Method and system for evaluating coupling state of transducer based on bubble characteristics in visual image

Abstract

The invention discloses a transducer coupling state evaluation method and system based on bubble characteristics in a visual image, and the scheme comprises the steps: S1, focusing ultrasonic coupling surface image collection: collecting an image of a focusing ultrasonic treatment transducer coupling surface, and transmitting the captured image data to a subsequent processing unit in real time; s2, image pre-processing: performing image pre-processing after judging the acquired image; s3, bubble detection and feature calculation: according to the image input in the step S2, performing bubble and contact surface area identification by using a deep learning-based framework model to obtain actual contour information of bubbles and contact surfaces for subsequent processing; and S4, safety evaluation and early warning: according to the contour information in the step S3, calculating an ultrasonic propagation loss L, and defining a level for the ultrasonic propagation loss. According to the invention, the coupling state of the transducer is monitored and evaluated, timely safety evaluation and early warning information is provided for operators, and the safety and effectiveness of the treatment process are ensured.

Description

Technical Field

[0001] This invention relates to the field of medical focused ultrasound, and in particular to a method and system for evaluating transducer coupling status based on bubble features in visual images. Background Technology

[0002] Focused ultrasound (FUS) therapy, as a non-invasive treatment method, has shown significant effectiveness in treating various diseases. However, its therapeutic effect is highly dependent on the effective transmission path of ultrasound waves. A coupling agent is needed at the contact surface between the water bag and the skin to ensure optimal sound wave transmission efficiency.

[0003] However, due to its inherent properties, the coupling agent itself may contain air gaps, i.e., bubbles. These bubbles can severely interfere with the normal propagation of ultrasound waves, leading to energy loss, focus shift, and other problems. These issues not only affect the treatment outcome but may also increase unnecessary risks. Therefore, accurately identifying and controlling bubbles on the contact surface is a crucial step in ensuring the safety and effectiveness of focused ultrasound therapy.

[0004] Currently, the treatment of bubbles on the contact surface mainly relies on the experience and manual inspection of doctors or operators. Some studies have attempted to improve the formulation of coupling agents and the design of application tools to reduce the possibility of bubble formation, such as optimizing the composition of coupling agents to improve their fluidity and smoothness, or developing specially designed applicators to distribute the coupling agent evenly. In addition, there are studies on vacuum degassing technology to remove dissolved gases from the coupling agent, thereby reducing the possibility of bubble formation.

[0005] While the above methods alleviate the problems caused by bubbles to some extent, they raise the following three issues: First, these methods can only guarantee no (or few) bubbles for a short period after the coupling agent is applied, and cannot be adapted to long-term treatment processes; second, even when combined with manual inspection, these methods still cannot reliably identify bubbles, because a major drawback of manual inspection is the inability to observe the bubble situation from a suitable angle and under suitable lighting; third, there is no quantitative evaluation model to analyze the impact of bubbles in the current treatment contact on the subsequent treatment effect. Summary of the Invention

[0006] The purpose of this invention is to provide a transducer coupling state assessment method and system based on bubble features in visual images. Through steps such as multi-view image acquisition, image stitching, bubble detection, parameter calculation, and safety assessment, the transducer coupling state is monitored and assessed, providing operators with timely safety assessment and early warning information to ensure the safety and effectiveness of the treatment process.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for evaluating transducer coupling state based on bubble features in visual images includes the following steps:

[0009] S1. Focused ultrasound coupling surface image acquisition: The image acquisition device is used to acquire images of the coupling surface of the focused ultrasound treatment transducer, and the captured image data is transmitted to the subsequent processing unit in real time through the data transmission interface.

[0010] S2. Image preprocessing: After judging the acquired image, image preprocessing is performed;

[0011] S3. Bubble Detection and Feature Calculation: Based on the image input in step S2, a deep learning framework model is used to identify bubbles and contact surface regions, and to obtain the actual contour information of bubbles and contact surfaces for subsequent processing.

