An acoustic aperture drug delivery spatial prediction method and system based on perfusion hard-gating and microbubble population threshold

Abstract

The application discloses a kind of space prediction method and system of ultrasonic mediated drug delivery, belong to medical ultrasonic image processing technical field.The pixel-by-pixel perfusion map is used as binary hard gate, and the non-perfusion area (such as tumor necrosis core) without microbubble arrival is zeroed regardless of its mechanical index, which corrects the specific failure mode of "high mechanical index is misjudged as high delivery" in the prior art, significantly improves the physical accuracy of spatial delivery prediction;The steady-state / inertial cavitation threshold derived from the scattering cross-section weighting of microbubble particle size distribution varies with microbubble model and frequency.The same set of thresholds is used to generate multi-channel delivery partition and as a parameter recommendation acoustic pore window constraint, making the prediction and planning physically self-consistent;Provide subject-specific cavitation window and quantifiable bleeding risk partition, significantly improve the physical accuracy and safety of spatial delivery prediction.

Description

Technical Field

[0001] This invention relates to the field of medical ultrasound image processing and computer-aided delivery dosimetry, and more specifically, to a spatial prediction method and system for ultrasound-mediated (acoustic aperture effect / ultrasound microbubble targeted delivery) drug delivery. Background Technology

[0002] The acoustic aperture effect of ultrasound combined with microbubbles can temporarily increase the permeability of vascular endothelium and cell membranes, thereby enhancing drug delivery to the target area. This is an important means of locally enhancing drug delivery for diseases such as tumors. To improve its efficacy and safety, it is necessary to perform forward prediction of the spatial delivery distribution of drugs in the target area in a computer and provide safe zones to assist doctors in treatment planning.

[0003] Currently, quantitative prediction of ultrasound-mediated drug delivery has the following main shortcomings:

[0004] First, existing cavitation dose imaging methods (such as passive cavitation imaging) are mainly used for post-event monitoring and imaging of "delivered" doses, which is a post-event evaluation rather than forward prediction; they only rely on sound pressure or mechanical index (MI) and cavitation dose for judgment, without introducing perfusion information of the target area at all.

[0005] Secondly, while existing contrast-enhanced ultrasound (CEUS) perfusion measurement software can logarithmically decompress and linearize contrast-enhanced videos and calculate perfusion parameter maps such as pixel-by-pixel peak enhancement, it is purely for diagnostic / monitoring purposes and never combines perfusion maps with in vivo mechanical indices or cavitation doses to predict drug delivery.

[0006] Third, existing methods that associate perfusion with drug uptake only use perfusion as a soft correlation or soft weight to predict uptake, without using perfusion as a binary hard gate. Therefore, for non-perfused areas without microbubble access (such as tumor necrosis nuclei), existing methods will still give non-zero delivery predictions when their mechanical index is high, thus overestimating drug delivery in the clinical context and producing false positive predictions.

[0007] Fourth, existing cavitation threshold modeling is mostly biophysical simulation detached from specific planning systems, and does not apply the steady-state / inertial threshold derived from microbubble particle size distribution to subject-specific delivery prediction and safety zoning; existing parameter selection is mostly table lookup or single-signal driven, lacking the integration of "prediction-safety zoning".

[0008] In summary, existing technologies either only consider pressure / MI, only consider perfusion, or only perform post-operative monitoring, leading to technical problems such as overestimating delivery in non-perfusion areas, lack of microbubbles and subject-specific cavitation windows, and lack of quantifiable safety zones. Summary of the Invention

[0009] To address the technical problems of existing technologies that overemphasize sound pressure / mechanical index or rely solely on perfusion for soft correlation, overestimate delivery in non-perfusion areas, and lack subject-specific cavitation windows and safety zones, this invention provides a spatial prediction method and system for acoustic aperture drug delivery based on perfusion hard gating and microbubble population thresholds.

