阅读说明：本技术 用于微血管超分辨率超声成像的方法 (Method for microvascular super-resolution ultrasound imaging ) 是由 宋鹏飞 J·D·奇萨思科 A·曼杜卡 S·陈 于 2018-05-30 设计创作，主要内容包括：描述了用于受试者体内微血管的超分辨率超声成像的系统和方法。从已经注射微气泡造影剂的受试者体内感兴趣的区域获取超声数据。当微气泡正穿过感兴趣的区域或以其他方式存在于感兴趣的区域中时获取超声数据。感兴趣的区域可以包括,例如,受试者体内的微血管或其他微脉管系统。通过隔离、定位、跟踪和积累超声数据中的微气泡,可以生成微血管的超分辨率图像。(Systems and methods for super-resolution ultrasound imaging of microvessels in a subject are described. Ultrasound data is acquired from a region of interest within a subject that has been injected with a microbubble contrast agent. Ultrasound data is acquired while microbubbles are traversing or otherwise present in a region of interest. The region of interest may include, for example, microvasculature or other microvasculature within the subject. By isolating, locating, tracking and accumulating microbubbles in the ultrasound data, a super-resolution image of the microvessels can be generated.)

1, a method for super resolution imaging of a microvascular using an ultrasound system, the steps of the method comprising:

(a) providing ultrasound data to a computer system, the ultrasound data having been acquired with an ultrasound system from a region of interest within a subject, microbubble contrast agent being present in the region of interest when the ultrasound data is acquired;

(b) generating, with the computer system, microbubble signal data by isolating microbubble signals in the ultrasound data from other signals in the ultrasound data;

(c) locating microbubbles in the microbubble signal data by processing the microbubble signal data with the computer system to determine spatial locations associated with microbubbles in the microbubble signal data;

(d) generating a super-resolution microvascular image based at least in part on the localized microbubble signal.

2. The method of claim 1, wherein isolating microbubble signals comprises removing background signals in the ultrasound data, wherein the background signals comprise signals associated with tissue and signals associated with stationary microbubbles.

3. The method of claim 2, wherein removing the background signal comprises filtering the ultrasound data using a high-pass temporal filter to isolate microbubble signals in the ultrasound data.

4. The method of claim 3, wherein the high-pass time filter comprises a cut-off frequency lower than a temporal frequency of the isolated microbubble signal and higher than a temporal frequency of the background signal.

5. The method of claim 2, wherein removing the background signal comprises frame-to-frame subtraction, wherein temporally successive frames of the ultrasound data are subtracted to isolate the microbubble signal while subtracting background signal.

6. The method of claim 2, wherein removing the background signal comprises performing singular value decomposition-based filtering on the ultrasound data to isolate microbubble signals in the ultrasound data.

7. The method of claim 6, wherein a singular value cutoff is selected to separate the background signal from the isolated microbubble signal, wherein the singular value cutoff is greater than a singular value associated with the background signal and less than a singular value associated with the isolated microbubble signal.

8. The method of claim 1, wherein locating microbubbles in the microbubble signal data comprises:

providing a point spread function of the ultrasound system used to acquire the ultrasound data to the computer system; and

identifying, with the computer system, a spatial location of the microbubbles based on an normalized cross-correlation between the microbubble signal data and the point spread function of the ultrasound system.

9. The method of claim 8, wherein the point spread function is a simulated point spread function.

10. The method of claim 9, in which the simulated point spread function is simulated based at least in part on a multivariate gaussian distribution.

11. The method of claim 8, wherein the point spread function is estimated based on measurements acquired by imaging a small point object with the ultrasound system.

12. The method of claim 11, wherein the small point object is smaller than an ultrasound wavelength used to image the small point object.

13. The method of claim 12, wherein the small point objects are microbubbles.

14. The method of claim 8, wherein identifying the spatial location of the microbubbles comprises:

generating a cross-correlation map by calculating the two-dimensional normalized cross-correlation between microbubble signal data and the point spread function of the ultrasound system, the cross-correlation map depicting correlation coefficients between the microbubble signal data and the point spread function of the ultrasound system;

removing entries in the cross-correlation map having correlation coefficients below a selected threshold; and

selecting the spatial position of the microbubbles based on a local area maximum of the correlation coefficient retained in the cross-correlation map.

15. The method of claim 1, wherein the spatial position of the microbubbles is a center position of the microbubbles in every time frames of the ultrasound data.

16. The method of claim 1, wherein isolating the microbubble signal comprises spatially interpolating the microbubble signal data to a finer spatial resolution prior to processing the microbubble signal data to isolate the microbubble signal.

17. The method of claim 1, wherein generating the microvascular image comprises globally tracking the microbubbles as a function of time, and generating the microvascular image based on the tracking of the microbubbles.

18. The method of claim 17, wherein globally tracking the microbubbles comprises performing bipartite graph minimum distance pairing of microbubble pairings between temporally successive image frames in the microbubble signal data.

19. The method of claim 18, wherein performing the bipartite graph minimum distance pairing comprises:

generating a distance map by calculating the distance between the spatial position of every microbubbles in the th image frame and the spatial position of every microbubbles in the second image frame;

thresholding the distance map based on physiological limits of blood flow velocity; and

minimizing the total distance of paired microbubbles.

20. The method of claim 19, wherein a total distance of paired microbubbles is minimized based at least in part on a partial distribution algorithm that pairs microbubbles in the th image frame with microbubbles in the second image frame based on a minimum pairing distance.