[0012] S4. Safety Assessment and Early Warning: Based on the contour information in step S3, calculate the ultrasonic propagation loss L and define the level of ultrasonic propagation loss.

[0013] Furthermore, in step S1, the image acquisition device uses a single camera or multiple cameras; if the treatment area is small, a single high-speed camera is used to cover the entire treatment contact surface, and the camera is ensured to have sufficient resolution and viewing angle to completely capture the target area; for larger treatment areas or complex geometries, multiple high-speed cameras are used in combination, and each camera needs to be precisely calibrated to ensure that there is no ghosting or distortion when the images are stitched together.

[0014] Furthermore, in step S2, firstly, the system determines whether the currently acquired image is a single image or multiple images; then, based on the determination result, the system performs different preprocessing steps.

[0015] The preprocessing steps for a single image are as follows:

[0016] Coverage check: Confirm whether a single image completely covers the treatment coupling surface; if the image can completely contain the entire treatment area, no additional processing is required; if the image does not completely cover the treatment coupling surface, adjust the camera zoom or increase the distance between the camera and the coupling surface to expand the field of view; if the treatment coupling surface still cannot be completely covered after adjustment, implement a multi-camera solution.

[0017] Direct transmission: If the image coverage is verified to be correct, the image is directly transmitted to the subsequent processing unit.

[0018] The preprocessing steps for multiple images are as follows:

[0019] First, each image is denoised, color corrected, and resized to ensure consistency and compatibility.

[0020] Next, SIFT is used to extract feature points from the image, and FLANN is used to find the best matching point pair.

[0021] Then, based on the matched feature point pairs, the homography matrix is ​​calculated, and the image is aligned accordingly.

[0022] Apply fade-in / fade-out to the overlapping areas to ensure the stitched image is seamless and natural, ultimately generating a high-quality, complete contact surface image for subsequent analysis.

[0023] Furthermore, in step S3, the bubble and contact surface area are identified as follows:

[0024] First, a rectangular target region is obtained through real-time object detection. Then, a secondary analysis of the rectangular region is performed to extract a more accurate contour. Specifically, the real-time object detection framework uses the YOLO model, and the secondary analysis process is as follows:

[0025] (1) Gaussian filtering is applied to the extracted bubble region to remove noise interference;

[0026] (2) Use the Canny edge detection algorithm to extract the boundary information of the bubble and the contact surface;

[0027] (3) The image is binarized using the Otsu thresholding method to achieve clear target segmentation;

[0028] (4) Perform morphological closing operations to fill small holes in the edges and connect broken boundaries;

[0029] (5) Use the contour search algorithm to locate and extract the specific contours of the bubble and the contact surface.

[0030] Furthermore, the contour information in step 3 includes:

[0031] Bubble percentage P: represents the proportion of the contact surface area occupied by bubbles; the area of ​​each bubble is summed up. and total contact area Through formula The proportion P of the bubble to the total contact surface area is obtained;

[0032] Bubble density D: represents the number of bubbles per unit area; the number of bubbles per unit area is expressed as... To indicate, among which It refers to the number of bubbles;

[0033] Bubble size distribution B: Calculates the proportion of bubble sizes distributed across several pixel intervals.

[0034] Furthermore, in step 4, based on the contour information in step S3: bubble proportion P, bubble density D, and bubble size distribution B, the contour information is imported into the evaluation and analysis model to calculate the ultrasonic propagation loss L, and the ultrasonic propagation loss is defined into levels of low risk, medium risk, and high risk. The specific risk level is given based on the output parameters of the model.