[0010] Technical solution

[0011] The technical solution of the present invention is as follows:

[0012] On one hand, this invention provides a spatial prediction method for ultrasound-mediated drug delivery, comprising: acquiring contrast-enhanced ultrasound or ultrasound microvascular imaging images and performing logarithmic decompression and linearization on their pixel data to obtain a pixel-by-pixel perfusion map; calculating the in vivo mechanical index distribution and cavitation dose distribution containing steady-state and inertial components of the target area based on acoustic parameters, tissue attenuation models, and capsule microbubble population dynamics models, and deriving subject-specific steady-state cavitation thresholds and inertial cavitation thresholds from the population summation of microbubble particle size distributions weighted by scattering cross sections; combining the pixel-by-pixel perfusion map as a hard gate term with the in vivo mechanical index distribution and cavitation dose distribution, setting the predicted delivery value to zero and marking it as a delivery cavity for pixels with perfusion values ​​lower than the perfusion threshold regardless of their mechanical index, and dividing the remaining pixels into multiple channels such as steady-state acoustic window effective delivery and inertial cavitation extravasation according to the mechanical index, generating a spatial delivery prediction map and a safety partition and outputting it; furthermore, under the constraint of the acoustic window, recommended acoustic parameters can be calculated and output, and delivery regression and safety classification models can be trained with paired data and updated by back-interpolation.

[0013] On the other hand, the present invention also provides a spatial prediction system for ultrasound-mediated drug delivery, comprising: a data input and linearization module for acquiring images and generating pixel-by-pixel perfusion maps; a physical modeling module for calculating in vivo mechanical index distribution and cavitation dose distribution; a threshold determination module for determining subject-specific steady-state and inertial cavitation thresholds using a microbubble population dynamics model; a perfusion gating and partitioning module for generating multi-channel spatial delivery prediction maps and safety partitions using perfusion maps as hard gates; and an interactive output module for presenting prediction maps, quantitative indicators, and safety quality control information.

[0014] It should be noted that this invention is an auxiliary prediction and planning tool for ultrasound-mediated drug delivery. The spatial delivery prediction map, quantitative indicators and safety quality control information output by this invention are only data for doctors to refer to in treatment planning. This invention does not include driving the ultrasound transducer or applying ultrasound to a living body, drug delivery or other treatment steps. The final clinical decision is still made by the doctor.

[0015] The beneficial effects of this invention are as follows:

[0016] Eliminating false-positive delivery predictions in non-perfusion areas: By using pixel-by-pixel perfusion maps as binary hard gating, the predicted delivery for non-perfusion areas without microbubble arrival (such as tumor necrosis nuclei) is set to zero regardless of their mechanical index. This corrects the specific failure mode of existing technology, where "high mechanical index is misjudged as high delivery", and significantly improves the physical accuracy of spatial delivery prediction. This effect is neither taught nor inspired by existing technology that only uses soft correlation for perfusion.

[0017] It provides subject-specific cavitation windows and supports integrated safety zoning: the steady-state / inertial cavitation thresholds derived from the microbubble size distribution through scattering cross-section weighting vary with microbubble type and frequency. The same set of thresholds is used to generate multi-channel delivery zoning and can also serve as constraints for recommended acoustic aperture windows, making prediction and planning physically consistent. The system further outputs multi-channel safety zoning for steady-state acoustic aperture effective delivery, inertial cavitation extravasation (with bleeding risk), and delivery cavities, providing doctors with quantifiable safety references. Attached Figure Description

[0018] Figure 1 The overall flowchart of the ultrasound-mediated drug delivery spatial prediction method provided in the embodiments of the present invention is shown.

[0019] Figure 2 This is a module architecture diagram of a spatial prediction system provided in an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of perfusion hard-gated and multi-channel delivery zoning provided in an embodiment of the present invention, showing the steady-state acoustic hole effective delivery zone, the inertial cavitation extravasation risk zone and the non-perfusion delivery cavity zone, as well as the tumor outline and necrotic nucleus.