21. The method of claim 17, wherein globally tracking the microbubbles comprises pairing microbubbles in temporally consecutive image frames based on an optimized dispense solution that dispenses microbubbles in an th image frame to microbubbles in a second image frame.

22. The method of claim 21, wherein the optimized allocation solution is implemented as a Kuhn-Munkres algorithm.

23. The method of claim 17, wherein globally tracking the microbubbles comprises pairing microbubbles in different image frames and performing frame-to-frame persistence control on the paired microbubbles to carefully examine microbubble pairing.

24. The method of claim 23, wherein the frame-to-frame persistence control rejects microbubble pairings that are not successfully paired in a selected number of temporally consecutive image frames.

25. The method of claim 24, wherein the selected number of temporally successive image frames is at least two temporally successive image frames.

26. The method of claim 24, wherein the selected number of temporally successive image frames is at least five temporally successive image frames.

27. The method of claim 23, wherein the microvascular image depicts a measure of local blood flow calculated based on tracking microbubbles over the selected number of temporally successive image frames.

28. The method of claim 1, further comprising processing the microvascular image to recover missing microvascular data .

29. The method of claim 28, wherein recovering the missing microvascular data comprises applying a spatial low pass filter to the microvascular image.

30. The method of claim 28, wherein recovering the missing microvascular data comprises applying a non-linear averaging filter to the microvascular image.

31. The method of claim 28, wherein recovering the missing microvascular data comprises inputting the microvascular image to a sparsity-promoting signal estimation.

32. The method of claim 31, wherein the sparsity-facilitating signal estimation comprises minimizing a cost function including a sparsity-facilitating term and a data fidelity term.

33. The method of claim 32, wherein the sparsity promoting terms are based on at least of fourier transforms, curvelet transforms, wavelet transforms, or singular value decompositions.

34. The method of claim 32, in which the sparsity promoting terms are learned empirically using a machine learning algorithm.

35. The method of claim 34, wherein the machine learning algorithm comprises a K-SVD algorithm.

36. The method of claim 32, wherein the data fidelity term comprises measuring a difference between the input microvascular image and the updated microvascular image over each iterations of the sparsity-facilitating signal estimation.

37. The method of claim 32, wherein the data fidelity term comprises maximizing a probability of viewing the input microvascular image in the case of an estimated microbubble signal.

38. The method of claim 37, wherein said estimated microbubble signal is based on a poisson distribution likelihood function.

39. The method of claim 37, wherein said estimated microbubble signal is based on a gaussian distributed likelihood function.

40. The method of claim 32, wherein the sparsity-facilitating signal estimates are constrained by a constraint that facilitates signal estimation along a particular spatial direction.

41. The method of claim 1, wherein the microvascular image comprises an accumulated microbubble location map depicting the number of times a microbubble is present at a given location in a region of interest.

42. The method of claim 1, wherein the microvascular image depicts morphological measurements of microvasculature within the subject, wherein the morphological measurements include at least of blood vessel density or blood vessel tortuosity.

43. The method of claim 1, wherein the microvascular image depicts a hemodynamic measurement of a microvascular within the subject, wherein the hemodynamic measurement comprises at least of blood flow velocity, a perfusion index derived from blood flow velocity, or a cross-sectional blood flow rate.

44. The method of claim 1, further comprising de-noising the isolated microbubble signal data prior to locating the microbubbles.

45. The method of claim 44, wherein de-noising the microbubble signal data comprises de-noising the microbubble signal data in a spatio-temporal domain.

46. The method of claim 45, wherein denoising the microbubble signal data in the spatio-temporal domain comprises applying a non-local mean filter to the microbubble signal data.

47. The method of claim 45, wherein denoising the microbubble signal data in the spatio-temporal domain comprises applying a Gaussian smoothing filter to the microbubble signal data.

48. The method of claim 45, wherein denoising the microbubble signal data in the spatio-temporal domain comprises applying a median filter to the microbubble signal data.

49. The method of claim 1, further comprising processing the ultrasound data to remove tissue motion prior to isolating the microbubble signals, wherein the tissue motion is caused by at least of transducer movement, a cardiovascular system of the subject, or a respiratory system of the subject.

50. The method of claim 49, wherein processing the ultrasound data to remove tissue motion comprises computing measurements of the tissue motion between time frames of the ultrasound data and a reference frame of the ultrasound data, and realigning the time frames of the two consecutive ultrasound data based on the measurements of the tissue motion.

51. A method according to claim 50, wherein the measure of tissue motion is calculated based on at least of image intensity-based image registration, image feature-based image registration, spectral phase-based image registration methods, transformation model-based image registration, or ultrasound speckle tracking-based methods.

52. The method of claim 50, wherein the reference frame of ultrasound data is th frame of ultrasound data.

53. The method of claim 1, wherein the provided ultrasound data comprises ultrasound data having at least two spatial dimensions and at least time dimensions.

54. The method of claim 1, wherein the provided ultrasound data comprises ultrasound data acquired from both a linear component and a non-linear component of an ultrasound signal.

55. The method of claim 1, wherein providing the ultrasound data comprises acquiring the ultrasound data from a subject using the ultrasound system.

56. The method of claim 55, wherein acquiring the ultrasound data comprises applying microbubble concentration control pulses that modulate microbubble concentration in the region of interest to facilitate isolation of microbubble signals.