[0035] Furthermore, the evaluation and analysis model is an SVM (Support Vector Machine) model, specifically represented as follows:

[0036] Radial basis functions as kernel functions:

[0037]

[0038] This function is used to measure the similarity between two samples, with parameters... It controls the shape of the function. Let represent the Euclidean distance between two vectors i and j;

[0039] The objective of SVM regression is to minimize the following objective function:

[0040]

[0041] In the objective function is the L2 norm of the weight vector, which controls the model complexity; C is the regularization parameter, used to control the degree of penalty for errors. These are slack variables used to handle samples that exceed the error range; their constraints are:

[0042]

[0043]

[0044] 0

[0045] Of the above constraints, It is the true value of sample i. It is the error tolerance, representing the allowable range of prediction deviation; where, Let represent the prediction of the support vector machine model for the i-th sample, and its expression is:

[0046]

[0047] In the formula:

[0048] , : Lagrange multipliers.

[0049] : Kernel function, here it is RBF kernel;

[0050] b: Bias term;

[0051] Summation index j: corresponds to all training samples or all support vectors;

[0052] above , Both b are learned automatically during the training process.

[0053] A transducer coupling state evaluation system based on bubble features in visual images, the system being used to implement the aforementioned evaluation method, specifically comprising:

[0054] The focused ultrasound coupling surface image acquisition module is used to acquire images of the coupling surface of the focused ultrasound therapy transducer and transmits the captured image data to the subsequent processing unit in real time through the data transmission interface.

[0055] The image preprocessing module is used to preprocess the acquired images after making a judgment.

[0056] Bubble detection and feature calculation module: used to identify bubbles and contact surface areas, and obtain the actual contour information of bubbles and contact surfaces for subsequent processing;

[0057] The safety assessment and early warning module is used to calculate the ultrasonic propagation loss L and define the level of ultrasonic propagation loss.

[0058] A computer device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the evaluation method described above when executing the computer-readable instructions.

[0059] A readable storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the evaluation method described above.

[0060] The beneficial effects of this invention are:

[0061] This invention achieves the monitoring and evaluation of transducer coupling status through steps such as multi-view image acquisition, image stitching, bubble detection, parameter calculation, and safety assessment. It provides operators with timely safety assessment and early warning information, ensuring the safety and effectiveness of the treatment process. Attached Figure Description

[0062] Figure 1 This is a flowchart of the overall method of the present invention;

[0063] Figure 2 This is a diagram illustrating the multi-camera image processing procedure.

[0064] Figure 3This is a diagram illustrating the secondary processing of the target detection results;

[0065] Figure 4 This is a schematic diagram of the basic structure of a bubble detection device;

[0066] Figure 5 This is a schematic diagram of image processing in an embodiment of the present invention;

[0067] Figure 6 The bubble marking tool and partial dataset in this embodiment of the invention;

[0068] Figure 7 These are the bubble detection results in the embodiments of the present invention;

[0069] Figure 8 This is the bubble boundary segmentation result in the embodiment of the present invention;

[0070] Figure 9 These are the actual prediction results and the true value results in the embodiments of this invention. Detailed Implementation

[0071] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0072] like Figures 1 to 3 This invention provides a method for evaluating the coupling state of a transducer based on bubble features in a visual image. The specific evaluation process is as follows:

[0073] S1. Focused ultrasound coupling surface image acquisition:

[0074] A camera is used to capture images of the coupling surface of the focused ultrasound therapy transducer, and the data transmission interface transmits the captured image data to the subsequent processing unit in real time.

[0075] The above image acquisition includes:

[0076] Single-camera solution: If the treatment area is small, a single high-speed camera is sufficient to cover the entire treatment contact surface. It is necessary to ensure that the camera has sufficient resolution and field of view to completely capture the target area.

[0077] Multi-camera solution: For large treatment areas or complex geometries, multiple high-speed cameras are used in combination. Precise calibration is required between each camera to ensure no ghosting or distortion during image stitching. Marker points can be used to assist in camera alignment.

[0078] The above-mentioned coupling surface images support color images (RGB), grayscale images, and infrared images.

[0079] S2. Image preprocessing:

[0080] After judging the acquired images, image preprocessing is performed.

[0081] A. Image Judgment Process: First, the system needs to automatically determine whether a single image or multiple images are being captured. This step is based on the camera configuration information (i.e., whether a single or multiple cameras are being used).