[0021] Figure 4 This is a schematic diagram illustrating the determination of microbubble population particle size distribution and steady-state / inertial mechanical index threshold provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to specific embodiments.

[0023] Example 1: Physical modeling, threshold determination, and model building process (offline stage)

[0024] This embodiment mainly illustrates the physical modeling of in vivo mechanical index and cavitation dose, the determination of subject-specific cavitation threshold, and the construction mechanism of data flywheel model.

[0025] Input: Contrast-enhanced ultrasound / ultrasound microvascular imaging image of the target area, acoustic parameters (frequency f, peak negative pressure PNP, pulse repetition frequency PRF, depth), microbubble type and its particle size distribution and capsule parameters.

[0026] Processing flow:

[0027] (1) Pixel-by-pixel perfusion map generation module (Linearization): Clinical CEUS display data is logarithmically compressed, and the pixel values ​​are not linearly related to microbubble concentration, so linearization is required first. Assume the 8-bit display pixel value is... The scanner's dynamic range is (dB, default value is 50), then the linearized echo power is:

[0028]

[0029] For pixel values saturated pixels and Truncated pixels are masked (these pixels are irrecoverable). For video, each pixel is taken in time series. The maximum value is used as peak enhancement ,by As the pixel injection value For URM color images, microvascular color saturation is used as... .

[0030] (2) In vivo mechanical index and cavitation dose modeling module (Physics): Mechanical index is calculated according to... Calculate (PNP in MPa, f in MHz) and calculate along the propagation path according to the tissue attenuation coefficient. (e.g., 0.5 dB / cm / MHz) derating correction for sound pressure level, depth The depreciation factor at the location is:

[0031]

[0032] The radius of microbubbles was determined using the Marmottant model. Solve the radial vibration equation (e.g., using the fourth-order Runge-Kutta integral) to obtain the maximum expansion ratio. For the cavitation emission spectrum, the root mean square of the subharmonic and superharmonic energies characterizes the steady-state cavitation dose component. The root mean square of broadband noise energy is used to characterize the inertial cavitation dose component. .

[0033] (3) Population Threshold Determination Module: Samples the representative radius of microbubble particle size according to a log-normal distribution. Each radius is determined according to its scattering cross section. Scattering weighting, with weights as follows:

[0034]

[0035] in Let be a log-normal probability density function. For each Using the Marmottant model under a series of MIs to find Let the steady-state oscillation criterion be satisfied (e.g.) The weighted proportion of the group The MI corresponding to the first time it exceeds the first preset proportion (e.g., 0.2) is the steady-state cavitation threshold. Let the inertial collapse criterion be satisfied (e.g.) The MI corresponding to the first time the weighted group proportion of ) exceeds the second preset proportion (e.g., 0.5) is the inertial cavitation threshold. . , The shell elasticity χ and shell viscosity κs vary depending on the microbubble type (such as SonoVue, Definity, and the self-developed CXCR4 microbubble) and frequency, thereby achieving microbubble / subject specificity.

[0036] (4) Data Flywheel Model Training Module: Collects paired data records according to a structured template. Each record contains acoustic parameters, in vivo MI distribution derived from the parameters, cavitation dose distribution, pixel-by-pixel perfusion map, and delivery outcome quantity measured by an objective gold standard (such as fluorescence, HPLC, mass spectrometry imaging, or PET). This is used to train a delivery quantity regression model (input includes MI distribution, cavitation dose distribution, pixel-by-pixel perfusion map, and perfusion...). (The pixel forced to set the predicted delivery amount to zero) and the safe classification model (depending on) , Outputs multi-channel safety partitions and bleeding / damage risk levels. When real data is insufficient, the training pipeline can be run with physically synthesized data first, and then a usable model can be obtained after connecting real paired data.

[0037] Example 2: Clinical Assisted Delivery Prediction Process (Online Operation Phase)

[0038] This embodiment illustrates the system's workflow in a clinical setting.

