Differential prediction for aberration correction in ultrasound therapy

Abstract

Various methods for delivering ultrasound energy to a target area during a treatment or diagnostic procedure include: implementing an adjustment mechanism with a machine learning model trained based on input vectors corresponding to differences and / or ratios of parameter values ​​between a plurality of transducer elements to generate one or more parameter values ​​for each of the transducer elements to compensate for anticipated beam aberrations caused by intermediate tissue; and activating the transducer elements according to the corresponding parameter values ​​to generate an optimal focused area at the target area during the treatment or diagnostic procedure.

Description

[0001] Cross-reference to related applications

[0002] This application claims priority and benefit to U.S. Provisional Patent Application No. 62 / 985,587, filed March 5, 2019, which is incorporated herein by reference in its entirety. Technical Field

[0003] This invention generally relates to ultrasound therapy, and more particularly to systems and methods for correcting aberrations that affect the delivery of ultrasound therapy. Background Technology

[0004] Tissues, such as benign or malignant tumors, organs, or other areas of the body, can be treated invasively by surgical removal of the tissue, or minimally invasively or entirely non-invasively by using methods such as thermal ablation. Both methods can effectively treat certain localized conditions, but the procedures involved are delicate to avoid damaging or harming other healthy tissues.

[0005] Thermal ablation using focused ultrasound is particularly attractive for treating lesions surrounded or adjacent to healthy tissue or organs because the effect of ultrasound energy can be confined to a well-defined target area. Due to the relatively short wavelength, ultrasound energy can be focused onto areas with a cross-section of only a few millimeters (e.g., as small as 1.5 millimeters (mm) at 1 MHz). Furthermore, since acoustic energy typically penetrates soft tissue well, intermediate anatomical structures generally do not obstruct the definition of the desired focal zone. Therefore, ultrasound energy can be focused on small targets to ablate lesions while minimizing damage to surrounding healthy tissue.

[0006] To focus ultrasonic energy onto a desired target, a drive signal can be sent to an acoustic transducer with several transducer elements, causing constructive interference in the focal zone. At the target, sufficient acoustic intensity can be delivered to heat the tissue until necrosis occurs, i.e., until the tissue is destroyed. Preferably, non-target tissue along the path of the acoustic energy outside the focal zone (“through zone”) is exposed to a low-intensity acoustic beam and will therefore be heated only to a minimal extent (if any), thereby minimizing damage to tissue outside the focal zone.

[0007] The non-invasive nature of ultrasound surgery is particularly attractive for the treatment of brain tumors. However, the treatment challenges posed by the human skull anatomy limit the clinical implementation of ultrasound therapy. Barriers to transcranial ultrasound procedures include strong attenuation and distortion caused by irregularities in skull shape, density, and sound velocity, which can disrupt the ultrasound focus and / or reduce the ability to spatially register diagnostic image information.

[0008] Therefore, solutions have been proposed to adjust for the effects of ultrasound energy absorbed by non-target tissue. In a representative method described in U.S. Patent Publication No. 2020 / 0085409 (the entire disclosure of which is hereby incorporated herein by reference), a patient-specific 3D skull replica is created and placed in an environment similar to that used for treating the patient; detector devices (e.g., hydrophones) can be deployed at target regions within the skull replica to measure acoustic signals from each ultrasound transducer element during a simulated treatment sequence. By analyzing the measured signals, corrections can be determined for ultrasound parameters (e.g., amplitude and / or phase shift) associated with each transducer element. During treatment, the ultrasound transducer elements can be activated according to the corrected ultrasound parameters to compensate for beam aberrations caused by the skull; this can thereby produce high-quality focusing and / or improve ultrasound beamforming at the target region. Minimizing the area of ​​the focused region increases the peak acoustic intensity at the target region and also allows for shaping of the ultrasound beam emitted from the transducer elements to avoid or minimize exposure of non-target tissue to the treatment energy. However, creating 3D skull replicas involves costs and time, leading practitioners to seek alternative technologies to correct for fasciculations caused by patient-specific anatomical features.

[0009] One technique utilizes machine learning. Predictors applicable to different patients can be developed based on the optimal focusing configuration identified for many specific patients, and these can be used as a training set. The trained predictor can estimate the optimal corrections for the amplitude, phase, and / or time delay of all transducer elements used in the treatment procedure.

[0010] One obstacle to using machine learning to predict optimal phase and / or amplitude settings is measurement bias, a general term used in this paper to describe various errors that share a common source but may differ for different measurements. "Error" refers to deviations from the optimal reference setting, not measurement inaccuracy. One of these errors can be a constant bias term. For example, suppose we are trying to predict the optimal focus configuration via regression, using, for instance, the following equation:

[0011]

[0012] in It predicts the phase, {x1,x2…x n} is the feature vector, and θ is the weight. To obtain a prediction, the deviation θ0 must be known, but without accurate knowledge of, for example, water and tissue temperatures, this value may not be accurately estimated for a specific measurement on the training set. Accurate knowledge of water and tissue temperatures can be achieved by creating a skull replica as described above. Another possible error is caused by the position of each transducer element relative to the target and is position-specific (e.g., due to inaccurate knowledge of the precise transducer position in space, or due to non-uniform water and skull temperatures, or due to inaccurate knowledge of the target position), causing different transducer elements contributing to the measurement to cause different error values. Other errors may arise from the input features provided to the predictor, for example, due to image filters used to process computed tomography (CT) scans. The input vector is typically the skull characteristics and / or images of the medium through which the element passes on its way to the target. Using data measured at different ultrasound frequencies may introduce further errors.

[0013] As with any machine learning exercise, effective training data is crucial for developing a predictor for optimal focusing. The aforementioned measurement errors hindered the creation of a training dataset that consisted of numerous, consistent, and learnable measurements from many patients, excised skulls, and / or skull replicas. Therefore, it was necessary to mitigate or eliminate the inaccuracies in predictive transducer setup caused by measurement errors, enabling the effective use of machine learning to provide high-quality focusing in the target region. Summary of the Invention

[0014] This invention provides systems and methods for focusing an ultrasound beam traversing a tissue (e.g., the human skull) with irregular structure, shape, density, and / or thickness onto a target area with high-quality focusing. For ease of reference, the following description relates only to ultrasound therapeutic procedures; however, it should be understood that this method is generally applicable to ultrasound imaging procedures as well. Furthermore, although the description herein relates to an ultrasound beam traversing the human skull, the methods described in conjunction with various embodiments can be applied to determine beam aberrations caused by any part of the human body (e.g., ribs), thereby allowing adjustment of parameter values ​​characterizing the sound beam (e.g., phase shift and / or amplitude) to compensate for the aberrations.

[0015] The methods described herein advantageously mitigate or eliminate measurement errors that share a common source but may affect different transducer elements in different ways, such that different measurements may exhibit different errors in the measured parameters (e.g., phase, amplitude, and / or time delay). In various embodiments, the predicted values ​​are relationships between elements, such as relationships between pairs of elements, rather than direct predictions of the parameters of each transducer element. When the predicted values ​​are relationships (e.g., differences or ratios) between values ​​associated with elements having the same input source, measurement errors caused by the input (e.g., by the type of imaging and image processing used to characterize intermediate tissues) are reduced, and measurement errors caused by different element locations are reduced by increasing constraints on the distances between multiple (e.g., pairs) elements; in particular, when the relationship is a difference or ratio between parameter predictions of selected elements, the predicted bias terms are canceled out—e.g., in the case of element pairs, the predicted values ​​are the relationships between the expected parameters of the two elements in each pair (rather than the actual values ​​for each individual element).

