Human organ autonomous scanning method and device based on Bayesian optimal search

Abstract

The invention discloses a Bayesian optimal search-based human organ autonomous scanning method and device, and the method comprises the steps: training an initial agent model according to an initial probe pose data set and an initial ultrasonic image scoring data set, and carrying out the calculation of the number of iterations and the number of artificial fishes of an artificial fish swarm algorithm according to the data size of the data sets, finding a next probe pose by adopting an artificial fish swarm algorithm; a probe of the ultrasonic robot is controlled to move to the next probe pose, a new ultrasonic image is acquired in real time, and an updated probe pose data set and an updated ultrasonic image scoring data set are obtained; updating the initial proxy model based on the updated probe pose data set and the updated ultrasonic image scoring data set to obtain an updated proxy model; and the Bayesian optimal search process is executed again until a preset stop condition is met, and the optimal probe pose and the corresponding optimal ultrasonic image are obtained. According to the method, the autonomous scanning quality and efficiency of the ultrasonic robot are remarkably improved.

Description

Technical Field

[0001] The invention relates to a method and a device for autonomously scanning human organs based on Bayesian optimal search. Background Art

[0002] With the advancement of medical technology, ultrasound examination has been widely used in clinical diagnosis due to its non-invasive, real-time and cost-effective characteristics. However, traditional ultrasound examination mainly relies on manual operation by doctors, which not only requires doctors to have rich experience and superb skills, but also the examination results are easily affected by the doctor's personal technical level, resulting in poor consistency and repeatability of the examination results. In order to overcome these limitations and improve the accuracy and efficiency of ultrasound examination, ultrasound robot technology has gradually become a research hotspot. This technology can not only reduce the workload of doctors through automation and intelligent methods, but also significantly improve the stability and reliability of examination results, thereby promoting ultrasound examination to develop in a more accurate and efficient direction. Summary of the invention

[0003] In order to obtain an optimal ultrasound image of a target organ, an embodiment of the present invention provides a method and device for autonomous scanning of human organs based on Bayesian optimal search.

[0004] In a first aspect, an embodiment of the present invention provides a method for autonomous scanning of human organs based on Bayesian optimal search, which may include:

[0005] An initial proxy model is trained based on the acquired initial probe pose dataset and initial ultrasound image scoring dataset;

[0006] Calculating the number of iterations of the artificial fish swarm algorithm and the number of artificial fish according to the data volume of the initial ultrasound image scoring data set;

[0007] Starting from the probe's initial posture, the artificial fish school algorithm is used to find the next probe posture based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish;

[0008] Controlling the probe of the ultrasonic robot to move to the next probe posture, and acquiring a new ultrasonic image in real time, and updating the initial probe posture data set and the initial ultrasonic image scoring data set according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set;

[0009] Update the initial proxy model based on the updated probe pose dataset and the updated ultrasound image scoring dataset to obtain an updated proxy model;

[0010] The above-mentioned Bayesian optimal search process is re-executed until a preset stop condition is reached to obtain an optimal probe posture, and the probe is controlled to acquire an optimal ultrasound image of the target organ at the optimal probe posture.

[0011] In one or some optional implementations of the embodiment of the present application, the step of calculating the number of iterations of the artificial fish swarm algorithm and the number of artificial fish according to the data volume of the initial ultrasound image scoring data set includes:

[0012] According to the data volume of the initial ultrasound image scoring data set, the number of iterations of the artificial fish swarm algorithm is calculated based on the following formula:

[0013]

[0014] Where G is the number of iterations of the artificial fish swarm algorithm, G max is the preset maximum number of increased iterations, G base is the preset iteration base, S p is the data volume of the initial ultrasound image scoring dataset, S max To preset the maximum data volume, is the ceiling rounding function;

[0015] According to the data volume of the initial ultrasound image scoring data set, the number of artificial fish of the artificial fish swarm algorithm is calculated based on the following formula:

[0016]

[0017] Where, F is the number of artificial fish in the artificial fish swarm algorithm, F max The maximum number of artificial fish to be added is F. base is the preset artificial fish base number, S p is the data volume of the initial ultrasound image scoring dataset, S max To preset the maximum data volume, is the ceiling function.