57. The method of claim 56, wherein the ultrasound data is acquired using an imaging pulse sequence comprising repeated applications of:

applying said microbubble concentration control pulse;

waiting a waiting time during which no ultrasonic pulse is applied;

applying a detection pulse after the waiting time to acquire the ultrasound data.

58. The method of claim 57, wherein the waiting time is selected such that an ultrasound signal generated by the microbubble concentration control pulse substantially attenuates prior to acquiring the ultrasound data.

59. The method of claim 57, wherein the power of the microbubble control pulses is adjusted during the imaging pulse sequence to maximize an amount of isolated microbubble signal in the acquired ultrasound data.

60, a method for generating an image indicative of microbubbles that have been tracked over time with an ultrasound system, the steps of the method comprising:

(a) acquiring ultrasound data from a region of interest in a subject, microbubble contrast agent being present in the region of interest when the ultrasound data is acquired, the ultrasound data having at least spatial dimensions and at least temporal dimensions including a plurality of time frames;

(b) generating microbubble signal data with a computer system by processing the ultrasound data to isolate microbubble signals from other signals in the ultrasound data;

(c) locating microbubbles in the microbubble signal data with a computer system by processing the microbubble signal data to determine spatial locations of microbubbles in every of the time frames;

(d) generating tracked microbubble data by globally tracking the spatial locations of microbubbles along a time dimension by pairing the spatial locations of the microbubbles in successive time frames; and

generating an image based on the tracked microbubble data.

61. The method of claim 60, wherein pairing spatial positions of microbubbles in consecutive time frames comprises performing bipartite graph minimum distance pairing.

62. The method of claim 61, wherein performing the bipartite graph minimum distance pairing comprises:

generating a distance map by calculating a distance between the spatial location of every microbubbles in an time frame and the spatial location of every microbubbles in a second time frame, the second time frame consecutive to the time frame;

thresholding the distance map based on physiological limits of blood flow velocity; and

minimizing the total distance of paired microbubbles.

63. The method of claim 60, wherein pairing spatial locations of microbubbles in different time frames comprises determining an optimized distribution solution that distributes microbubbles in a th time frame to microbubbles in a second time frame, the second time frame being different from the th time frame.

64. The method of claim 63, wherein the optimized allocation solution is determined based on a Kuhn-Munkres algorithm.

65. The method of claim 60, further comprising performing frame-to-frame persistence control on the tracked microbubble data to scrutinize microbubble pairing.

66. The method of claim 65, wherein the frame-to-frame persistence control rejects microbubble pairings that are not successfully paired in a selected number of temporally consecutive time frames.

67. The method of claim 66, wherein said selected number of temporally consecutive time frames is at least two temporally consecutive time frames.

68. The method of claim 66, wherein said selected number of temporally successive time frames is at least five temporally successive time frames.

69. The method of claim 60, wherein the image comprises an accumulated microbubble location map depicting a number of times a microbubble is present at a given location in the region of interest.

70. The method of claim 60, wherein the image comprises morphological measurements of microvessels in the region of interest, wherein the morphological measurements comprise at least of blood vessel density or blood vessel tortuosity.

71. The method of claim 60, wherein the image depicts a hemodynamic measurement of a microvasculature in the region of interest, wherein the hemodynamic measurement includes at least of blood flow velocity, a perfusion index derived from blood flow velocity, or a cross-sectional blood flow rate.

72, a method for generating an image depicting microbubbles with an ultrasound system, the steps of the method comprising:

(a) providing a spatially sparse image depicting microbubbles in a region of interest within a subject to a computer system;

(b) inputting the spatially sparse image to a sparsity-promoting signal estimation cost function comprising at least sparsity-promoting terms and at least data fidelity terms;

(c) generating, with the computer system, an updated image by minimizing the sparsity-facilitating signal estimation cost function to estimate signals associated with microbubbles that are not depicted in the spatially sparse image;

(d) storing the updated image for later use.

73. The method of claim 72, wherein the sparsity promoting terms are based on at least of Fourier transforms, curvelet transforms, wavelet transforms, or singular value decompositions.

74. The method of claim 72, wherein the sparsity promoting terms are learned empirically using a machine learning algorithm.

75. The method of claim 74, wherein the machine learning algorithm comprises a K-SVD algorithm.

76. The method of claim 72, wherein the data fidelity term comprises measuring a difference between the input microvascular image and an updated microvascular image over each iterations of the sparsity-facilitating signal estimation.

77. The method of claim 72, wherein the data fidelity term comprises maximizing a probability of viewing the input microvascular image in the case of an estimated microbubble signal.

78. The method of claim 77, wherein the estimated microbubble signal is based on a Poisson distribution likelihood function.

79. The method of claim 77, wherein the estimated microbubble signal is based on a Gaussian distribution likelihood function.

80. The method of claim 72, wherein the sparsity-facilitating signal estimates are constrained by a constraint that facilitates signal estimation along a particular spatial direction.

Background

Microbubbles typically range in size from 1-5 μm, much smaller than the wavelength of ultrasound, which is typically on the order of 100-800 μm.

The common approach to breaking the diffraction limit of imaging systems is to create isolated point sources that are spaced at least -half of the ultrasound wavelength from each other and use the center positions of these blurred point sources (e.g., at the maximum intensity values of the blurred point sources) to form an image.

In ultrasound imaging, two approaches have recently been proposed to address this challenge, method promotes microbubble isolation by diluting the microbubble solution, the second approach uses high frame rate ultrasound to monitor the "blinking" events of microbubbles, which can isolate individual microbubble signals from a large number of microbubbles.