[0016] In the case of amplitude prediction, the measurement is biased by a multiplicative factor, which is canceled out when a ratio is used. In the case of phase prediction, the measurement is biased by an additive factor. Assuming the bias term is mainly associated with different uniform water temperatures, equal element-to-target distances ensure that the bias term is canceled out; if the difference between these distances is much smaller than the average element-to-target distance, the term can be ignored.

[0017]

[0018] in This indicates that in the skull medium at temperature T i The phase shift measured by the lower element j; This indicates that in an aqueous medium at temperature T i The phase shift measured by the lower element j; This indicates that in an aqueous medium at temperature T i The phase shift caused by Δd is defined as the element-target distance Δd = |target-el1| - |target-el2|. For example, in a representative therapeutic system, the distance between the transducer element and its natural focus is 150 mm, while the typical brain target distance is a few millimeters from the natural focus (therefore Δd is at most twice that). The temperature range of 15°–25° corresponds to a change in ultrasound velocity of approximately 2.5% in water. Therefore, the error caused by this temperature change will be reduced from 2.5% at 150 mm (approximately 1.6λ for a 650 kHz transducer) to 2.5% at a few millimeters (less than 0.1λ).

[0019] For phase shift, the prediction difference represents the difference between the phase shifts of elements that may cause constructive interference at the focal point. All differences relate to the same focal point; if element pairs are considered independently, rather than as a whole, it is impossible to obtain focus for the entire system. It is the combination of all differences that produces accurate correction for each transducer element, partly because redundancy is enforced during the inference phase (i.e., the same element should appear in multiple pairs). This redundancy produces an overdetermined set of equations that can be analyzed as an optimization problem. Constant deviations affecting all elements in the transducer have no effect on focus quality or focus position. Therefore, a reference phase can be assigned to selected elements, and then the phase differences can be combined into a table specifying the phase correction for each element using conventional optimization algorithms. Deviations caused by measurement location or non-uniform temperature may produce small shifts in the focal position; if clinically insignificant, these can be corrected or ignored. As used in this paper, “clinically insignificant” means that the effect on the tissue is not expected (sometimes means that the expected effect does not exist), and the clinician considers the effect to be insignificant, whether temporary or permanent, such as before injury or other adverse clinical effects occur.

[0020] One embodiment uses features corresponding to two sets of skull characteristics (e.g., thickness, density, angle) obtained from CT images (or any other skull imaging technique), each set associated with a different transducer element; therefore, a 2k-length vector is used instead of a k-length vector as input for element phase shift prediction. Another option is to use two sets of CT blocks of the skull medium for each element, for example, combined in a fourth dimension. Yet another option is to use a Siamese network, where each component receives input corresponding to a single element, and its total output is based on a comparison of component outputs. Other embodiments utilize a single block corresponding to the differences in the CT blocks of the element for each element pair. Groups consisting of more than two elements can be utilized, in which case the features will be a combination of skull characteristics relative to each element.

[0021] The predictor's output, i.e., element-specific adjustments (or corrections) for ultrasound parameters such as amplitude and / or phase, can be stored in a data structure, such as a table relating parameter values ​​to associated transducer elements used to produce optimal focus at the target. As used herein, the term "optimal" generally refers to a more significant improvement than that obtained using prior art methods (e.g., more than 10%, more than 20%, or more than 30%), but does not necessarily mean that theoretically possible optimal focusing characteristics have been achieved.

[0022] Therefore, various embodiments of the invention advantageously mitigate or eliminate various measurement errors in order to more accurately predict the optimal settings (e.g., phase and / or amplitude) associated with the transducer elements, taking into account the effects of intermediate tissue on the acoustic energy exerted on the target tissue (e.g., defocus, acoustic coupling mismatch, etc.). In one embodiment, transducer adjustments based on predictions are dynamically updated during treatment to ensure treatment efficiency during the ultrasound procedure.

[0023] Therefore, in one aspect, the present invention relates to a system for delivering ultrasound energy to a target region during a treatment or diagnostic procedure. In various embodiments, the system comprises: an ultrasound transducer having a plurality of transducer elements, wherein at least some of the transducer elements are designated to be in action during the treatment or diagnostic procedure; and an adjustment mechanism comprising a machine learning model trained based on an input vector corresponding to the differences and / or ratios of parameter values ​​between the plurality of transducer elements. The adjustment mechanism is configured to: (i) receive a numerical quantity characterizing the tissue between the in-action transducer elements and the target region (e.g., tissue density and / or thickness); and (ii) based thereon, generate one or more parameter values ​​for each in-action transducer element to compensate for anticipated beam aberrations. A controller is configured to activate the in-action transducer elements according to the corresponding parameter values ​​to generate an optimal focused area at the target region during the treatment or diagnostic procedure. In one embodiment, the generated parameter values ​​specify corrections for amplitude, phase, frequency, duty cycle, ultrasound processing mode, and / or time delay.

[0024] The machine learning model can be a neural network. Alternatively, the neural network can be trained based on input vectors, each with parameters associated with a pair of transducer elements. In one embodiment, the adjustment mechanism includes a lookup table of adjustment values ​​generated by a machine learning algorithm. The machine learning algorithm is a neural network trained based on input vectors, each with parameters associated with more than two transducer elements. In some embodiments, the adjustment mechanism is further configured to generate parameter values ​​at least in part based on one or more geometric parameters associated with the associated transducer elements. Geometric parameters can be, for example, angles relative to the skull surface between the associated transducer elements and the target region. In some embodiments, the differences and / or ratios of the ultrasound parameter values ​​are obtained using automated focused therapy measurements or ex vivo or replica skull measurements.

[0025] In some embodiments, the target region comprises multiple portions, and the resulting optimal focal zone is located in a first portion of the target region; the controller is further configured to predict using a physical model: (i) the real-time temperature of a second portion of the target region different from the first portion or a non-target region, (ii) the real-time temperature of the focal zone, (iii) the focal shape, and / or (iv) treatment success. Additionally, the controller may be further configured to cause the operating transducer elements to: (i) emit ultrasonic waves toward the target region, and (ii) measure the reflection of ultrasonic waves from the target region; classify each operating transducer element as a measuring transducer element or a non-measuring transducer element based on the measurements; and update the adjustment mechanism at least in part based on the measurements provided by the measuring transducer elements.

[0026] In one embodiment, the updated adjustment mechanism is configured to: (i) receive a numerical quantity characterizing the tissue between the non-measuring transducer element and the target region; and (ii) based on this, generate one or more updated parameter values ​​for each non-measuring transducer element to compensate for the expected beam aberration. The controller is then further configured to activate the non-measuring transducer element according to the corresponding updated parameter value.

[0027] In various embodiments, the controller is further configured to cause the operating transducer element to: (i) emit ultrasonic waves toward the target region and (ii) measure the reflection of ultrasonic waves from the target region; and, at least in part based on the measurement, update the adjustment mechanism. The updated adjustment mechanism may be configured to: (i) receive a numerical quantity characterizing the tissue between the operating transducer element and a secondary target region different from the target region, and (ii) based thereon, generate one or more updated parameter values ​​for each operating transducer element to compensate for anticipated beam aberrations. The controller is then further configured to activate the operating transducer element according to the corresponding updated parameter value to generate an optimal focusing area at the secondary target region.

