the number of microbubbles detected in each voxel, resulting in microbubbles count maps, commonly known as microbubbles density maps. 3D ULM image may be constructed by selecting a voxel size and by counting stored microbubbles within each voxel. For instance, 3D ULM microbubble density image may be computed by counting, for each voxel, the number ofstored microbubbles in the corresponding voxel of the filtered images. For instance, the voxel size for 3D ULM microbubble density image construction may be between 10 and 20 pm.

[0111] Similarly, velocity maps are reconstructed from the amplitude velocity of every stored microbubble in each voxel. The velocity value in each voxel may be taken as the average velocity measured for each stored microbubble passing through this voxel.

[0112] Successive positions of ultrasound contrast agents in a track may be used to compute velocity parameters. For example, interframe microbubble velocity vector components may be computed along the probe x axis and y axis and the depth z axis for 3D imaging. It is possible to calculate the three velocity components Vx, Vyand Vzalong respectively the three directions X, Y, Z.

[0113] These velocity maps may lead to quantitative and localized maps of brain blood flow velocity.

[0114] In one embodiment, the method comprises an automatic segmentation step of the vessels in at least one area of a vascular network from a 3D ULM image. Thus, it is possible to quantify at least one dimensional parameter of the vessels from the segmented image. For example, the dimensional parameter may be the vessels diameter or the length of each segment of vessels.

[0115] The supplementary sorting step combined with the localization technique provides a 3D ULM image with higher quality. As the phantom microbbubles have been removed, the reconstruction of the vascular network is based on the real microbubbles. In the example of figures 1A and 1 B, the suppress of the phantom microbubbles 13 allows the reconstruction of the real blood vessel 13 as illustrated in Figure 8.

[0116] Based on the 3D ULM microbubbles density images, at least one vascular dynamic parameter may be computed accurately. The at least one vascular dynamic parameter may be chosen in the group comprising: blood flow rate, blood velocity, blood volume, blood pressure, vascular vessels’ diameters, and any combination thereof.

[0117] For instance, quantification of vessel diameter may be performed on the 3D density maps reconstructed with isotropic voxels of size between 2 pm and 500 pm.

[0118] For instance, the amplitude of the flow velocity may be measured at the centerline points of a skeleton computed on the 3D ULM map and assumed to be the maximum velocity of the velocity profile. Using the maximum velocity and diameter measured, the flow rate may be computed assuming a Poiseuille flow: Q = Vmax / 2 x KR2, where Vmax : maximum velocity, R: radius of the detected vessel.

[0119] In the present disclosure, the vascular network described is the vascular network of a nervous system. Nevertheless, the methods and apparatus described may be adapted to any vascular network.

Claims

Claims

1. Method for 3D ultrasound localization microscopy (ULM) imaging at a microscopic scale a region of a vascular network of an organ, of a human or animal, said region containing microbubbles, the method comprising the following steps:(a) transmitting incident ultrasonic waves in the region by a 2D array of ultrasonic transducers and acquiring raw data from backscattered ultrasonic waves by said 2D array of transducers;(b) generating a series of successive 3D ultrasound images from said raw data;(c) filtering the 3D ultrasound images to remove tissue signal;(d) detecting and localizing a plurality of isolated spots in each filtered 3D ultrasound images, each detected and located spot being a candidate main spot formed by the main lobes of the PSF of a microbubble, said candidate main spot assigned to a potential microbubble (PMB) ;(e) detecting for each candidate main spot the candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot, said candidate secondary spot assigned to a potential phantom microbubble (PPMB);(f) assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the main spot corresponding to the microbubble according to a selection parameter;(g) sorting the detected and localized spots by keeping the main spots corresponding to the microbubbles and removing the secondary spots corresponding to the phantom microbubbles;(h) tracking the kept spots through successive 3D filtered ultrasound images;(i) generating a 3D ULM image by attributing for each voxel an intensity value given by the number of kept spots in each voxel.

2. Method according to claim 1 , wherein the selection parameter being the intensity value of spots, the step of assigning (f) includes:- comparing for each candidate main spot its intensity value with the intensity value of the identified candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot;- assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the real main spot corresponding to the microbubble if the intensity value of the candidate main spot is higher than that of the candidate secondary spots.