Therefore, there is still a need for a super-resolution processing method that can significantly improve microbubble tracking and accumulation.

Disclosure of Invention

The present disclosure addresses the above-mentioned shortcomings by providing a method for super-resolution imaging of a microvasculature using an ultrasound system. Ultrasound data acquired with an ultrasound system from a region of interest in a subject is provided to a computer system, and microbubble contrast agent is present in the region of interest when the ultrasound data is acquired. Microbubble signal data is generated with a computer system by isolating microbubble signals in the ultrasound data from other signals in the ultrasound data. Microbubbles are localized in the microbubble signal data by processing the microbubble signal data with a computer system to determine spatial locations associated with microbubbles in the microbubble signal data. Generating a super-resolution microvascular image based at least in part on the localized microbubble signal.

Another aspect of the present disclosure is to provide a method for generating images indicative of microbubbles that have been tracked over time with an ultrasound system.

Another aspect of the present disclosure is to provide a method for generating an image depicting microbubbles using an ultrasound system.A spatially sparse image depicting microbubbles in a region of interest within a subject is provided to a computer system.A spatially sparse image is input to a sparsity-promoting signal estimation cost function comprising at least sparsity-promoting terms and at least data fidelity (data-fidelity) terms.

The foregoing and other aspects and advantages of the present disclosure will appear from the following description, in this specification, reference is made to the accompanying drawings which form a part hereof , and in which is shown by way of illustration a preferred embodiment, which is not, however, which is indicative of the full scope of the invention and is therefore intended to be referenced by the claims and herein for interpreting the scope of the invention.

Drawings

FIG. 1 is a flow chart setting forth the steps of an exemplary method for generating super-resolution images of microvessels with an ultrasound system.

Fig. 2 depicts an example of a microvascular moving between successive time frames due to tissue motion.

Fig. 3 is a flow chart setting forth the steps of an exemplary method for isolating microbubble signals based on cross-correlation with the point spread function of the ultrasound system.

Fig. 4 is an example of a two-dimensional cross-correlation map calculated between microbubble signal data and the point spread function of the ultrasound system used to acquire the microbubble signal data.

Fig. 5 is an example of microbubble signal data in which signals not associated with microbubbles have been rejected based on the two-dimensional cross-correlation map of fig. 4.

Fig. 6 is an example of the center position of microbubbles identified on the microbubble signal data of fig. 5, where the center position is determined by identifying local region maxima in the two-dimensional cross-correlation map of fig. 4.

Fig. 7 depicts an example of tracking the position of a microbubble over multiple time frames using a local window in which an error may occur when there are other microbubbles in the same local window.

Fig. 8 is an exemplary diagram showing the microbubble positions in the region of interest for two consecutive time frames.

Fig. 9 is a flow chart setting forth the steps of an exemplary method for pairing microbubble positions in consecutive time frames based on a bipartite graph and minimum distance pairing.

Fig. 10 is an example of a distance map depicting distances between pairs of microbubble positions in successive time frames of microbubble signal data.

FIG. 11 is an example of the distance map of FIG. 10 after thresholding the distance map based on physiological limits of blood flow.

Fig. 12 is an exemplary diagram showing paired microbubble positions in a region of interest for two consecutive time frames.

Fig. 13A is an example of an accumulated microbubble map before persistent control is implemented to remove spurious microbubble signals.

Fig. 13B is an example of the accumulated microbubble map of fig. 13A after implementing five frame-to-frame persistence controls to remove spurious microbubble signals.

Fig. 14 depicts an example of filtering microvessels to "patch" or "fill in" blank regions between detected microbubble locations.

Fig. 15 depicts an example of a microvascular image and its corresponding 4-scale curvelet transform representation.

Fig. 16 depicts an example of microvascular images before (left) and after (right) updating the microvascular images using sparsity-facilitating signal estimation.

Fig. 17 is an exemplary image sequence containing repeated applications of microbubble control pulses followed by series detection pulses.

FIG. 18 is a block diagram of an example ultrasound system in which methods described in this disclosure may be implemented.

Fig. 19 is a flow chart illustrating steps implemented by an example partial dispense algorithm that minimizes the pairing distance between microbubbles.

Detailed Description

Systems and methods for super-resolution ultrasound imaging of microvessels in a subject are described herein. Ultrasound data is acquired from a region of interest within a subject that has been injected with a microbubble contrast agent. Ultrasound data is acquired as microbubbles traverse or otherwise reside in a region of interest. The region of interest may include, for example, microvasculature or other microvasculature within the subject. By isolating, locating, tracking, and accumulating microbubbles in ultrasound data, super-resolution images of the microvessels may be generated as described in the present disclosure.

Referring now to fig. 1, a flow chart illustrating the steps of an exemplary method for using an ultrasound system to generate super-resolution images of microvessels in a subject that has been injected with a microbubble contrast agent is shown. Generally, super-resolution refers to enhanced resolution relative to that obtainable by an imaging system. For example, a super-resolution ultrasound image may refer to an image having a resolution finer than the diffraction limit.

In embodiments, providing the ultrasound data to the computer system may include retrieving previously acquired ultrasound data from a memory or other data storage device, which may be part of the computer system or separate from the computer system in other embodiments, providing the ultrasound data may include acquiring such data with the ultrasound system and providing the acquired data to the computer system, which may be part of the ultrasound system or separate from the ultrasound system.