[0039] Input: CEUS / URM image of the target area to be analyzed, physical resolution (mm / pixel), and acoustic parameters.

[0040] Processing flow:

[0041] [Step S101] Image and perfusion map acquisition: The system acquires the image to be analyzed and performs preprocessing. According to Example 1 (1), the pixel data is linearized and saturated / truncated pixels are masked to generate a pixel-by-pixel perfusion map. .

[0042] [Step S102] Physical modeling: Calculate the in vivo mechanical index distribution of the target area according to Example 1 (2). With cavitation dose distribution ( , The MI displayed by diagnostic ultrasound is the in vivo value after tissue devaluation. In practice, the in vivo focal MI is made consistent with the displayed MI to avoid secondary devaluation.

[0043] [Step S103] Determine the subject-specific threshold: Derive the steady-state cavitation threshold from the particle size distribution of the microbubbles used by the subject according to Example 1 (3). With inertial cavitation threshold .

[0044] [Step S104] Inject hard gating and multi-channel partitioning: for each pixel ,like ( To differentiate between necrotic nuclei and perfusion thresholds without blood supply, the delivery prediction value is... And it is identified as a delivery hole channel, with that pixel's High or low is irrelevant; if Then partition according to the following formula:

[0045]

[0046] Thanks to the hard gating, the microbubble-free arrival area will not be misjudged as a delivery area due to its high mechanical index.

[0047] [Step S105] Generate safety quality control information: The system generates a safety map that is registered with the prediction map, according to each pixel. The degree of exceedance relative to the preset dose upper limit is graded; and the area ratio of each channel within the target area is calculated as a clinically relevant quantitative indicator.

[0048] [Step S106] Interactive output: On the interface, based on the ultrasound B-mode image, the steady-state acoustic hole effective delivery area, inertial cavitation extravasation risk area and non-perfusion delivery cavity area, clinical quantitative indicators and safety quality control information are superimposed with different identifiers to assist doctors in treatment planning; optionally, the partitions are recalculated and updated in real time when parameters change.

[0049] [Step S107] Parameter Recommendation (Optional): Based on , The defined acoustic aperture window serves as a constraint. Candidate acoustic parameters are traversed to calculate and output the recommended parameter window that maximizes the effective delivery area of ​​the steady-state acoustic aperture and ensures that the predicted inertial cavitation dose does not exceed the safety limit. This window serves as data for treatment planning reference.

[0050] Example 3: Comparative Analysis of Real Clinical Cases

[0051] To further demonstrate the accuracy of this invention and its superiority over traditional and existing methods, the following analysis and comparison are conducted using a real case of difficult hepatocellular carcinoma (with a large area of ​​necrotic nucleus in the center and rich blood supply around the tumor).

[0052] Case Background: Contrast-enhanced ultrasound images of a patient with advanced hepatocellular carcinoma, with a physical resolution of approximately 0.08 mm / pixel. CEUS showed no contrast enhancement in the central necrotic nucleus of the tumor (no microbubble perfusion), but significant enhancement in the peritumoral capsule area. A SonoVue microbubble delivery protocol with a frequency of 1.7 MHz and a screen mechanical index (MI) of 0.25 was proposed.

[0053] 1. Traditional / Mechanical Index-Only Method (Subjective Experience-Based Parameter Setting): Physicians rely solely on the mechanical index on the screen and their experience, assuming that "acoustic delivery can be achieved throughout the entire tumor area at MI 0.2–0.3," without distinguishing between necrotic nuclei. This overestimates the delivery of the central necrotic nucleus and fails to indicate the bleeding risk area. This method cannot accurately calculate delivery in each area and is highly subjective.