[0082] Single-camera acquisition scheme: If the system detects that only one camera is being used for acquisition, it defaults to a single image.

[0083] Multi-camera acquisition scheme: If the system detects that multiple cameras are being used, it is expected to receive multiple images, with each image corresponding to the viewpoint of one camera.

[0084] B. Preprocessing Flow: Based on the above judgment results, the system will perform different preprocessing steps:

[0085] Preprocessing for single-camera solutions:

[0086] Coverage check: Confirm that a single image completely covers the treatment coupling surface (this function can be achieved through training on pre-defined boundary markers and sizes). If the image completely encompasses the entire treatment area, no additional processing is required. If the image does not completely cover the treatment coupling surface, one approach is to adjust the camera zoom or increase the distance between the camera and the coupling surface to expand the field of view. If the treatment coupling surface still cannot be completely covered, a multi-camera solution should be considered.

[0087] Direct transmission: Once the image coverage is verified to be correct, the image will be directly transmitted to the subsequent processing unit.

[0088] Multi-camera solution preprocessing:

[0089] For multi-camera solutions, image stitching is required, the process is as follows: Figure 2 As shown:

[0090] First, each image undergoes preprocessing operations such as noise reduction, color correction, and resizing to ensure image consistency and compatibility.

[0091] Next, SIFT is used to extract feature points from the image, and FLANN is used to find the best matching point pair.

[0092] Then, based on the matched feature point pairs, the homography matrix is ​​calculated, and the image is aligned accordingly.

[0093] Feathering is applied to the overlapping areas to ensure a seamless and natural stitched image. This ultimately generates a high-quality, complete contact surface image for subsequent analysis.

[0094] S3. Bubble Detection and Feature Calculation:

[0095] Based on the stitched image input in step S2, an algorithm model based on a deep learning framework is used to identify the bubble and contact surface regions. This step obtains the actual contour information of the bubble and contact surface, which can be used for subsequent processing.

[0096] A. The target (bubble and contact surface) can be extracted using the following two methods.

[0097] First, a real-time object detection model is combined with object boundary extraction. The advantages of using a real-time object detection model are its fast detection speed and high efficiency. First, a rectangular target region is obtained through real-time object detection. Then, a secondary analysis of this rectangular region is performed to extract a more accurate contour. The real-time object detection framework primarily uses YOLO. The rectangular target region output by object detection needs to undergo secondary analysis to extract more accurate boundary contour information. The process is as follows:

[0098] The extracted bubble region is processed by Gaussian filtering to remove noise interference.

[0099] The Canny edge detection algorithm was used to extract the boundary information of the bubble and the contact surface.

[0100] The Otsu thresholding method was used to binarize the image, achieving clear target segmentation.

[0101] Perform morphological closing operations to fill small holes in the edges and connect broken boundaries.

[0102] The contour lookup algorithm is used to locate and extract the specific contours of the bubble and the contact surface.

[0103] Secondly, an image semantic segmentation and instance segmentation task model is adopted, which can perform pixel-level segmentation. The Mask-RCNN model can be used.

[0104] B. The data features of the contour information as described above include:

[0105] Bubble percentage P: This represents the proportion of the contact surface area occupied by bubbles, i.e., the sum of the areas of each bubble. and total contact area Through the formula The proportion P of the bubble to the total contact area was obtained.

[0106] Bubble density D: The number of bubbles per unit area. The number of bubbles per unit area (bubble density) can be expressed as... To indicate, among which It is the number of bubbles; a high-density bubble distribution can significantly affect the propagation path of ultrasound.

[0107] Bubble size distribution B: Calculates the proportion of bubble sizes distributed across several pixel intervals (bins).

[0108] S4. Safety Assessment and Early Warning:

[0109] Based on the bubble data characteristics (bubble percentage P, bubble density D, and bubble size distribution B) from step S3, the model is imported into the evaluation and analysis model to calculate the ultrasonic propagation loss L and define its risk level as low, medium, or high. The specific risk level is then determined based on the model's output parameters.