[0018] In one or some optional implementations of the embodiments of the present application, the control of the probe of the ultrasonic robot to move to the next probe posture, and acquiring a new ultrasonic image in real time, and updating the initial probe posture data set and the initial ultrasonic image scoring data set according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set, including:

[0019] Controlling the probe of the ultrasonic robot to move to the next probe posture, and collecting multiple new ultrasonic images and the probe posture corresponding to each new ultrasonic image in real time, and determining the score corresponding to each new ultrasonic image;

[0020] The probe poses and scores corresponding to the multiple new ultrasound images are added to the initial probe pose dataset and the initial ultrasound image score dataset respectively, and data sparse downsampling is performed at the same time to obtain an updated probe pose dataset and an updated ultrasound image score dataset.

[0021] In one or some optional implementations of the embodiment of the present application, the updated probe posture dataset, the updated ultrasound image scoring dataset, and the multiple new ultrasound images are sorted from old to new according to acquisition time;

[0022] The step of adding the probe poses and scores corresponding to the multiple new ultrasound images to the initial probe pose dataset and the initial ultrasound image score dataset respectively, and performing data sparse downsampling at the same time to obtain an updated probe pose dataset and an updated ultrasound image score dataset, comprises:

[0023] Calculate the L2 norm of the probe pose corresponding to the last probe pose in the initial probe pose data set and the first new ultrasound image, and determine whether the L2 norm is greater than a preset sampling interval:

[0024] If yes, then adding the probe pose corresponding to the first new ultrasound image to the initial probe pose data set, and adding the score corresponding to the first new ultrasound image to the initial ultrasound image score data set;

[0025] If not, the above process of calculating the L2 norm is re-executed based on the subsequent new ultrasound image until there is no subsequent new ultrasound image, thereby obtaining an updated probe pose data set and an updated ultrasound image scoring data set.

[0026] In one or some optional implementations of the embodiment of the present application, the data in the updated probe posture dataset and the updated ultrasound image scoring dataset are sorted by acquisition time;

[0027] After obtaining the updated probe pose dataset and the updated ultrasound image scoring dataset, it also includes:

[0028] The preset maximum number of probe poses and scores sorted at the end of the updated probe pose data set and the updated ultrasound image score data set are retained, and the other data are deleted.

[0029] In one or some optional implementations of the embodiments of the present application, starting from the probe starting posture, based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish, the artificial fish school algorithm is used to find the next probe posture, including:

[0030] Initializing a plurality of artificial fishes according to the starting position of the probe and the number of the artificial fishes; wherein each of the artificial fishes corresponds to a position;

[0031] Calculating a sampling potential value of each artificial fish according to the initial proxy model and the acquisition function;

[0032] Controlling all the artificial fish to perform the flocking behavior, tail-chasing behavior, foraging behavior and random behavior in the artificial fish school algorithm based on the corresponding sampling potential values, iteratively updating all the artificial fish until the iteration number of the artificial fish school algorithm is reached, and obtaining a plurality of updated artificial fish;

[0033] Calculating a sampling potential value of each updated artificial fish according to the initial proxy model and the acquisition function;

[0034] The posture corresponding to the updated artificial fish with the largest update sampling potential value is used as the next probe posture.