[0028] In another aspect, the present invention relates to a method for generating correction values ​​for delivering ultrasound energy through intermediate tissue via an ultrasound transducer comprising a plurality of transducer elements, the correction values ​​correcting for beam aberrations. In various embodiments, the method includes the steps of: training a machine learning model based on received input vectors, each input vector containing (i) differences and / or ratios of ultrasound parameters between the plurality of transducer elements and (ii) numerical quantities (e.g., tissue density and / or thickness) characterizing tissue between the plurality of transducer elements and a target region, and for generating correction values ​​of the ultrasound parameters as output to compensate for expected beam aberrations through the intermediate tissue. Training may include enabling the machine learning model to learn relationships between input vector values ​​and corresponding correction values. A plurality of input vectors are provided to the machine learning model, each input vector having a numerical quantity characterizing patient tissue between the plurality of transducer elements to be activated and the target region of the patient. Based on the provided input vectors, correction values ​​of the ultrasound parameters are received as output of the machine learning model. The method may further include activating transducer elements according to the corresponding correction values ​​of the parameter values ​​to generate an optimal focal zone at the target region. In one implementation, the correction value specifies corrections for amplitude, phase, frequency, duty cycle, ultrasonic processing mode, and / or time delay.

[0029] The machine learning model can be a neural network. Alternatively, the neural network can be trained based on input vectors, each input vector having parameters associated with a pair of transducer elements. In one embodiment, the difference and / or ratio of the ultrasound parameters is obtained using automated focused therapy measurements or ex vivo or replica skull measurements. In various embodiments, the input vectors further include one or more geometric parameters associated with each transducer element; the machine learning model is further trained to generate corrected values ​​for the ultrasound parameters, at least in part, based on the geometric parameters. The geometric parameters may include, for example, angles relative to the skull surface between each transducer element and the target region.

[0030] In various embodiments, the target region comprises multiple portions, and the resulting optimal focal zone is located in a first portion of the target region. The method further includes using a physical model to predict: (i) the real-time temperature of a second portion of the target region different from the first portion or a non-target region, (ii) the real-time temperature of the focal zone, (iii) the focal shape, and / or (iv) treatment success. Additionally, the method may further include activating at least some of the transducer elements to: (i) emit ultrasound waves toward the target region and (ii) measure the reflection of ultrasound waves from the target region; classify each transducer element as a measuring transducer element or a non-measuring transducer element based on the measurements; and update the machine learning model at least in part based on the measurements provided by the measuring transducer elements. In one embodiment, the method further includes providing the updated machine learning model with a numerical quantity characterizing the tissue between the non-measuring transducer elements and the target region; and for each non-measuring transducer element, receiving one or more updated parameter values ​​to compensate for anticipated beam aberrations. Additionally, the method may further include activating the non-measuring transducer elements according to the corresponding updated parameter values.

[0031] In some embodiments, the method further includes activating at least some of the transducer elements to: (i) emit ultrasonic waves toward a target region and (ii) measure the reflection of ultrasonic waves from the target region; and updating a machine learning model, at least in part, based on the measurements. Additionally, the method may further include providing the updated machine learning model with numerical quantities characterizing tissue between the transducer elements and secondary target regions different from the target region; and for each transducer element, receiving at least one updated parameter value to compensate for anticipated beam aberrations. In one embodiment, the method further includes activating the transducer elements according to the corresponding updated parameter values ​​to generate an optimal focusing area at the secondary target region.

[0032] Another aspect of the invention relates to a method for delivering ultrasound energy through intermediate tissue using an ultrasound transducer having multiple transducer elements, employing a machine learning model trained with input vectors. Each input vector has (i) differences and / or ratios of ultrasound parameters between the multiple transducer elements and (ii) a numerical quantity (e.g., tissue density and / or thickness) characterizing the tissue between the transducer element and the target region, and is used to generate corrected values ​​for the ultrasound parameters as output to compensate for anticipated beam aberrations through the intermediate tissue. The training involves enabling the machine learning model to learn the relationship between the values ​​of the input vectors and the corresponding corrected values. In various embodiments, the method includes the steps of: providing a multiple input vector to the machine learning model, each input vector having a numerical quantity characterizing the patient tissue between the multiple transducer elements to be activated and the target region of the patient; and treating the patient by activating the transducer elements to deliver ultrasound energy according to the corrected values ​​generated by the machine learning model based on the provided input vectors in order to create an optimal focal zone at the target region. In one embodiment, the corrected values ​​specify corrections for amplitude, phase, frequency, duty cycle, ultrasound processing mode, and / or time delay.

[0033] The machine learning model can be a neural network. Alternatively, the neural network can be trained based on input vectors, each input vector having parameters associated with a pair of transducer elements. In some embodiments, the input vectors further include one or more geometric parameters associated with each transducer element; the machine learning model is further trained to generate corrected values ​​for the ultrasound parameters, at least in part, based on the geometric parameters. The geometric parameters can be, for example, angles relative to the skull surface between each transducer element and the target region. In one embodiment, the difference and / or ratio of the ultrasound parameters are obtained using automated focused therapy measurements or ex vivo or replica skull measurements.

[0034] In various embodiments, the target region comprises multiple portions, and the resulting optimal focal zone is located in a first portion of the target region. The method further includes using a physical model to predict: (i) the real-time temperature of a second portion of the target region different from the first portion or a non-target region, (ii) the real-time temperature of the focal zone, (iii) the focal shape, and / or (iv) treatment success. Additionally, the method may further include activating at least some of the transducer elements to: (i) emit ultrasound waves toward the target region and (ii) measure the reflection of ultrasound waves from the target region; classify each transducer element as a measuring transducer element or a non-measuring transducer element based on the measurements; and update the machine learning model at least in part based on the measurements provided by the measuring transducer elements. In one embodiment, the method further includes providing the updated machine learning model with a numerical quantity characterizing the tissue between the non-measuring transducer elements and the target region; and for each non-measuring transducer element, receiving one or more updated parameter values ​​to compensate for anticipated beam aberrations. The method may further include activating the non-measuring transducer elements according to the corresponding updated parameter values.

[0035] In some embodiments, the method further includes activating at least some of the transducer elements to: (i) emit ultrasonic waves toward a target region and (ii) measure the reflection of ultrasonic waves from the target region; and updating a machine learning model, at least in part, based on the measurements. Additionally, the method may further include providing the updated machine learning model with numerical quantities characterizing the tissue between the transducer elements and secondary target regions different from the target region; and for each transducer element, receiving one or more updated parameter values ​​to compensate for anticipated beam aberrations. Furthermore, the method may include activating the transducer elements according to the corresponding updated parameter values ​​to generate an optimal focusing area at the secondary target region.

[0036] As used herein, the term "generally" means ±10%, and in some embodiments, ±5%. Throughout this specification, references to "an example," "an instance," "an embodiment," or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with that example is included in at least one instance of the invention. Therefore, the phrases "in an example," "in an example," "an embodiment," or "an embodiment" appearing in different places throughout this specification do not necessarily refer to the same example. Furthermore, particular features, structures, routines, steps, or characteristics may be combined in any suitable manner in one or more instances of the technology. The headings provided herein are for convenience only and are not intended to limit or interpret the scope or meaning of the claimed technology. Attached Figure Description

[0037] In the accompanying drawings, similar reference numerals are generally used throughout different views to refer to the same parts. Furthermore, the drawings are not necessarily drawn to scale; the focus is usually on illustrating the principles of the invention. In the following description, various embodiments of the invention are described with reference to the following drawings, wherein:

[0038] Figure 1 An exemplary ultrasound system according to various embodiments of the present invention is schematically depicted;

[0039] Figure 2 The tissue layers of the human skull are shown schematically.

[0040] Figure 3 The prediction module according to various embodiments of the present invention and the manner of training and using the prediction module are illustrated schematically.