Ultrasound data may be ultrasound radio frequency data, ultrasound in-phase-quadrature ("IQ") data, etc. typically, ultrasound data contains spatial dimensions, which spatial dimensions may include a transverse dimension, an axial dimension, an elevation dimension, and combinations thereof.

In embodiments, ultrasound data is acquired in a manner that facilitates microbubble localization, such as by using dilute microbubble solutions, "high frame rate monitoring of" blinking "microbubble events, or other such methods that may produce isolated microbubble sources.

As another example, ultrasound pulses having energy sufficient to break a number of microbubbles are used, wherein the broken microbubbles then release free bubbles from the microbubbles and generate ultrasound signals having different amplitudes than the intact microbubbles.

The microbubble signal can be acquired from both the linear and nonlinear components of the ultrasound wave. The linear component is typically at the fundamental frequency of the applied ultrasonic waves, while the nonlinear component may be at harmonic frequencies of the applied ultrasonic waves, at the fundamental frequency of the applied ultrasonic waves, or both. For example, the non-linearity introduced by amplitude modulation based imaging methods may be at the fundamental frequency.

The ultrasound data is processed to remove tissue motion, as indicated by step 104. For in vivo imaging, there may be transducer movement and tissue motion caused by cardiovascular (e.g., heart beat and beat from arteries) and respiratory systems. The amplitude of these motions can be significantly larger than the size of the microvessels to be resolved, thus significantly blurring the microvessel image and causing inaccurate measurements of blood flow velocity. Thus, the ultrasonically detected microbubble signals can be processed to remove these tissue movements.

For example, as shown in FIG. 2, tissue 202 is moved to a different tissue location 204 and target microvasculature 206 is moved to a different microvasculature location 208 due to physiological motion, such as respiration.

As another examples, an ultrasound speckle tracking based method (such as two-dimensional recursive -dimensional cross-correlation) may be used to estimate a motion vector between a reference signal and moving tissue and microvascular signals.

Referring again to FIG. 1, after tissue motion is removed from the ultrasound data, the microbubble signals are isolated in the ultrasound data, as indicated at step 106. generally, isolating the microbubble signals includes isolating the microbubble signals from background signals, such as tissue signals and signals from unchanged microbubbles that do not change between acquisition frames (e.g., when microbubbles do not move between frames). In embodiments, the microbubble signals may be isolated using frame-to-frame signal subtraction, high pass filtering along the time direction of the signals, singular value decomposition ("SVD") based filtering, and the like.

In these implementations, the ultrasound data may be filtered using a cutoff frequency that is lower than the temporal frequency of the isolated microbubble signal but higher than the temporal frequency of the background signal to isolate the microbubble signal.

As another examples, SVD-based filtering may be used in which singular value cutoff may be used to separate background signals (e.g., tissue signals and non-moving microbubble signals, which are typically projected to low-order singular values) from isolated moving microbubble signals (typically projected to medium-to-high-order singular values) — as examples, the block-wise adaptive SVD filter (the entire contents of which are incorporated herein by reference) described in co-pending PCT application No. PCT/US2017/016190 may be used to implement SVD-based filtering to extract microbubble signals.

Optionally, the isolated microbubble signal can then be denoised, as indicated at step 108. the microbubble super-resolution imaging techniques described in the present disclosure are based at least in part on locating the center position of the microbubbles.

In general, noise has similar characteristics to microbubble signals, and when noise becomes stronger in deeper regions of tissue and microbubble signals become weaker, it may be challenges to distinguish the two.

For example, by suppressing pixels having intensity values less than a selected value (e.g., -30dB relative to the maximum intensity value in the current field of view), a significant amount of background noise may be suppressed.

As another examples, a non-local mean ("NLM") denoising filter may be applied to raw, noisy microbubble data, as microbubbles move with the blood stream, microbubble movement is a deterministic event that may be continuously tracked over multiple acquisition frames, while noise events are random and do not show any trajectory-like features over multiple acquisition frames.

Another advantages of the spatio-temporal denoising filter described above is that because denoising is performed in the spatio-temporal domain, the underlying microbubble signal has little spatial blur, other denoising methods (e.g., convolutional Gaussian smoothing, Gaussian spectral apodization, wavelet thresholding, or iterative total variation ("TV") minimization) can also be used in the spatio-temporal domain to achieve similar denoising effects in implementations, the axial-temporal microbubble signal data can be used for denoising, while in other implementations, the transverse-temporal data or the complete axial-transverse-temporal 3D data can also be used for denoising.

After denoising the microbubble signal data, the microbubbles are localized in the denoised microbubble signal data, as indicated at step 110. Typically, the process includes identifying the location at which the microbubbles are located in each time frame of microbubble signal data. For example, locating the center of each isolated microbubble signal allows the movement of the microbubbles to be tracked over time. The location of the center of the localized microbubbles can also be used to construct super-resolution microvascular images and track the movement of the microbubbles to calculate hemodynamic measurements, such as blood flow velocity.

In implementations, the microbubbles may be located in the de-noised microbubble signal data using a de-blurring and de-convolution method (such as a CLEAN algorithm, a sparsity-based or entropy-based iterative regression method, a blind de-convolution method, etc.).

In some other implementations, microbubbles may be located based on a two-dimensional regression -based cross-correlation method that focuses on detecting structures that have good correlation with the point spread function ("PSF") of the ultrasound system used to acquire microbubble signal data referring now to FIG. 3, a flow chart setting forth the steps of non-limiting examples of such a method is shown.