[0110] A. The feature data is represented as follows:

[0111] Bubble percentage P: This represents the proportion of the contact surface area occupied by bubbles.

[0112] Bubble density D: The number of bubbles per unit area.

[0113] Bubble size distribution B: can be represented by a vector, for example, , where Bi represents the value of bubble size distribution in the i-th interval (such as proportion or quantity).

[0114] Input features: The features of each sample are

[0115] Output target L: L is the acoustic transmission loss (a continuous value, e.g., in dB units).

[0116] B. The evaluation and analysis model is the SVM (Support Vector Machine) model, specifically represented as follows:

[0117] Radial basis functions (RBFs) as kernel functions:

[0118]

[0119] This function is primarily used to measure the similarity between two samples. Parameters It controls the shape of the function. Let represent the Euclidean distance between two vectors i and j.

[0120] The goal of SVM regression is to minimize the following objective function:

[0121]

[0122] In the objective function is the L2 norm of the weight vector, controlling model complexity (margin size). C is the regularization parameter, used to control the degree of penalty for errors. These are slack variables used to handle samples that exceed the error range. Their constraints are:

[0123]

[0124]

[0125] 0

[0126] Of the above constraints, It is the true value of sample i. This is the error tolerance, representing the allowable range of prediction deviation. Let represent the prediction of the support vector machine model for the i-th sample, and its expression is:

[0127]

[0128] in:

[0129] , Lagrange multipliers.

[0130] : Kernel function, here it is RBF kernel.

[0131] b: Bias term.

[0132] Summation index j: corresponds to all training samples (or all support vectors).

[0133] above , Both b are learned automatically during the training process.

[0134] To support the above-mentioned evaluation method, this invention also provides a transducer coupling state evaluation system based on bubble features in visual images, specifically including:

[0135] The focused ultrasound coupling surface image acquisition module is used to acquire images of the coupling surface of the focused ultrasound therapy transducer and transmits the captured image data to the subsequent processing unit in real time through the data transmission interface.

[0136] 1. Image preprocessing module, used to preprocess the acquired images after judgment;

[0137] Bubble detection and feature calculation module: used to identify bubbles and contact surface areas, and obtain the actual contour information of bubbles and contact surfaces for subsequent processing;

[0138] The safety assessment and early warning module is used to calculate the ultrasonic propagation loss L and define the level of ultrasonic propagation loss.

[0139] The following is a further explanation of the application of the above methods in specific cases:

[0140] Focused ultrasound coupling surface image acquisition:

[0141] The embodiment uses a multi-camera solution, that is, installing two high-speed cameras.

[0142] As shown in Figure 4: The data acquisition device includes a water tank H1, a transducer H2, a water bladder H3, and two cameras H4 and H5.

[0143] Water tank H1: It is the basic support structure for all related hardware (such as cameras, transducers and water bladders), ensuring that they remain in the correct position during use.

[0144] Two high-speed cameras, H4 and H5, are used for image acquisition. They are mounted symmetrically to the center of the treatment head and attached to a structural member located at the center of the transducer. Parameters of the two cameras:

[0145]

[0146] Transducer H2: Used to emit ultrasound waves and radiate energy to the treatment area to achieve a therapeutic effect;

[0147] Water-filled bladder H3: Used to adhere to the skin surface, ensuring that ultrasound waves can be efficiently transmitted from the treatment head to the target area inside the body.

[0148] 2. Image preprocessing:

[0149] For the dual-camera solution used above, image stitching of the two acquired images is required. The stitching process is as described in the preprocessing flow of the multi-camera solution in part B of step S2 above. The actual stitching effect is as follows. Figure 5 As shown.

[0150] 3. Bubble detection and feature calculation:

[0151] For the fused images, a YOLO algorithm combined with secondary detection is primarily used to extract the actual contour information. During YOLO training, labelImg is used to annotate the bubbles and contact surfaces. The annotation process is as follows: Figure 6 As shown, the dataset contains 2000 images, with a training dataset to test dataset ratio of 7:3.