[0041] Figure 4 An imaging block of the skull is schematically depicted according to various embodiments of the present invention;

[0042] Figure 5 A siamese neural network for training a prediction model is schematically depicted according to various embodiments of the present invention;

[0043] Figure 6A This is a flowchart of various embodiments of the present invention, illustrating an exemplary method for training a prediction model to generate corrections to ultrasonic parameter values ​​associated with transducer elements;

[0044] Figure 6B This is a flowchart of various embodiments of the present invention, illustrating an exemplary method for delivering ultrasound energy to a target area during a treatment or diagnostic procedure. Detailed Implementation

[0045] Figure 1 An exemplary ultrasound system 100 for focusing ultrasound waves through the skull onto a target region 101 is shown. However, those skilled in the art will understand that the ultrasound system 100 described herein can be applied to any part of the human body. In various embodiments, system 100 includes a phased array 102 of transducer elements 104, a beamformer 106 driving the phased array 102, a controller 108 communicating with the beamformer 106, and a frequency generator 110 providing input electronic signals to the beamformer 106.

[0046] Array 102 may have a curved (e.g., spherical or parabolic) shape, making it suitable for placement on or near the surface of the skull or other body parts (e.g., separated by a water-filled pad), or may comprise one or more planar or other shaped segments. Depending on the application, its size can vary between millimeters and tens of centimeters. The transducer elements 104 of array 102 may be piezoelectric ceramic elements and may be mounted in silicone rubber or any other material suitable for suppressing mechanical coupling between elements 104. Piezoelectric composite materials or any material generally capable of converting electrical energy into acoustic energy may also be used. To ensure maximum power delivery to transducer elements 104, elements 104 may be configured for electrical resonance at 50 Ω, matched to the input connector impedance.

[0047] Transducer array 102 is coupled to beamformer 106, which drives individual transducer elements 104 such that they together generate a focused ultrasonic beam or ultrasonic field. For n transducer elements, beamformer 106 may contain n driver circuits, each circuit comprising an amplifier 118 and a phase delay circuit 120, or a combination thereof; the driver circuits drive one of the transducer elements 104. Beamformer 106 receives a radio frequency (RF) input signal, typically in the range of 0.1 MHz to 10 MHz, from frequency generator 110, which may be, for example, a DS345 generator available from Stanford Research Systems. The input signal may be split into n channels for the n amplifiers 118 and delay circuits 120 of beamformer 106. In some embodiments, frequency generator 110 is integrated with beamformer 106. The radio frequency generator 110 and the beam shaper 106 are configured to drive the individual transducer elements 104 of the transducer array 102 at the same frequency, but with different phases and / or amplitudes, so that the transducer elements 104 together form a "phased array".

[0048] Acoustic waves / pulses emitted from transducer element 104 form an acoustic beam. Typically, the transducer element is driven such that the waves / pulses converge in a focal zone within target tissue 101. Within the focal zone, the beam's wave energy is absorbed (at least partially) by the tissue, generating heat and raising the tissue temperature for therapeutic and / or diagnostic purposes. For example, the tissue can be heated to the point of cell denaturation and / or ablation. Alternatively or additionally, variations in the beam or beam can produce cavitation, interact with air bubbles in the tissue (causing mechanical effects or acoustic emissions), and / or stimulate or inhibit neural activity. For effective treatment of the target tissue, the acoustic beam must be precisely focused onto the target location 101 to avoid damaging healthy tissue surrounding the target area. Reference Figure 2A typical human skull 200 is non-uniform and has multiple tissue layers, including an outer layer 202, a bone marrow layer 204, and an inner or cortical layer 206; each layer of the skull 202 may be highly irregular in shape, thickness, and density, and is unique to each patient. As a result, when ultrasound / pulses emitted from system 100 encounter the skull 200, beam scattering, absorption, reflection, and / or refraction may occur due to tissue non-uniformity; this can lead to beam aberrations, which may then distort the focus and reduce intensity, thus affecting treatment efficiency. Therefore, it is desirable to adjust the parameters of the drive signal associated with the transducer elements (e.g., phase shift a1-a1). n and / or amplification or attenuation factor α1-α n This is done to compensate for acoustic image aberrations, thereby improving the focusing characteristics of the target area 101.

[0049] Typically, controller 108 can be used to calculate amplification factor and phase shift, and the controller can provide the calculation functionality through software, hardware, firmware, hardwiring, or any combination thereof. For example, controller 108 can utilize a general-purpose or special-purpose digital data processor programmed in a conventional manner with software to determine a set of baseline parameters (e.g., frequency, phase shift, and / or amplification factor) for transducer element 104 without requiring excessive experimentation. Controller 108 can determine the parameters based on information about the characteristics of the skull (e.g., structure, thickness, density, etc.) and their effect on sound energy propagation. (See again...) Figure 1 In one embodiment, such information is obtained from an imager 112, which may be a magnetic resonance (MR) imaging device, a computed tomography (CT) device, a positron emission tomography (PET) device, a single-photon emission computed tomography (SPECT) device, or an ultrasound device. Image acquisition may be three-dimensional (3D), or alternatively, the imager 112 may provide a set of two-dimensional (2D) images suitable for reconstructing a three-dimensional image of the target region 101 and / or other regions (e.g., the region surrounding the target 101, the region in the passageway between the transducer and the target, or another target region). Image manipulation functions may be implemented in the imager 112, the controller 108, or a separate device. The controller 108 may be used to calculate the amplification factor and phase shift, and the controller may provide the relevant calculation functions through software, hardware, firmware, hardwiring, or any combination thereof. For example, the controller 108 may utilize a general-purpose or special-purpose digital data processor programmed in software in a conventional manner and determine the frequency, phase shift, and / or amplification factor of the transducer element 104 without excessive experimentation. In some embodiments, the controller calculations are based on information about the characteristics (e.g., structure, thickness, density, etc.) of the intermediate tissue located between the transducer 102 and the target 101 (e.g., the passage zone) and its effect on acoustic energy propagation.

[0050] In some embodiments, ultrasound therapy involves acoustic reflectors (e.g., microbubbles). For example, microbubbles can be introduced via acoustic energy generation and / or through system injection for autofocusing. Ultrasonic waves emitted from all (or at least some) of the transducer elements 104 are reflected by the reflectors; the reflected signals can be detected by acoustic signal sensors and / or transducer elements 104. The measured signals can then be provided to controller 108 to obtain information associated with the reflections, such as amplitude and / or phase; these can be compared with the amplitude and / or phase associated with the ultrasonic waves emitted from transducer elements 104. Based on the deviations between them, the drive signals of transducer elements 104 can be adjusted to compensate for the deviations, thereby improving focusing characteristics. In some embodiments, this autofocusing procedure is performed iteratively until optimal focusing characteristics are achieved. For example, methods for automatically focusing an ultrasound beam at a target region are provided in PCT Publications WO 2018 / 020315 and WO 2020 / 128615; and methods for generating microbubbles and / or introducing microbubbles into target region 101 are provided in PCT Publications WO 2018 / 020315, WO 2019 / 116107, WO 2019 / 058171, WO 2019 / 116097, WO 2019 / 002947 and WO 2019 / 116095, and U.S. Patent Publications 2019 / 0083065 and 2019 / 0178851. The entire contents of the foregoing applications are incorporated herein by reference.

[0051] In various embodiments, reference is made to Figure 1 and 2 The adjustment mechanism 126 may include an automatic focus correction database 160 that correlates anatomical features identified in the images created by the imager 112 with corrections for one or more ultrasound parameters, such as frequency, amplitude, and / or phase. These features may include measurements of the thickness and / or density of layers 202, 204, 206, and other intermediate tissue layers (estimated based on automated analysis of CT or MR images), and the angle of the beam output of the transducer element relative to the patient's skull surface 208 (where the beam extends from the transducer element to the desired focal location). For example, methods for determining beam-skull angles are provided in U.S. Patents Nos. 8,617,073 and 10,456,603, the entire contents of which are incorporated herein by reference.