In examples, the PSF can be obtained from simulations, e.g., the PSF can be simulated based on a multivariate Gaussian distribution.

As another examples, the PSF may be obtained from experimental measurements of very small point objects (such as objects much smaller than the ultrasound wave length).

To improve the accuracy of microbubble location, the raw microbubble signal data may optionally be interpolated to a finer data grid, as indicated at step 304. As an example, the microbubble signal data may be spatially interpolated by a selected amount, such as to ten times the original spatial resolution.

A two-dimensional normalized cross-correlation is computed between the microbubble signal data (whether or not interpolated to a finer data grid) and the provided ultrasound system PSF, as indicated by step 306. the result of this step is a 2D normalized cross-correlation coefficient map, an example of which is shown in FIG. 4. the normalized cross-correlation coefficient will be higher in regions where there is better correlation between the detected microbubble signal and the ultrasound system PSF.

By selecting the cross-correlation cutoff threshold, as indicated at step 308, features not associated with microbubbles may be rejected as indicated at step 310 to preserve signals associated with microbubbles. An example of microbubble signal data in which signals not associated with microbubbles have been rejected using this method is shown in fig. 5.

The center position of each microbubble may then be estimated, as indicated at step 312. in implementations, the center position may be estimated by identifying the local area maximum of the thresholded cross-correlation coefficient map. in other implementations, the center position may be estimated by curve fitting the cross-correlation map to find the local maximum position. an example output of localization is shown in fig. 6, where a cross indicates the estimated center position of the microbubble.

Referring again to fig. 1, after microbubbles are located, the location of microbubbles is accumulated and tracked, as indicated at step 112. in addition to being useful in generating super-resolution microvascular images and vessel dynamics measurements, the microbubble tracking process may also provide a quality control step for the microbubble location step.

As illustrated in FIG. 7, this local tracking method generally uses a local window 702 to track the movement of a single microbubble over time. As an example of , two-dimensional cross-correlation may be used to track the movement of a microbubble over time. the original microbubble signal in frame n704 is tracked in frames n +1, n +2, n +3, and n +4, as indicated by the dashed circle 706. the true microbubble movement trajectory is indicated by 708.

The drawbacks of this local window tracking based approach are that it is susceptible to other microbubble events that may occur in the same local tracking window 702. for example, another microbubbles 710 may appear in the same local tracking window 702 in frames n +2 and n + 3. in cases, microbubble tracking may be tracking this second microbubble 708, which may result in an incorrect microbubble movement trajectory indicated by 712. as a result, in the case of the local tracking approach, complex rules and models are typically built to robustly track microbubbles that may take into account all possible situations (e.g., bubble generation, bubble death, new bubble occurrence, bubble trajectory merging, bubble trajectory splitting).

For illustrative purposes, FIG. 8 shows an example of localized microbubble center positions for two consecutive frames (i.e., frame n and frame n + 1). The bipartite graph and minimum distance pairing described in this disclosure is based on the principle that each authentic microbubble signal in frame n will have and only paired microbubble signals in frame n +1, and vice versa, and that the total distance between all paired microbubbles should be minimal.

Referring now to fig. 9, a flow chart setting forth the steps of an example method for tracking microbubbles based on a bipartite graph and minimum distance pairings is shown, the method includes calculating distances between all microbubbles in th frame and all microbubbles in a second frame, as indicated at step 902. as examples, th and second frames may be temporally consecutive frames (e.g., frame n and frame n + 1). in some other examples, the th and second frames need not be temporally consecutive, but may be different time frames.

Next, the distance between microbubbles that exceeds the physiological limit of blood flow velocity is rejected, as indicated in step 904. for example, microbubbles cannot move a distance in the time interval between two frames because the blood flow indicating microbubble movement cannot exceed a velocity As examples, a blood flow limit of 100cm/s may be used.

The overall distance of all paired microbubbles is then minimized, as indicated by step 906. the minimization problem may be based on the paired distance in a thresholded distance map (such as the map shown in fig. 11. as non-limiting examples, a partial distribution algorithm may be used to pair microbubbles based on the minimum paired distance fig. 19 shows an example of a partial distribution algorithm implemented for this purpose the basic principle behind partial distribution algorithm implementation is that each microbubble will find paired microbubbles, the distance between them must be minimal among all other pairing options for both microbubbles.

As another examples, the pairing process can also be implemented as an assignment problem by assigning microbubbles in frame n to microbubbles in frame n +1, or assigning microbubbles in frame n +1 to microbubbles in frame n examples of this type of implementation can use the classical Hungarian algorithm (also known as Kuhn-Munkres algorithm) to arrive at an assignment solution.

Fig. 12 shows examples of two-part map global microbubble pairing results applied to the localized microbubble signals from fig. 8 after pairing, many microbubble signals from fig. 8 are rejected because they cannot be paired.

After bipartite graph minimum distance or other pairing process, a frame-to-frame pairing persistence control method may be used to further scrutinize microbubble pairing and tracking results.

Fig. 13A and 13B show example results of five frame-to-frame persistence control (i.e. identical microbubbles must be paired in at least five consecutive frames). fig. 13A indicates accumulated super-resolution microvascular images before persistence control, and fig. 13B shows results after persistence control, it can be seen in fig. 13B that the persistence control method rejects many noise signals, especially in deeper regions of the tissue where the noise is high.