[0035] In a second aspect, an embodiment of the present invention provides a human organ autonomous scanning device based on Bayesian optimal search, which may include:

[0036] A first training module is used to train an initial proxy model according to the acquired initial probe posture dataset and initial ultrasound image scoring dataset;

[0037] A first calculation module, configured to calculate the number of iterations of the artificial fish swarm algorithm and the number of artificial fish according to the data volume of the initial ultrasound image scoring data set;

[0038] A first optimization module is used to start from the probe's initial posture and find the next probe posture by using the artificial fish school algorithm based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish;

[0039] A first updating module is used to control the probe of the ultrasonic robot to move to the next probe posture, and acquire a new ultrasonic image in real time, and update the initial probe posture data set and the initial ultrasonic image scoring data set according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set;

[0040] A second updating module is used to update the initial proxy model based on the updated probe pose dataset and the updated ultrasound image scoring dataset to obtain an updated proxy model;

[0041] The first judgment module is used to judge whether the preset stop condition is met: if so, the optimal probe posture is obtained, and the probe is controlled to acquire the optimal ultrasound image of the target organ at the optimal probe posture; if not, the above-mentioned first calculation module is re-executed.

[0042] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the above-mentioned autonomous scanning method of human organs based on Bayesian optimal search.

[0043] In a fourth aspect, an embodiment of the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the above-mentioned method for autonomous scanning of human organs based on Bayesian optimal search.

[0044] In a fifth aspect, an embodiment of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory, wherein when the processor executes the computer program, the method for autonomous scanning of human organs based on Bayesian optimal search as described above is implemented.

[0045] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0046] The embodiment of the present invention provides a method for autonomous scanning of human organs based on Bayesian optimal search. The method obtains an initial proxy model through training, and calculates the number of iterations and the number of fish swarms of an artificial fish swarm algorithm (AFSA) according to the data volume of the initial ultrasound image scoring data set, and uses the artificial fish swarm algorithm as an optimization algorithm to find the optimal next probe posture, control the ultrasound probe to move to the next probe posture, collect new ultrasound images during the movement, update the data set and the proxy model, repeat the above-mentioned Bayesian optimal search algorithm process until the stop condition is met, obtain the optimal probe posture, and control the probe to collect the optimal ultrasound image of the target organ under the optimal probe posture. The method establishes a proxy model through the Bayesian optimal search algorithm, predicts the scores under different probe postures, uses the artificial fish swarm algorithm as the optimization algorithm of the Bayesian optimal search algorithm, combines the acquisition function to determine the next posture that is most likely to improve the scanning result, guides the ultrasonic robot device to approach the optimal solution, and then gradually approaches the optimal probe posture by continuously updating the proxy model and optimizing the acquisition function.

[0047] This method can also dynamically adjust the number of artificial fish and the number of iterations of the artificial fish swarm algorithm according to the amount of data in the data set during the iteration round, so that when the amount of data is small and the training time of the proxy model is short, the number of iterations and the number of artificial fish are increased, so that the artificial fish swarm algorithm can perform a deeper search; and when the amount of data increases and the training time of the proxy model is long, the number of iterations and the number of artificial fish are reduced to reduce the search time. This dynamic adjustment mechanism effectively solves the problem of fluctuations in processing time of different iteration rounds in the Bayesian optimal search process, ensuring the stability and efficiency of the Bayesian optimal search.

[0048] This method significantly improves the acquisition quality and efficiency of ultrasound images through intelligent and automated means. This method not only reduces the workload of ultrasound doctors and improves the utilization efficiency of medical resources, but also ensures the consistency and repeatability of examination results by continuously optimizing the search path and image quality, thereby improving the accuracy and reliability of clinical diagnosis.

[0049] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0050] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0052] Figure 1 A schematic diagram of a flow chart of a method for autonomous scanning of human organs based on Bayesian optimal search provided by an embodiment of the present invention;

[0053] Figure 2 A schematic diagram of an ultrasonic robot device provided by an embodiment of the present invention;

[0054] Figure 3 A framework diagram of a method for autonomous scanning of human organs based on Bayesian optimal search provided by an embodiment of the present invention;

[0055] Figure 4 A schematic diagram of the structure of a human organ autonomous scanning device based on Bayesian optimal search provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0057] The inventors found that in the prior art, due to the significant differences in physical conditions between different human bodies, it is difficult to build a universal human model to effectively plan the ultrasound scanning path. When scanning a certain human organ, it is still necessary to rely on the professional skills and experience of ultrasound doctors to find the best ultrasound probe posture, which not only increases the workload of doctors, but also greatly consumes limited medical resources. Based on this, the inventors made the present invention after further research and development, providing a method and device for autonomous scanning of human organs based on Bayesian optimal search.