[0052] Adjustment mechanism 126 utilizes these features to modify baseline ultrasound parameters based on a trained predictor. In practice, adjustment mechanism 126 can directly implement the predictor, for example, as a regression module operating as described below; based on the feature values, the regression module calculates appropriate adjustments to one or more ultrasound parameters associated with one or more transducer elements based on its prior training. In other embodiments, the relationship between features and corresponding ultrasound parameters can be stored as a database with entries specific to each transducer element. These entries are obtained by providing a series of inputs to a trained predictor, specifically, a series of inputs sufficient to cover situations a patient might encounter and a number sufficient to facilitate interpolation between database entries, where one or more input feature values ​​lie between stored values. Appropriate corrections are obtained through feature-value lookups (and, if necessary, interpolation), where at least some feature values ​​are specific to transducer elements 104 (based on their respective geometric relationships to target 101 and / or skull 200). It should be noted that, more generally, ultrasound signals can be wrapped with smoothing windows (e.g., Hamming windows) to improve signal quality.

[0053] More specifically, the predictor can be used to provide initial focus correction when initiating an autofocus procedure, and / or for elements with missing or known incorrect measurements, and / or for aiming at different locations. In the latter two cases, the predictor can be adjusted during treatment using current treatment measurements. For example, as described in PCT Publication No. WO 2019 / 234495 (the entire disclosure of which is hereby incorporated herein by reference), the ultrasound signal reflected from the acoustic reflector during the autofocus procedure can first be analyzed to determine its quality (e.g., signal-to-noise ratio, SNR). Based on the determined quality, the reflected signal is classified as: (i) a sufficient quality signal that can be further analyzed to obtain correction for ultrasound parameter values ​​used for autofocus, or (ii) a insufficient quality signal (or an incorrect or missing signal) that will not be used for autofocus. Transducer elements receiving a sufficient quality signal are classified as measuring elements, while transducer elements receiving an insufficient quality signal are classified as non-measuring elements. The predictor can be adjusted using corrections to ultrasound parameter values ​​obtained using the measuring elements. Based on characteristics associated with tissue between the target region and non-measuring elements (e.g., thickness, density, and bundle-skull angle), the adjusted predictor generates corrections to parameter values ​​associated with the non-measuring elements to create optimal focus at the target. The determined correction values ​​associated with the measuring and non-measuring elements can be stored in a database 160 along with their respective transducer elements. Additionally, the database can serve as a basis for generating synthetic data—that is, generating data to train predictors for use in automated focusing or other diagnostic or therapeutic purposes.

[0054] In some cases, the target region may span a large volume that cannot be fully treated in a single treatment phase (e.g., a focused area) and / or multiple target regions with discrete locations may exist. Therefore, it may be necessary to treat a second portion of the target region that differs from the first portion and / or a second target region that differs from the first target region. In one embodiment, the controller optionally adjusts the predictor using corrections to ultrasound parameter values ​​acquired during the automated focusing procedure and / or during treatment of the first target region (or the first portion of the target region). Based on characteristics (e.g., thickness, density, and beam-skull angle) associated with the intermediate tissue located between the second target region (or the second portion of the target region) and the transducer element, the adjusted predictor can generate corrections to parameter values ​​associated with the transducer element to treat the second target region (or the second portion of the target region).

[0055] Figure 3 The diagram illustrates the prediction module 300 (implemented in controller 108, a separate external controller, or other computing entity) and how it is trained and used. The prediction module can be based on any learning algorithm suitable for the prediction task, such as linear or nonlinear regression or Bayesian prediction, or it can be a neural network. Neural networks are organized in a highly interconnected, brain-like manner and can analyze and recognize patterns in a variety of complex inputs. The learning algorithm can involve supervised or unsupervised learning.

[0056] Figure 3 The supervised learning scenario is illustrated, where the learning algorithm receives labeled training data 310. Each item in the training data contains a vector of feature values ​​and an associated label, i.e., the correct (or acceptable) parameter tuning values. A predictor 300 assimilates the training data 310 and generates a model 315 based on its architecture—a set of coefficients or "weights" in the case of simple linear regression, or a more complex interconnected weight architecture in the case of a neural network. Training may involve passing the training data 310 through the predictor 300 multiple times, and once fully trained, the model 315 is tested with labeled test data 320. Like the training data 310, each item in the test data 320 contains a value vector and a label. The predictions made by the trained model 315 based on the test data items are compared to the labels associated with those test items. If the overall prediction accuracy is sufficiently high, then the model 315 is considered fully trained; otherwise, additional training data can be provided to the predictor and / or its architecture can be changed, and the training and testing process is repeated until satisfactory prediction results are obtained.

[0057] In this paper, data associated with two or more transducer elements are considered in a single data sample (i.e., feature vector). For example, consider a transducer with 1024 elements, where one sample is for {element #3, element #46}, and the input data for each element corresponds only to three skull characteristics that are very important to the bundle of the offset element—namely, thickness, skull density, and bundle-skull angle—the values ​​of which are associated with the portion of the skull that intersects with the bundle of the element. Thus, the feature referring to the data sample for {element #3, element #46} is (thickness...). el#3 ,density el#3 ,angle el#3 ,thickness el#46 ,density el#46 ,angle el#46 ), making density el#3 and density el#46 This refers to the density at different locations within the same skull. Therefore, the input size is 3 features × 2 elements = 6. If the input is in the form of 200×32 images, then the input size will be 200×32×2, with each 200×32 image corresponding to a different element. The label is the difference between the measured parameter values ​​(e.g., phase) of the element, and based on inference, the output is a relative prediction; if the relative values ​​are represented by differences, then the prediction is... in This represents the phase shift associated with transducer element i.

[0058] Now assume that the total number of samples describing the transducer is 10,000, where element combinations are selected according to one or more criteria, for example, to ensure sufficient redundancy (each element appears in more than one pair) to maximize prediction confidence. Through inference, we obtain 10,000 equations.

[0059] Regarding the time delay, we obtain:

[0060] Where P {i,j} The model predicts the difference in time delay between transducer elements i and j, i.e., a known (geometrically determined) quantity. The goal is to find the set that best satisfies the above difference equation based on an error metric (e.g., least squares minimization). The errors associated with each equation can be weighted (based on skull translucency, model confidence, etc.).

[0061] Taking advantage of the fact that constant phase deviation does not affect focus quality, the equations can be expressed as a set of overdetermined linear equations:

[0062]

[0063] The first row corresponds to setting the reference phase, and each subsequent row corresponds to the equation predicting P for one of the element pairs.{i,j} Thus, the matrix is ​​10,001 × 1024, and can be solved, for example, using ordinary least squares.

[0064] The case of wrapped phase shift (instead of amplitude, time delay, or full phase shift) introduces additional difficulties due to phase wrapping. The resulting set of 10,000 equations would be:

[0065]

[0066] Among them, phase shift Restricted to [-π,π], P {i,j} This is the model's prediction of the difference in the wrapping phase shifts of elements i and j (usually rewrapped and also restricted to [-π,π]). Similarly, P {i,j} Given the quantities, the goal is to find the set that best satisfies the above equation. The phase wrap equation can be written as:

[0067]

[0068] The first row corresponds to setting the reference phase, and the quantity k {i,j} Unknown. For simplicity, we will express the above equation as follows: This is referred to as Equation 1. One method to solve this set of equations is as follows. First, set the phase shift difference to the range [0, 2π] instead of [-π, π]:

[0069]

[0070] This change will not affect the solution. Equation 1 can then be rewritten as:

[0071]

[0072] The range [0, 2π] can be discretized by defining the variable p as a large prime number and rewriting equation 1 as follows:

[0073]

[0074] Assume p is sufficiently high in accuracy relative to the model. Using Given that the 2π cycles in the matrix are irrelevant to achieving optimal focus, and that all entries in matrix A are either 1 or -1, equation 1 can be further rewritten as:

[0075]

[0076] A set of overdetermined linear equations over a finite field is generated. This set of linear equations can be solved on integers, for example, by Gaussian elimination of multiple combinations of N equations, where N is the number of elements, and by selecting the best solution through the correspondence between the best solutions.