[0152] After the model training is completed, the actual inference results are as follows: Figure 7 As shown.

[0153] The YOLO detection area requires secondary processing. The secondary processing procedure is as described in Part A of step S3, and the actual processing effect is as follows: Figure 8 As shown.

[0154] Feature information is extracted based on the image calculation results. Among them, the bubble size distribution B, as described in part B of step S3, is mainly divided into small bubble proportion B1, medium bubble proportion B2, and large bubble proportion B3. The ultrasonic loss L is the corresponding sample measurement value.

[0155] ;

[0156] ;

[0157] .

[0158] The data characteristics of some samples from 1000 bubble images are shown in the table below:

[0159]

[0160] 4. Safety assessment and early warning:

[0161] The primary model used is the Support Vector Machine (SVM), with the following parameters:

[0162] Kernel function: Radial basis function (RBF);

[0163] Error tolerance ;

[0164] Regularization parameters 100;

[0165] Parameters of RBF core .

[0166] For example, the actual prediction results and the true values ​​are compared. Figure 9 As shown.

[0167] Definition of ultrasonic propagation loss level L (corresponding to the transducers involved in the implementation):

[0168] Low level: ;

[0169] Intermediate level: 10 ;

[0170] High level: 0 .

[0171] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0172] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the scope of protection of the present invention in any way, and all technical solutions obtained by equivalent substitution or other means fall within the scope of protection of the present invention. Parts not covered in this invention are the same as or can be implemented using existing technology.

Claims

1. A method for evaluating the coupling state of a transducer based on bubble features in a visual image, characterized in that, Includes the following steps: S1. Focused ultrasound coupling surface image acquisition: The image acquisition device is used to acquire images of the coupling surface of the focused ultrasound treatment transducer, and the captured image data is transmitted to the subsequent processing unit in real time through the data transmission interface. S2. Image preprocessing: After judging the acquired image, image preprocessing is performed; S3. Bubble Detection and Feature Calculation: Based on the image input in step S2, a deep learning framework model is used to identify bubbles and contact surface regions, and to obtain the actual contour information of bubbles and contact surfaces for subsequent processing. S4. Safety Assessment and Early Warning: Based on the contour information in step S3, calculate the ultrasonic propagation loss L and define the level of ultrasonic propagation loss.

2. The transducer coupling state evaluation method based on bubble features in a visual image according to claim 1, characterized in that, In step S1, the image acquisition device uses a single camera or multiple cameras. If the treatment area is small, a single high-speed camera is used to cover the entire treatment contact surface, and the camera is ensured to have sufficient resolution and viewing angle to completely capture the target area. For larger treatment areas or complex geometries, multiple high-speed cameras are used in combination, and each camera needs to be precisely calibrated to ensure that there is no ghosting or distortion when the images are stitched together.

3. The transducer coupling state evaluation method based on bubble features in a visual image according to claim 1, characterized in that, In step S2, firstly, the system determines whether the current acquisition is of a single image or multiple images; then, based on the determination result, the system performs different preprocessing steps. The preprocessing steps for a single image are as follows: Coverage check: Confirm whether a single image completely covers the treatment coupling surface; if the image can completely contain the entire treatment area, no additional processing is required; if the image does not completely cover the treatment coupling surface, adjust the camera zoom or increase the distance between the camera and the coupling surface to expand the field of view; if the treatment coupling surface still cannot be completely covered after adjustment, implement a multi-camera solution. Direct transmission: If the image coverage is verified to be correct, the image is directly transmitted to the subsequent processing unit; The preprocessing steps for multiple images are as follows: First, each image is denoised, color corrected, and resized to ensure consistency and compatibility. Next, SIFT is used to extract feature points from the image, and FLANN is used to find the best matching point pair. Then, based on the matched feature point pairs, the homography matrix is ​​calculated, and the image is aligned accordingly. Apply fade-in / fade-out to the overlapping areas to ensure the stitched image is seamless and natural, ultimately generating a high-quality, complete contact surface image for subsequent analysis.