The use of persistence control may also facilitate more robust microbubble tracking based blood flow velocity estimation, for example, after persistence control, any given microbubble may be successfully paired in n consecutive frames, where n >2 is the level of persistence control, the local blood flow velocity and direction may be estimated by the distance of movement and time interval between each pairs of consecutive frames, yielding a blood flow velocity measurement between each persistence control frame.

Referring again to fig. 1, after microbubbles have been located and tracked in the microbubble signal data, or more microvascular images are generated based on the location and tracking results, as indicated by step 114. as an example of , the microvascular images may include accumulated microbubble location maps throughout all acquisition frames. as another example of , the microvascular images may include blood flow velocity maps with blood velocity values assigned to all locations where microbubbles are detected.

The accumulated microbubble location map depicts the number of times the microbubbles appear at a location. In general, larger vessels have more microbubbles flowing through them in a given time interval, and thus they will appear brighter than smaller vessels, which have fewer microbubbles flowing through them in the same time interval. In fact, when more limited time intervals are used for in vivo imaging due to physiological motion effects and microbubble dose limitations, the resulting microbubble location map may be spatially sparse.

In such cases when the microvascular images (e.g., microbubble location map, blood flow map) are spatially sparse, the images may be processed to recover data, as indicated at step 116. As examples, a two-dimensional spatial low pass filter (e.g., 2D Gaussian smoothing filter) may be used to "fill in" or "patch" the empty regions between detected microbubble locations FIG. 14 shows an example using a 2D Gaussian smoothing filter on the sparse microbubble location map.

These potential drawbacks of these "fill-in" methods based on local smoothing filters are that the final microvascular image may be blurred.

The underlying principle of the data recovery problem is that an ideal microvascular image will display sparsity (either intrinsically, or in transform domains) simultaneously, and when transformed by a forward system, the measured dataset should be closely reflected.

Thus, as examples, a curvelet-based sparsity promoting term and a Poisson distribution-based statistical data fidelity term may be used to achieve robust microvascular data recovery_iAnd recovered microvascular data sample x_iThe probability P (x; y) (i.e., the probability of accurately seeing detected y data samples when the average recovered data sample is x) is given by:

where i is the pixel index or local block index of the microvascular data. Maximizing the probability term in equation (1) can be considered to minimize the following:

-ln(P(x；y))＝-y^Tln(x)+1^Tx+C (2)；

where C is a constant term that is generally not relevant from an optimization perspective. The cost function j (u) for this microvascular data recovery problem may then be initialized to, for example, an analysis-type canonical regression:

where Φ is the curvelet transform and ν is the curvelet coefficient. The curvelet transform decomposes the raw microvascular data into curvelet coefficients of different scales representing different directions and dimensions of the microvascular structure. For example, as shown in FIG. 15, the 4 th order curvelet transform of the microvascular structure image 1502 gives a curvelet coefficient map 1504. An inner portion 1506 of the curved-wave coefficient plot 1504 represents larger microvascular structures having lower spatial frequencies, while an outer portion 1508 of the curved-wave coefficient plot 1504 represents smaller microvascular structures having higher spatial frequencies. For example, in the direction from the central portion 1506 to the outer portion 1508, the coefficient of the curvelet represents smaller and smaller microvascular structures. The different quadrants of the curved-wave coefficient plot 1504 represent microvascular structures growing in different directions. For example, region 1510 represents a microvascular structure having a lateral growth direction, and region 1512 represents a microvascular structure having an axial growth direction.

Referring again to the minimization problem of equation (3), item on the right side of the equation is a sparsity promoting item, the second and third items are data fidelity items the goal of minimizing the cost function is to achieve optimal data recovery as examples, a multiplier alternating direction method ("ADMM") technique can be used to solve equation (3) with which equation (3) is first recast as a functionally equivalent constraint optimization problem,

thereby making v ═ Φ u (4).

Subsequently, a corresponding (e.g., scaled dual form) augmented lagrangian function of equation (4) is constructed, which is given by:

to solve equation (4), the saddle point of equation (5) may be determined, which may be effectively implemented by performing the following gaussian-Seidel (Gauss-Seidel) iterations until convergence, as examples:

ξ_k+1＝ξ_k-(v_k+1-Φu_k+1) (6)。

for the minimization term of item , the derivative of L (with respect to each pixel or small local block with a set of pixels) is derived with respect to u and is forced to zero:

for frequency-wound curvelet transformation, phi^*I (i.e., the adjoint and inverse curvelet operations are equivalent), equation (7) may result,

and therefore the number of the first and second channels,

for the second term minimization, soft thresholding of the curvelet coefficients may be used,

v_k+1＝T_λ/μ{Φu_k+1+ξ_k} (10)。

substituting equations (9) and (10) into equation (6) yields the following three iterative steps:

v_k+1＝T_λ/μ{Φu_k+1+ξ_k}

ξ_k+1＝ξ_k-(v_k+1-Φu_k+1) (11)。

FIG. 16 shows an example of an initial microvascular image 1602 and a microvascular image 1604 that has been processed using the data recovery techniques described above it can be seen that the data recovery method robustly recovers the locations of missing microvascular data without obscuring the microvascular structure in some implementations the data recovery results can be further improved by applying constraints to the curvelet coefficient soft thresholding step for example in the example shown in FIG. 16 most of the microvascular structure is growing in the top to bottom direction so more curvelet coefficients that promote the top to bottom microvascular structure direction (e.g., the coefficients associated with region 1512 in FIG. 15, etc.) can be retained and more curvelet coefficients corresponding to other microvascular structure directions can be rejected.