[0077] Another embodiment creates a package phase shift list based on the difference by solving a set of overdetermined quadratic equations:

[0078]

[0079] The objective function to be minimized is as follows:

[0080] f(x 1…n ,y 1…n )=∑ eq. x i x j +y i y j -cos(P ij )+y i x j -x i y j -sin(P ij )+x i 2 +y i 2 -1+x j 2 +y j 2 -1,

[0081] The sum can be weighted by factors such as transmittance and confidence level, and its gradient can be calculated using gradient descent.

[0082]

[0083] Regarding amplitude, it was found that the ratio of the amplitude of the received reflected signal to that of the transducer element (rather than the difference in phase measurement) provides the best result for the ultimate goal of compensating for beam aberration.

[0084] Therefore, for the amplitude, we obtain:

[0085] Similarly, these equations can be represented as a set of overdetermined linear equations:

[0086]

[0087] The first line corresponds to setting the reference amplitude, and each subsequent line corresponds to the equation predicting P for one of the element pairs. {i,j}The use of ratios eliminates errors caused by bias, also because the constant bias applied to all components of the transducer has no effect on focus quality.

[0088] Predicting phase is very similar to predicting time delay. Typically, the term "time delay" refers to the overall time (or phase) difference caused by the skull medium relative to the water medium, for example, 10 radians in terms of phase. The term "phase shift" typically refers to a bounding phase confined to the interval [-π, π]. Achieving accurate time delay measurements can be more difficult than bounding phase shifts.

[0089] In summary, the method described above replaces the k-length input vector used for predictions based on individual elements with a 2k-length vector for element pairs (or an n×k-length vector for larger groups of elements). However, other methods of representing the data are also possible. In one alternative, the two groups of images from the skull media can be combined, for example, in a fourth dimension. For example, refer to... Figure 4 The skull and brain media through which the element's bundle passes can be described by 3D volumes aligned with the bundle, i.e., volume v1 corresponds to element e1, and volume v2 corresponds to element e2. The first dimension extends along the path from the element to the target and is extracted by interpolation of the skull CT image representatively indicated at 400; that is, the figure shows the CT image at the depth corresponding to the first layer of volumes v1 and v2. Skull blocks 401 and 402 correspond to v1 and v2, respectively, and the CT image layer 400 shown appears on the first layers 4011 and 4021 of skull blocks 401 and 402. Subsequent layers of skull blocks 401 and 402 will contain image information from the CT layers corresponding to the depth of the block layers along the third dimension. If each block layer is 200×32 pixels and there are 32 layers, then the size of each block volume is 200×32×32. Therefore, when considering a single data sample corresponding to a pair of elements, the skull media (imaged by CT) of both elements are included in the sample input. Connecting these blocks along the fourth dimension produces a 200×32×32×2 input array.

[0090] exist Figure 5 In another method shown, a Siamese neural network 500 is employed, in which each of the neural networks 501, 502 receives input to a single element. That is, the skull block / feature vector of each element is the input to one of two identical neural networks 501, 502, which have the same architecture and share weights. The outputs of the sister networks 501, 502 are then combined to form a loss that depends only on the relative target value of the element. By having a sister network for each sample element and by extending the Siamese neural network by including an additional neural network that receives the combined output of multiple sister networks before loss evaluation, this format can be easily extended to accommodate more than two elements in a sample.

[0091] In another embodiment using more than two elements, the feature is a combination of their skull properties. For example, assuming all samples are of equal size, each data sample combines the values ​​of three elements, with input parameters being thickness, density, and angle. Then, when one of the samples consists of elements 3, 46, and 15, the feature might be (thickness...) el#3 ,density el#3 ,angle el#3 ,thickness el#46 ,density el#46 ,angle el#46 ,thickness el#15 ,density el#15 ,angle el#15 )

[0092] Alternatively, the connection volume can be 200×32×32×3. The output can be a vector of relative values ​​(e.g., second and third values ​​relative to the first value), and the loss can be the predicted stress class score relative to the true value, weighted by factors such as the predicted amplitude or difficulty.

[0093]

[0094] Now assume the sample sizes are different. For example, transducer elements can be grouped or the transducer can be divided into segments containing multiple elements. To illustrate, the transducer could be divided into 16 regions or segments, each containing 64 adjacent elements. Each of these segments may have a different number of elements suitable for use in a particular procedure and thus efficiently considered in the dataset. A sample generation network can be used to group the transducer into groups of eligible elements. For example, these segments and their varying numbers of eligible elements can be modeled using recursive or graph neural networks. Alternatively, the entire transducer can be represented as a graph, with the goal of optimizing sample generation and label learning. For example, a sample generation neural network can be initiated by having it perform random traversals (possibly with constraints) with varying numbers and step sizes. The output of this network is used to select elements in each data sample that serve as input to a graph neural network, which learns labels (as relative values) and returns predictions and prediction confidence levels. These values ​​can also be used to estimate the loss of the sample generation model, as well as constraints on whether providing true labels to the generated samples will achieve perfect focusing (e.g., by using all elements and enforced redundancy).

[0095] Another approach is to use unsupervised learning to detect skull behavior. For example, in treatment where an autofocusing procedure is performed on a portion of a transducer, the learned behavior can be used to find the acoustic parameters of elements with missing measurements, for instance, by using cluster analysis to fill all elements with missing measurements in a cluster with the median phase of the measured elements in the cluster.

[0096] Refer again Figure 1 In various embodiments, ultrasound parameter corrections (including amplitude, time delay, and / or phase shift) determined using the aforementioned predictor and / or appropriate modes for element activation and deactivation for specific procedures and patient anatomy are stored together with their respective transducer elements in a database 160 in a memory 162, accessible by an adjustment mechanism 126 and / or a controller 108. In one embodiment, the database stores transducer elements and their corresponding parameter corrections generated by the skull in a table, the entries of which are populated using the aforementioned predictor. The memory may comprise or consist primarily of one or more volatile or non-volatile storage devices, such as random access memory (RAM) devices, read-only memory (ROM) devices, magnetic disks, optical disks, flash memory devices, and / or other solid-state memory devices, such as DRAM, SRAM, etc. All or part of the memory may be located remotely from the ultrasound system 100 and / or the imager 112, for example, as one or more storage devices connected to the ultrasound system 100 and / or the imager 112 via a network (e.g., Ethernet, WiFi, cellular telephone network, Internet, or any combination of local area network or wide area network or network capable of supporting data transmission and communication). As used herein, the term “storage” broadly refers to any form of digital storage, such as optical storage, magnetic storage, semiconductor storage, etc.

[0097] In other embodiments, the predictor is part of the adjustment mechanism 126 (i.e., stored in a memory 162 accessible by the adjustment mechanism 126). For example, prior to treatment, the phase and / or amplitude of the reflected signal can be sensed by a transducer element to be used in the treatment procedure, and the input vector can be created by: (i) the difference or ratio between the sensed signals as described above, (ii) geometric parameters specific to each transducer element (e.g., angles), and (iii) anatomical features (thickness, density) obtained through CT, MR, or other imaging modalities. Each input vector is processed by a trained neural network that outputs a correction value for the associated transducer element.