4. The transducer coupling state evaluation method based on bubble features in a visual image according to claim 1, characterized in that, In step S3, the bubble and contact surface area are identified as follows: First, a rectangular target region is obtained through real-time object detection. Then, a secondary analysis of the rectangular region is performed to extract a more accurate contour. Specifically, the real-time object detection framework uses the YOLO model, and the secondary analysis process is as follows: (1) Gaussian filtering is applied to the extracted bubble region to remove noise interference; (2) Use the Canny edge detection algorithm to extract the boundary information of the bubble and the contact surface; (3) The image is binarized using the Otsu thresholding method to achieve clear target segmentation; (4) Perform morphological closing operations to fill small holes in the edges and connect broken boundaries; (5) Use the contour search algorithm to locate and extract the specific contours of the bubble and the contact surface.

5. The transducer coupling state evaluation method based on bubble features in a visual image according to claim 1, characterized in that, The contour information in step 3 includes: Bubble percentage P: represents the proportion of the contact surface area occupied by bubbles; the area of ​​each bubble is summed up. and total contact area Through formula The proportion P of the bubble to the total contact surface area is obtained; Bubble density D: represents the number of bubbles per unit area; the number of bubbles per unit area is expressed as... To indicate, among which It refers to the number of bubbles; Bubble size distribution B: Calculates the proportion of bubble sizes distributed across several pixel intervals.

6. The transducer coupling state evaluation method based on bubble features in a visual image according to claim 1, characterized in that, In step 4, based on the contour information in step S3: bubble proportion P, bubble density D, and bubble size distribution B, the contour information is imported into the evaluation and analysis model to calculate the ultrasonic propagation loss L. The ultrasonic propagation loss is defined into levels, which are low risk, medium risk, and high risk. The specific risk level is given according to the output parameters of the model.

7. The transducer coupling state evaluation method based on bubble features in a visual image according to claim 6, characterized in that, The evaluation and analysis model is the Support Vector Machine (SVM) model, and the specific model representation is as follows: Radial basis functions as kernel functions: This function is used to measure the similarity between two samples, with parameters... It controls the shape of the function. Let represent the Euclidean distance between two vectors i and j; The objective of SVM regression is to minimize the following objective function: In the objective function is the L2 norm of the weight vector, which controls the model complexity; C is the regularization parameter, used to control the degree of penalty for errors. These are slack variables used to handle samples that are outside the error range; Its constraints are: 0 Of the above constraints, It is the true value of sample i. It is the error tolerance, representing the allowable range of prediction deviation; where, Let represent the prediction of the support vector machine model for the i-th sample, and its expression is: In the formula: , Lagrange multipliers; : Kernel function, here it is RBF kernel; b: Bias term; Summation index j: corresponds to all training samples or all support vectors; above , Both b are learned automatically during the training process.

8. A transducer coupling state evaluation system based on bubble features in visual images, characterized in that, The system is used to implement the evaluation method as described in any one of claims 1 to 7, and the system specifically includes: The focused ultrasound coupling surface image acquisition module is used to acquire images of the coupling surface of the focused ultrasound therapy transducer and transmits the captured image data to the subsequent processing unit in real time through the data transmission interface. The image preprocessing module is used to preprocess the acquired images after making a judgment. Bubble detection and feature calculation module: used to identify bubbles and contact surface areas, and obtain the actual contour information of bubbles and contact surfaces for subsequent processing; The safety assessment and early warning module is used to calculate the ultrasonic propagation loss L and define the level of ultrasonic propagation loss.

9. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, characterized in that, When the processor executes the computer-readable instructions, it implements the evaluation method as described in any one of claims 1 to 7.

10. A readable storage medium storing computer-readable instructions, characterized in that, When the computer-readable instructions are executed by one or more processors, the one or more processors cause the one or more processors to perform the evaluation method as described in any one of claims 1 to 7.

Images