[0054] 2. Existing AI baseline / perfusion-only soft-correlation method (comparative case): The image was processed using a baseline method that soft-weights perfusion. Because the necrotic nucleus perfusion is low but soft-weighted, the model still gives a non-zero delivery prediction for the necrotic nucleus (approximately 15% of the peritumoral area), incorrectly indicating partial delivery of the necrotic nucleus. Furthermore, this method does not provide steady-state / inertia thresholds from the microbubble population model and does not provide inertial hemorrhage risk zoning for areas with high mechanical indices; directly using this as a planning reference could be misleading.

[0055] 3. System analysis results of this invention (hard gating and security partitioning):

[0056] (1) Execution steps S101–S103: The system completes the analysis within 1 second. After linearization, a pixel-by-pixel perfusion map is obtained, with necrotic nucleus perfusion close to zero; the subject-specific steady-state threshold is derived from the particle size distribution of SonoVue at 1.7 MHz. Inertia threshold .

[0057] (2) Execution step S104: Central necrosis nucleus Hard-gated areas are considered delivery voids, with their delivery prediction value consistently at 0, unaffected even by artificially increasing the mechanical index—structurally eliminating false positives; the peritumoral hypervascular area falls into the in vivo MI. The steady-state window was judged as the effective delivery of the steady-state acoustic aperture, with an effective delivery coverage (perfusionable area) of approximately 62%, an intratumoral delivery cavity (necrotic nucleus) of approximately 26%, and an inertial risk of approximately 0%.

[0058] (3) Execution steps S105–S107: The system outputs a multi-channel safety partition map, clearly marking the necrotic core as "delivery cavity" and the peritumoral area as "stable-state acoustic hole effective delivery", and providing the parameter window that maximizes effective delivery and limits inertia in the parameter recommendations; when the scheme is changed to microbubble bursting (MI 1.13), the system according to The central area was reclassified as "inertial cavitation extravasation (with bleeding risk)" and a risk warning was given.

[0059] Comparative Conclusion: The analysis of the aforementioned real-case studies demonstrates that this invention successfully overcomes the pain point of traditional and existing methods—overestimating delivery in non-perfused necrotic areas. This system structurally ensures that delivery prediction in non-microbubble arrival areas is always zero through perfusion hard gating, eliminating false positives. Furthermore, the microbubble population model provides subject-specific cavitation windows and steady-state / inertial / void multi-channel safety partitions, offering physicians objective, stable, and safe structured delivery prediction and planning references, demonstrating significant clinical application value and beneficial effects.

[0060] Electronic device example:

[0061] Embodiments of the present invention also provide an electronic device (such as a clinical ultrasound workstation or a cloud edge computing server). This electronic device includes a processor, a memory, a communication interface, and a communication bus. The processor, memory, and communication interface communicate with each other via the communication bus. The memory stores a computer program, and when the processor executes this computer program, it implements the ultrasound-mediated drug delivery spatial prediction method described in Embodiments 1 and 2 above. The processor can be a central processing unit (CPU), graphics processing unit (GPU), application-specific integrated circuit (ASIC), or digital signal processor (DSP), etc. The memory may include high-speed random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0062] Examples of computer-readable storage media:

[0063] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the ultrasound-mediated drug delivery spatial prediction method described in the above embodiments. The computer-readable storage medium can be any medium capable of storing program code, such as a read-only memory (ROM), random access memory (RAM), magnetic disk, optical disk, or USB flash drive. Those skilled in the art should understand that the above embodiments are merely illustrative of the technical solutions of the present invention and not intended to limit it. Equivalent substitutions or improvements made within the scope of the inventive concept should be included within the protection scope of the present invention.