[0098] In one embodiment, controller 108 implements a physical model to predict treatment outcomes (e.g., real-time temperature) at target region 101 and / or non-target regions using tissue properties (e.g., energy absorption coefficient) and ultrasound parameter values ​​(e.g., phase, amplitude, etc.) stored in database 160. Additionally, the physical model can predict the real-time temperature and / or shape of the focal zone (which may or may not coincide with the target region). Based on the predicted treatment outcomes of the target / non-target regions, the real-time temperature of the focal zone, and / or the shape of the focal zone, controller 108 can use, for example, the physical model to determine the patient's suitability for ultrasound treatment (e.g., treatment success rate). In various embodiments, imager 112 is used to acquire tissue properties of the target region and the through region. For example, based on the acquired images, a tissue model characterizing the material properties of the target region and the through region can be constructed. The tissue model can take the form of a 3D table of cells corresponding to voxels representing the target tissue; the cells have values ​​representing attributes of tissue properties related to energy absorption, such as the absorption coefficient. Voxels are acquired by an imaging device via computed tomography, and the tissue type represented by each voxel can be automatically determined by conventional tissue analysis software. Using the determined tissue type and tissue parameter lookup table (e.g., absorption coefficient by tissue type), cells of the tissue model can be populated. During the autofocusing procedure, measurements can be used to optimize tissue properties in the physical model (depending on the specific patient and the imager 112 used), which can then be used to target new locations or for missing measurements. For more detailed information on creating tissue models that identify the energy absorption coefficient, thermosensitivity, and / or thermal tolerance of various tissues, please refer to U.S. Patent Publication No. 2012 / 0029396, the entire disclosure of which is hereby incorporated by reference.

[0099] Figure 6AAn exemplary method 600 is illustrated for training a predictor to generate corrected ultrasound parameter values ​​associated with transducer elements. According to the invention, these parameter values ​​help compensate for anticipated beam aberrations caused by tissue between the target region and the transducer element. In a first step 602, an imager 112 is activated to acquire images of the target tissue and / or intermediate tissue. Based on this, a controller 108 identifies numerical quantities characterizing the intermediate and / or target tissue (e.g., anatomical features such as thickness and / or density, and / or tissue properties such as energy absorption coefficient) and / or geometric parameters associated with each transducer element (e.g., beam-skull angle) (in a second step 604). In one embodiment, coordinate systems need to be registered in different imaging modalities (e.g., ultrasound imaging, MRI imaging, and / or CT imaging) to calculate the beam-skull angle associated with each element. For example, an exemplary registration method is provided in U.S. Patent No. 9,934,570, the entire disclosure of which is hereby incorporated herein by reference. Additionally, the controller may acquire / determine one or more correction values ​​for ultrasound parameters (e.g., frequency, amplitude, and / or phase) associated with the transducer elements to compensate for aberrations caused by intermediate tissues with identified anatomical features, in order to create optimal focus at the target region (in step 606). The correction values ​​may be acquired / determined based on, for example, a retrospective study of previously performed ultrasound procedures. Furthermore, the controller may calculate the differences and / or ratios of correction values ​​between multiple (e.g., two) transducer elements (in step 608). The differences and / or ratios of the anatomical features identified in step 604 and the ultrasound parameter values ​​determined in step 608 are then provided as training data for a predictor that estimates ultrasound correction values ​​based on the anatomical features (in step 610). Training may be based on any suitable learning algorithm and may involve supervised or unsupervised learning.

[0100] Once the predictor is trained, the controller 108 can implement the trained predictor to determine the optimal setting of the transducer elements, thereby producing the best therapeutic effect at the patient's target area during treatment or diagnostic procedures. (Reference) Figure 6BIn the first step 652, controller 108 may implement a predictor that has been trained to generate corrections to ultrasound parameter values. In the second step 654, controller 108 may provide the predictor with multiple input vectors, each containing a numerical quantity characterizing the tissue between the target area of ​​the patient and the transducer element to be activated during the treatment phase. The numerical quantity may include anatomical features (e.g., thickness and / or density) and / or tissue properties associated with the target and / or intermediate tissue and / or geometric parameters (e.g., bundle-skull angle) associated with each transducer element. Similarly, the numerical quantity may be obtained from images acquired using imager 112. Based on the input vectors, the predictor may generate correction values ​​as output associated with the active transducer element to create optimal focus at the target area (in step 656). Controller 108 may then activate the transducer element based on the corresponding correction value to initiate an autofocusing procedure (in step 658). Subsequently, controller 108 may treat the target area by operating the transducer element using the ultrasound parameter values ​​determined in the autofocusing procedure (in step 660). The transducer elements in operation may comprise all elements of the transducer array. Alternatively, the transducer elements in operation may comprise only some elements of the transducer array. For example, based on reflected signals received during the autofocus procedure, controller 108 may determine whether there are missing and / or incorrect measurements associated with one or more transducer elements during the autofocus procedure (in step 662). If so, then the transducer elements in operation comprise only those elements without missing and / or incorrect measurements. In some embodiments, controller 108 optionally adjusts the predictor using corrections to ultrasound parameter values ​​acquired during the autofocus procedure (in step 664). Additionally, the controller may provide the adjusted predictor with multiple input vectors, each containing a numerical quantity characterizing the tissue between the target region and the transducer elements with missing / incorrect measurements (in step 666). The adjusted predictor may then generate corrections to the parameter values ​​associated with the transducer elements with missing / incorrect measurements; and the controller may use the newly generated correction values ​​to activate these transducer elements. In some embodiments, controller 108 further determines whether a second portion of the target region needs treatment (e.g., when the target region spans a large volume that cannot be fully treated in step 660) and / or a second target region (e.g., when the first and second target regions have discrete locations) (in optional step 668). If so, controller 108 may optionally adjust the predictor (in step 670) using corrections to ultrasound parameter values ​​acquired during the autofocusing procedure (performed in step 658) and / or during treatment of the target region (or a portion of the target region) (performed in step 660).Additionally, the controller can provide the adjusted predictor with multiple input vectors, each containing a numerical value representing the tissue between the new target region (or a new portion of the target region) and the transducer element (in step 672). Subsequently, the adjusted predictor can generate corrections to the parameter values ​​associated with the transducer element for treating the new target region (or a new portion of the target region); and the controller can activate the transducer element based on the corresponding correction values.

[0101] Generally, the functions of implementing the predictor as described above, measuring quantities constituting the input vector or facilitating database lookups, training the predictor, providing input to the trained predictor, receiving output from the predictor, executing autofocusing procedures, and adjusting ultrasound parameter values ​​during ultrasound procedures, whether integrated within the controller of the imager and / or ultrasound system or provided by a separate external controller, can be constructed in one or more modules implemented in hardware, software, or a combination of both. For embodiments where the functionality is provided as one or more software programs, the programs can be written in any of a variety of high-level languages, such as Python, Java, C, C++, C#, BASIC, various scripting languages, and / or HTML. For neural networks, machine learning libraries or frameworks such as Tensorflow, Pytorch, Theano, or Caffe can be used directly. Alternatively, the software can be implemented in assembly language for a microprocessor residing on a target computer (e.g., the controller); for example, if the software is configured to run on an IBM PC or a PC clone, then the software can be implemented in Intel 80x86 assembly language. Software can be embodied in an article of art, including but not limited to floppy disks, jump drives, hard disks, optical disks, magnetic tapes, PROMs, EPROMs, EEPROMs, field-programmable gate arrays, or CD-ROMs. Embodiments using hardware circuitry systems can be implemented using, for example, one or more FPGAs, CPLDs, or ASIC processors.

[0102] Furthermore, the term "controller" as used herein broadly encompasses all necessary hardware components and / or software modules for performing any of the functions described above; a controller may contain multiple hardware components and / or software modules, and functionality may be extended across different components and / or modules. Additionally, the terms "prediction module" and "predictor" are used interchangeably herein.