3. Method according to claim 1 , wherein the selection parameter being the number of candidate secondary spots detected associated to the candidate main spot, the step ofassigning (f) includes :- counting for each candidate main spot the number of candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot;- assigning the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the real main spot corresponding to the microbubble if the number of candidate secondary spots detected corresponds to the number of side lobes of the point spread function (PSF) of the microbubble associated to said candidate main spot.

4. Method according to any of claims 1 to 3, before the step of sorting (g), further comprising:- storing the location of the main spots corresponding to the microbubbles and the location of the secondary spots corresponding to the phantom microbubbles;- correcting the location of the main spots by fitting the location of secondary spots to the theoretical secondary lobe location of a microbubble model.

5. Method according to any of claims 1 to 4, wherein the step of detecting of the candidate secondary spots for each candidate main spot comprises:- calculating for each candidate main spot the side lobes of the point spread function (PSF) of the microbubble associated to said candidate main spot;- detecting the spots at the location of the side lobes, said detected spots being the candidate secondary spots.

6. Method according to any of claims 1 to 4, wherein the step of detecting of candidate secondary spots for each candidate main spot comprises:- detecting for each candidate main spot the systematic presence of a set of spots surrounding said candidate spot over the successive 3D ultrasound images, said detected set of spots corresponding to the candidate secondary spots generated by the side lobes of the point spread function (PSF) of the microbubble associated to said candidate main spot.

7. Method according to any of claims 1 to 6, wherein the point spread function (PSF) of the candidate main spot is varied spatially.

8. Method according to any of claims 1 to 7, wherein detecting and localizing a plurality of isolated spots in each 3D ultrasound image comprises:- detecting local maxima in each 3D ultrasound image;- determining the center position of each local maxima by fitting the voxel around the local maxima with a 3D paraboloid on the voxels around the local maxima.

9. Method according to any one of claims 1 to 8, further comprising a step of computing of the speed velocity components Vx, Vyand Vzalong respectively the three directions X, Y, Z for each kept spot to generate a 3D velocity map.

10. Method according to any one of claim 1 to 9, wherein the 2D array of ultrasonic transducers is a row-column array (RCA).

11. Method according to any one of claim 1 to 9, wherein the 2D array of ultrasonic transducers is a sparse array having a reduced number of active elements being distributed randomly or according to a predetermined pattern.

12. Method according to any one of claims 1 to 11 , wherein at least one vascular parameter is extracted from the 3D ULM image, said one vascular parameter being chosen in the group comprising: vessel diameter, blood flow, blood velocity, blood volume, blood pressure, and any combination thereof.

13. Apparatus (1) for 3D ultrasound localization microscopy (ULM) imaging at a microscopic scale in a region of a vascular network of an organ, of a human or animal, said apparatus including at least an ultrasound probe (5) having a 2D array of ultrasonic transducers (7) and a control system (2) configured to:(a) transmit incident ultrasonic waves in the region though said 2D array of ultrasonic transducers and acquire raw data from backscattered ultrasonic waves through said 2D array of ultrasonic transducers;(b) generate a series of successive 3D ultrasound images from said raw data;(c) filter the 3D ultrasound images to remove tissue signal;(d) detect and localize a plurality of isolated spots in each filtered 3D ultrasound images, each detected and localized spot being a candidate main spot formed by the main lobes of the PSF of a microbubble, said candidate main spot assigned to a potential microbubble (PMB);(e) detect for each candidate main spot the candidate secondary spots generated by the side lobes of the PSF of the microbubble associated to the candidate main spot, said candidate secondary spot assigned to a potential phantom microbubble (PPMB);(f) assign the detected candidate secondary spots to the secondary spots corresponding to the phantom microbubbles and the candidate main spot to the main spot corresponding to the microbubble according to a selection parameter;(g) sort the detected and localized spots by keeping the main spots corresponding to the microbubbles and removing the secondary spots corresponding to the phantom microbubbles;(h) track the kept spots through successive 3D filtered ultrasound images;(i) generate a 3D ULM image by attributing for each voxel an intensity value given by the number of kept spot in each voxel.

Images