Claims

1. A spatial prediction method for ultrasound-mediated drug delivery, characterized in that, The method includes: Acquire contrast-enhanced ultrasound (CEUS) videos or ultrasound microvascular imaging (URM) images to be analyzed, and linearize the pixel data of the images to obtain pixel-by-pixel perfusion maps; wherein, when the pixel data is logarithmically compressed display data, according to... An inverse transformation is performed, DR is the dynamic range, and saturated pixels and truncated pixels are masked and not included in the pixel-by-pixel infusion map; Based on acoustic parameters, tissue attenuation models, and capsule microbubble population dynamics models, the image and acoustic parameters are processed to calculate the in vivo mechanical index (MI) spatial distribution and cavitation dose spatial distribution of the target region; wherein, the cavitation dose spatial distribution includes at least a steady-state cavitation dose component characterized by subharmonic and superharmonic energy, and an inertial cavitation dose component characterized by broadband noise energy. The weighted population summation of the particle size distribution of the microbubbles used by the subject was performed using the encapsulated microbubble population dynamics model to derive the subject-specific steady-state cavitation MI threshold and inertial cavitation MI threshold. The pixel-by-pixel perfusion map is used as a hard gate and combined with the spatial distribution of the in vivo mechanical index (MI) and the spatial distribution of the cavitation dose to generate a spatial delivery prediction map of the target area. For pixels with perfusion values ​​below the perfusion threshold, regardless of their MI level, their delivery prediction value is set to zero and they are identified as delivery cavitation channels. For pixels with perfusion values ​​not lower than the perfusion threshold, based on the relationship between their MI and the steady-state cavitation MI threshold and the inertial cavitation MI threshold, they are divided into multiple delivery channels, including at least a steady-state acoustic hole effective delivery channel and an inertial cavitation extravasation channel. Based on the spatial distribution of the cavitation dose, safety quality control information is generated, and the spatial delivery prediction map, clinically relevant quantitative indicators, and the safety quality control information are output.

2. The method according to claim 1, characterized in that, The classification of pixels with perfusion values ​​not lower than the perfusion threshold includes: classifying pixels whose in vivo mechanical index (MI) is between the steady-state cavitation MI threshold and the inertial cavitation MI threshold as steady-state acoustic aperture effective delivery channels and marking them as endothelial permeable delivery; classifying pixels whose in vivo mechanical index (MI) is not lower than the inertial cavitation MI threshold as inertial cavitation extravasation channels and marking them as extravasable delivery channels with bleeding risk; and marking pixels whose in vivo mechanical index (MI) is lower than the steady-state cavitation MI threshold as subthreshold ineffective delivery.

3. The method according to claim 1, characterized in that, The process of obtaining the pixel-by-pixel perfusion map includes: classifying pixels with a value greater than or equal to 254 as saturated pixels and pixels with a value less than or equal to 1 as truncated pixels, both of which are excluded from the calculation of pixel-by-pixel peak enhancement; when the image is a CEUS video, after linearizing it frame by frame, the maximum value of the linearized value of each pixel in the time series is taken as the peak enhancement of that pixel, and the peak enhancement is used as the perfusion value of that pixel; when the image is a URM color image, the saturation of the color representing the microvessels in it is used as the perfusion surrogate value of that pixel.

4. The method according to claim 1, characterized in that, The process of deriving the subject-specific steady-state cavitation MI threshold and inertial cavitation MI threshold includes: The particle size distribution of the microbubbles is sampled according to a log-normal distribution to obtain several representative radii, and each representative radius is weighted by R² scattering according to its scattering cross section to obtain the group weight of each representative radius; For each representative radius, the maximum expansion ratio of the radial vibration of the microbubble is solved using the membrane microbubble population dynamics model under a series of mechanical indices MI, and the weighted population proportions that satisfy the steady-state oscillation criterion and the inertial collapse criterion under each mechanical index MI are obtained. The mechanical index MI corresponding to the weighted population proportion that satisfies the steady-state oscillation criterion exceeding the first preset proportion is taken as the steady-state cavitation MI threshold, and the mechanical index MI corresponding to the weighted population proportion that satisfies the inertial collapse criterion exceeding the second preset proportion is taken as the inertial cavitation MI threshold.