[0058] Embodiment 1

[0059] In the first embodiment of the present invention, a method for autonomous scanning of human organs based on Bayesian optimal search is provided. Figure 1 As shown, the method may include the following steps S101-S106:

[0060] S101: Train an initial proxy model based on the acquired initial probe pose dataset and initial ultrasound image scoring dataset.

[0061] S102: Calculate the number of iterations of the artificial fish swarm algorithm and the number of artificial fish according to the data volume of the initial ultrasound image scoring data set.

[0062] S103: Starting from the probe's initial position, the artificial fish school algorithm is used to find the next probe position based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish.

[0063] S104: Control the probe of the ultrasonic robot to move to the next probe posture, and acquire a new ultrasonic image in real time, and update the initial probe posture data set and the initial ultrasonic image scoring data set according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set.

[0064] S105: updating the initial proxy model based on the updated probe pose dataset and the updated ultrasound image scoring dataset to obtain an updated proxy model.

[0065] S106: Determine whether the preset stop condition is met: if so, obtain the optimal probe posture, and control the probe to acquire the optimal ultrasound image of the target organ at the optimal probe posture. If not, re-execute the above steps S102-S105.

[0066] The embodiment of the present invention provides a method for autonomous scanning of human organs based on Bayesian optimal search. The method obtains an initial proxy model through training, and calculates the number of iterations and the number of fish schools of the artificial fish school algorithm according to the data volume of the initial ultrasound image scoring data set, and uses the artificial fish school algorithm as an optimization algorithm to find the optimal next probe posture, control the ultrasound probe to move to the next probe posture, collect new ultrasound images during the movement, update the data set and the proxy model, repeat the above-mentioned Bayesian optimal search algorithm process until the stop condition is met, obtain the optimal probe posture, and control the probe to collect the optimal ultrasound image of the target organ under the optimal probe posture. The method establishes a proxy model through the Bayesian optimal search algorithm, predicts the scores under different probe postures, uses the artificial fish school algorithm as the optimization algorithm of the Bayesian optimal search algorithm, combines the acquisition function to determine the next posture that is most likely to improve the scanning result, guides the ultrasound robot device to approach the optimal probe posture, and then gradually approaches the optimal probe posture by continuously updating the proxy model and optimizing the acquisition function.

[0067] This method can also dynamically adjust the number of artificial fish and the number of iterations of the artificial fish swarm algorithm according to the amount of data in the data set during the iteration round, so that when the amount of data is small and the training time of the proxy model is short, the number of iterations and the number of artificial fish are increased, so that the artificial fish swarm algorithm can perform a deeper search; and when the amount of data increases and the training time of the proxy model is long, the number of iterations and the number of artificial fish are reduced to reduce the search time. This dynamic adjustment mechanism effectively solves the problem of fluctuations in processing time of different iteration rounds in the Bayesian optimal search process, ensuring the stability and efficiency of the Bayesian optimal search.

[0068] This method significantly improves the acquisition quality and efficiency of ultrasound images through intelligent and automated means. This method not only reduces the workload of ultrasound doctors and improves the utilization efficiency of medical resources, but also ensures the consistency and repeatability of examination results by continuously optimizing the search path and image quality, thereby improving the accuracy and reliability of clinical diagnosis.

[0069] In order to facilitate those skilled in the art to understand the present invention, the following is an example diagram of the ultrasonic robot in the embodiment of the present application: Figure 2 As shown, the ultrasonic robot device includes a mechanical arm end and a computer end. The method will be deployed on the computer end to control the movement of the mechanical arm end to autonomously scan the target organs of the human body.