[0103] Some embodiments of the present invention have been described above. However, it is clearly stated that the present invention is not limited to these embodiments; rather, additions and modifications to the content expressly described herein are also included within the scope of the present invention.

Claims

1. A system for delivering ultrasound energy to a target area during a treatment or diagnostic procedure, the system comprising: An ultrasonic transducer comprising a plurality of transducer elements, at least some of which are designated to be in operation during the treatment or diagnostic procedure; The adjustment mechanism includes a machine learning model trained based on an input vector corresponding to the difference and / or ratio of parameter values ​​between a plurality of said transducer elements, the adjustment mechanism being configured to: (i) receive a numerical quantity characterizing the tissue between the operating transducer elements and the target region; and (ii) based thereon, generate at least one parameter value for each of said operating transducer elements to compensate for expected beam aberrations. as well as A controller configured to activate the transducer element in action according to corresponding parameter values ​​in order to generate an optimal focusing area at the target region during the treatment or diagnostic procedure.

2. The system according to claim 1, wherein, The machine learning model is a neural network.

3. The system according to claim 2, wherein, The neural network is trained based on input vectors, each of which includes parameters associated with a pair of transducer elements.

4. The system according to claim 1, wherein, The adjustment mechanism includes a lookup table of adjustment values ​​generated by a machine learning algorithm.

5. The system according to claim 4, wherein, The machine learning algorithm is a neural network trained on input vectors, each of which includes parameters associated with more than two transducer elements.

6. The system according to claim 1, wherein, The numerical values ​​include tissue density and thickness.

7. The system according to claim 1, wherein, The adjustment mechanism is configured to generate the at least one parameter value based at least in part on at least one geometric parameter associated with the associated transducer element.

8. The system according to claim 7, wherein, The at least one geometric parameter is an angle relative to the skull surface between the associated transducer element and the target region.

9. The system according to claim 1, wherein, The resulting parameter value specifies a correction for at least one of the following: amplitude, phase, frequency, duty cycle, ultrasonic processing mode, or time delay.

10. The system according to claim 1, wherein, The target region comprises multiple parts, and the resulting optimal focus area is located in the first part of the target region. The controller is configured to predict at least one of the following using a physical model: (i) the real-time temperature of a second part of the target region that is different from the first part or a non-target region; (ii) the real-time temperature of the focus area; (iii) the focus shape; or (iv) treatment success.

11. The system according to claim 1, wherein, The controller is configured to: The transducer element in operation shall: (i) emit ultrasonic waves toward the target area, and (ii) measure the reflection of the ultrasonic waves from the target area; Based on measurement, each of the transducer elements in operation is classified as a measuring transducer element or a non-measuring transducer element; and The adjustment mechanism is updated based at least in part on measurements provided by the measurement transducer element.

12. The system according to claim 11, wherein, The updated adjustment mechanism is configured to: (i) receive a numerical quantity characterizing the tissue between the non-measuring transducer element and the target region; and (ii) based thereon, generate at least one updated parameter value for each of the non-measuring transducer elements to compensate for the expected beam aberration.

13. The system according to claim 12, wherein, The controller is configured to activate the non-measurement transducer element based on the corresponding updated parameter value.

14. The system according to claim 1, wherein, The controller is configured to: The transducer element in operation shall: (i) emit ultrasonic waves toward the target region, and (ii) measure the reflection of the ultrasonic waves from the target region; and The adjustment mechanism is updated based at least in part on measurements.

15. The system according to claim 14, wherein, The updated adjustment mechanism is configured to: (i) receive a numerical value characterizing the organization between the transducer element in action and a secondary target region different from the target region; and (ii) based thereon, generate at least one updated parameter value for each of the transducer elements in action to compensate for the expected beam aberration.

16. The system according to claim 15, wherein, The controller is configured to activate the transducer element in action according to the corresponding updated parameter values ​​in order to generate an optimal focusing area at the secondary target region.

17. The system according to claim 1, wherein, The input vector is based at least in part on at least one of the differences or ratios of parameter values ​​between the transducer elements in the operation.

18. A method for generating correction values ​​for delivering ultrasonic energy through intermediate tissue via an ultrasonic transducer comprising a plurality of transducer elements, the correction values ​​correcting for beam aberration, the method comprising the steps of: A machine learning model is trained based on received input vectors, each input vector including (i) the difference and / or ratio of ultrasound parameters between multiple transducer elements and (ii) a numerical quantity characterizing the tissue between the multiple transducer elements and the target region, and used to generate corrected values ​​of the ultrasound parameters as output to compensate for expected beam aberrations passing through the intermediate tissue, the training including enabling the machine learning model to learn the relationship between the values ​​of the input vectors and the corresponding corrected values. The machine learning model is provided with multiple input vectors, each input vector including a numerical quantity representing patient tissue between the multiple transducer elements to be activated and the target region of the patient; as well as Based on the provided input vector, the corrected values ​​of the ultrasound parameters are received as the output of the machine learning model.

19. The method according to claim 18, wherein, The machine learning model is a neural network.

20. The method according to claim 19, wherein, The neural network is trained based on input vectors, each of which includes parameters associated with a pair of transducer elements.

21. The method according to claim 18, wherein, The differences and / or ratios of the ultrasound parameters are obtained using measurements taken from an ex vivo or replica skull.

22. The method according to claim 18, wherein, The numerical values ​​include tissue density and thickness.

23. The method according to claim 18, wherein, The input vector includes at least one geometric parameter associated with each of the transducer elements, and the machine learning model is trained to generate the corrected values ​​of the ultrasound parameters based at least in part on the geometric parameters.

24. The method according to claim 23, wherein, The geometric parameters are angles relative to the skull surface between each of the transducer elements and the target region.

25. The method according to claim 18, wherein, The correction value specifies a correction for at least one of amplitude, phase, frequency, duty cycle, ultrasonic processing mode, or time delay.

26. The method of claim 18, further comprising activating the transducer element according to a corresponding correction value of the ultrasound parameters to generate an optimal focusing area at a target region of an ex vivo or replica skull.

27. The method according to claim 18, wherein, The target region comprises multiple parts, and the resulting optimal focal area is located in the first part of the target region. The method includes using a physical model to predict at least one of the following: (i) the real-time temperature of a second part of the target region that is different from the first part or a non-target region; (ii) the real-time temperature of the focal area; (iii) the focal shape; or (iv) treatment success.

28. The method of claim 18, comprising: Activate at least some of the transducer elements to: (i) emit ultrasonic waves toward a target region of an ex vivo or replica skull, and (ii) measure the reflection of the ultrasonic waves from the target region; Based on measurement, each of the transducer elements is classified as a measurement transducer element or a non-measurement transducer element; and The machine learning model is updated based at least in part on measurements provided by the measurement transducer elements.

29. The method of claim 28, comprising: The updated machine learning model is provided with numerical quantities that characterize the tissue between the non-measurement transducer element and the target region. as well as For each of the non-measurement transducer elements, at least one updated parameter value is received to compensate for the expected beam aberration.

30. The method of claim 29, further comprising activating the non-measurement transducer element according to the corresponding updated parameter value.

31. The method of claim 18, comprising: Activating at least some of the transducer elements to: (i) emit ultrasonic waves toward a target region of an ex vivo or replica skull, and (ii) measure the reflection of the ultrasonic waves from the target region; and The machine learning model is updated based at least in part on measurements.

32. The method of claim 31, comprising: The updated machine learning model is provided with numerical quantities that characterize the organization between the transducer element and a secondary target region different from the target region. as well as For each of the transducer elements, at least one updated parameter value is received to compensate for the expected beam aberration.

33. The method of claim 32, further comprising activating the transducer element according to a corresponding updated parameter value to generate an optimal focusing area at a secondary target region of an ex vivo or replica skull.

Images