5. The method according to claim 1, characterized in that, The perfusion threshold is the lower limit of perfusion that distinguishes between the necrotic nucleus region and the perfusion region, and is determined based on the perfusion value distribution of the target region or the perfusion value of the reference tissue. When the target region contains a tumor necrotic nucleus, the pixel corresponding to the necrotic nucleus is hard-gated as a delivery cavity because its perfusion value is lower than the perfusion threshold, and its delivery prediction value does not increase with the increase of its in vivo mechanical index MI.

6. The method according to claim 1, characterized in that, The generation of safety quality control information includes: generating a safety map registered with the spatial delivery prediction map, wherein the safety map is classified according to the degree to which the inertial cavitation dose component of each pixel exceeds the preset dose upper limit; and displaying the delivery cavity channel, the steady-state acoustic hole effective delivery channel, and the inertial cavitation extravasation channel with different identifiers on a display device; the clinically relevant quantitative indicators include the area ratio of the steady-state acoustic hole effective delivery channel, the area ratio of the inertial cavitation extravasation channel, and the area ratio of the delivery cavity channel in the target area.

7. The method according to any one of claims 1 to 6, characterized in that, The method also includes a parameter recommendation step: using the acoustic aperture window defined by the steady-state cavitation MI threshold and the inertial cavitation MI threshold as constraints, for the pixel-by-pixel perfusion map and the candidate acoustic parameter set, the spatial delivery prediction map corresponding to each candidate acoustic parameter is calculated according to the method described in claim 1, and under the constraints of maximizing the area ratio of the effective delivery channel of the steady-state acoustic aperture and ensuring that the inertial cavitation dose component does not exceed the preset dose upper limit, recommended acoustic parameters are selected from the candidate acoustic parameter set and output; the recommended acoustic parameters are output as data for treatment planning reference, and the method does not include driving the ultrasound transducer or applying ultrasound to the living body.

8. The method according to claim 1, characterized in that, It also includes a model update step: collecting multiple paired data records according to a structured template, each paired data record containing acoustic parameters, an in vivo mechanical index distribution derived from the acoustic parameters, a cavitation dose distribution, a pixel-by-pixel perfusion map, and a delivery outcome quantity measured by an objective gold standard; training a delivery quantity regression model and a safety classification model with the paired data records, wherein the delivery quantity regression model forces the predicted delivery quantity to be zero for pixels in the pixel-by-pixel perfusion map whose perfusion value is lower than the perfusion threshold; and interpolating the output of the trained model back into the generation of the spatial delivery prediction map of claim 1 or the calculation of the recommended acoustic parameters of claim 7.

9. A spatial prediction system for ultrasound-mediated drug delivery, characterized in that, include: The data input and linearization module is used to acquire the contrast-enhanced ultrasound or ultrasound microvascular imaging images to be analyzed, and to perform logarithmic decompression and linearization on the pixel data to generate a pixel-by-pixel perfusion map that makes the pixel value linearly correspond to the microbubble concentration in the target area. The physical modeling module is used to calculate the in vivo mechanical index distribution of the target area, as well as the cavitation dose distribution including steady-state cavitation dose components and inertial cavitation dose components, based on acoustic parameters, tissue attenuation models and capsule microbubble population dynamics models. The threshold determination module is used to perform a scattering cross-section weighted summation of the microbubble size distribution by the swarm dynamics model of the enveloped microbubble to determine the subject-specific steady-state cavitation mechanical index threshold and inertial cavitation mechanical index threshold. The perfusion gating and partitioning module is used to combine the pixel-by-pixel perfusion map as a hard gating term with the in vivo mechanical index distribution and cavitation dose distribution. For pixels with perfusion values ​​lower than the perfusion threshold, the delivery prediction value is set to zero and marked as a delivery cavity. For the remaining pixels, they are divided into steady-state acoustic hole effective delivery channels and inertial cavitation extravasation channels according to the mechanical index, generating a spatial delivery prediction map and a safety partition. The interactive output module is used to present the spatial delivery prediction map, clinically relevant quantitative indicators, and safety and quality control information that assists in manual confirmation.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 8.

Images