Referring again to fig. 1, after processing, the microvascular images may be displayed to a user or stored for later use, such as for later analysis, as indicated by step 118. in implementations, microvascular morphological measurements (e.g., blood vessel density and blood vessel tortuosity) may be estimated from the microvascular images as another examples, microvascular hemodynamic measurements (e.g., blood flow velocity and blood flow) may be estimated from the microvascular images.

The blood vessel tortuosity may be measured by a method such as a distance metric that provides a ratio of the actual path length of the blood vessel normalized by the linear distance between the vessel curve endpoints to . the microvascular blood flow velocity may be averaged over the entire region of interest to represent the perfusion index, or the blood flow velocity may be integrated by the total cross-sectional area of the microvasculature within the region of interest to derive a cross-sectional blood flow rate that may represent the perfusion index, or the blood flow velocities from all of the microvasculature may be used to generate a histogram (e.g., the x-axis represents the blood flow velocity and the y-axis represents the total number of pixels with a blood flow velocity of in each x-axis bin) to represent the perfusion index.

The more frames that are accumulated, the more microbubble motion trajectories that can be seen at time, for example, the higher frame accumulation can be used to visualize slower-flowing vessels, while the lower frame accumulation can be used to visualize faster-flowing vessels.

In addition to the above described method for isolating microbubble signals, microbubble control pulses may be implemented to modulate microbubbles in a particular microvasculature so that the isolated microbubble signals may be more easily acquired. As an example, microbubble control pulses may enable higher transmit power ultrasound pulses (e.g., mechanical index >1.0) to locally or globally destroy microbubbles to reduce the number of microbubbles, thus increasing the chances of observing the isolated microbubble signal.

The microbubble control pulses may comprise an series of focused ultrasound beams that are transmitted at localized or arbitrary locations within the field of view, which may be set by the user, as examples, a series of focused ultrasound beams aimed at a fixed focal depth may be used, as another examples, two series of focused ultrasound beams may be used, where each series of focused ultrasound beams is aimed at of two different focal depths.

As shown in FIG. 17, microbubble control pulses 1702 may be inserted into the detection pulses 1704 to modulate microbubbles continuously.A wait time (e.g., several hundred microseconds) may be inserted just after the microbubble control pulses 1702 to attenuate the ultrasound signal generated by the microbubble control pulses 1702. the power of the microbubble control pulses 1702 may be adjusted during microbubble signal acquisition to maximize the amount of isolated microbubble signal.

Fig. 18 shows an example of an ultrasound system 1800 in which the methods described in this disclosure may be implemented. The ultrasound system 1800 includes a transducer array 1802, the transducer array 1802 including a plurality of individually driven transducer elements 1804. The transducer array 1802 may include any suitable ultrasound transducer array, including a linear array, a curved array, a phased array, and the like.

Ultrasonic energy (e.g., echoes) reflected back to the transducer array 1802 from a subject or subject under study are converted to electrical signals (e.g., echo signals) by every transducer elements 1804 and may be individually applied to the receivers 1808 through sets of switches 1810 the transmitter 1806, receivers 1808, and switches 1810 operate under control of a controller 1812, which may include or more processors, as examples, the controller 1812 may include a computer system.

In configurations, transmitter 1806 may also be programmed to transmit a diverging wave, a spherical wave, a cylindrical wave, a plane wave, or a combination thereof, furthermore, transmitter 1806 may be programmed to transmit a spatially or temporally encoded pulse.

The receiver 1808 may be programmed to implement a suitable detection sequence for the imaging task at hand in embodiments, the detection sequence may include or more of progressive scan, complex plane wave imaging, synthetic aperture imaging, and complex diverging beam imaging.

Thus, in configurations, the transmitter 1806 and receiver 1808 may be programmed to achieve a high frame rate, for example, a frame rate associated with an acquisition pulse repetition frequency ("PRF") of at least 100Hz may be achieved in configurations, the ultrasound system 1800 may sample and store a set of at least hundred echo signals in the temporal direction.

Using techniques described in this disclosure or otherwise known in the art, the controller 1812 may be programmed to design an imaging sequence in embodiments, the controller 1812 receives user input defining various factors used in the design of the imaging sequence.

The scan may be performed by setting the switches 1810 to their transmit positions, thus directing the transmitter 1806 to turn on momentarily to excite every transducer elements 1804 during a single transmit event according to a designed imaging sequence the switches 1810 then may be set to their receive positions and subsequent echo signals produced by the transducer elements 1804 in response to or more detected echoes are measured and applied to the receiver 1808 the individual echo signals from the transducer elements 1804 may be combined in the receiver 1808 to produce a single echo signal the image produced by the echo signals may be displayed on the display system 1814.

In embodiments, receiver 1808 may include a processing unit to process the echo signals or images generated from the echo signals, which may be implemented by a hardware processor and memory, such a processing unit may, as examples, isolate microbubble signals to produce microbubble signal data, locate microbubbles in the microbubble signal data, track microbubble locations in a time frame, accumulate microbubble locations, and produce microvascular images using the methods described in this disclosure.

The disclosure has described preferred embodiments, and it should be understood that many equivalents, substitutions, variations, and modifications, aside from those expressly stated, are possible and within the scope of the present invention.