[0070] In the embodiment of the present application, before executing step S101, it is necessary to execute step S107 to obtain an initial probe posture data set and an initial ultrasound image scoring data set.

[0071] Specifically, the probe may be controlled to move along a random path within the scanning area of ​​the target organ, and ultrasound images are collected in real time during the movement, and the probe posture at each collection is recorded, and the initial ultrasound image dataset and the initial probe posture dataset are obtained by sorting the collection time from old to new. Then, a score corresponding to each ultrasound image in the initial ultrasound image dataset is obtained to obtain the initial ultrasound image score dataset.

[0072] For different target organs, the way to obtain the corresponding score based on the ultrasound image will be different. For example, when the target organ is the carotid artery, the ultrasound image needs to determine the score based on the area and length of the carotid artery, while when the target organ is the liver, the ultrasound image may also need to consider the edge clarity of the liver.

[0073] In a specific embodiment, when the target organ is the carotid artery, the score of the ultrasound image can be determined based on the area and length of the carotid artery in the ultrasound image. Specifically, each ultrasound image in the initial ultrasound image data set can be input into a pre-trained image segmentation model to obtain the contour of the carotid artery in the ultrasound image, and then the area and length of the carotid artery in the ultrasound image are calculated based on OpenCV (Open Source Computer Vision Library), and the area and length of the carotid artery are weighted and summed and normalized to obtain the score of the ultrasound image.

[0074] Those skilled in the art can select a suitable neural network for pre-training based on the detailed description of the prior art to obtain a pre-trained image segmentation model. The training process may specifically include:

[0075] In the first step, carotid artery ultrasound images are collected and the contours of the carotid arteries in the carotid artery ultrasound images are marked respectively. After image preprocessing, a carotid artery contour dataset is obtained.

[0076] The second step is to select a suitable neural network model as the initial image segmentation network, such as U-Net model, SegNet model, etc.

[0077] The third step is to divide the carotid artery contour dataset into training set and test set.

[0078] The fourth step is to define the loss function (such as cross entropy loss function, mean square error loss function, etc.), optimization algorithm (such as Adam, SGD, etc.), etc.

[0079] The fifth step is to train the initial image segmentation network using the training set of the carotid artery contour dataset to obtain a trained image segmentation model.

[0080] Repeat the training process of the above image segmentation model until the preset conditions are met, stop training, and obtain a pre-trained image segmentation model. The preset conditions can be set to reach a preset number of iterations, the accuracy reaches a threshold, the accuracy does not change within a preset number of iterations, etc. No specific limitation is made here.

[0081] In an embodiment of the present application, after executing the above step S107 to obtain the initial probe pose data set and the initial ultrasound image scoring data set, it is also necessary to execute step S108 to perform data sparse downsampling on the initial probe pose data set and the initial ultrasound image scoring data set.

[0082] Specifically, all probe poses in the initial probe pose data set are traversed, and the L2 norm between every two adjacent probe poses is calculated. If the L2 norm is smaller than a preset sampling interval, the probe pose that is sorted later in the two adjacent probe poses is deleted, and the scoring data corresponding to the probe pose in the initial ultrasound image scoring data set is deleted.

[0083] In an embodiment of the present application, after executing the above step S107 to obtain the initial probe pose data set and the initial ultrasound image scoring data set, it is also necessary to execute step S109 to perform data time distance filtering on the initial probe pose data set and the initial ultrasound image scoring data set.

[0084] Specifically, the preset maximum number of probe poses and scores that are ranked later in the initial probe pose data set and the initial ultrasound image score data set, i.e., the latest preset maximum number of data, may be retained, and other data that are farther away in time may be deleted. The preset maximum number may be exemplarily set to 200.

[0085] In the above step S101, the initial proxy model is trained according to the acquired initial probe posture dataset and initial ultrasound image scoring dataset.

[0086] Specifically, the initial probe pose dataset is used as training data, and the initial ultrasound image score dataset is used as the corresponding label to train an initial proxy model, so that the initial proxy model can learn the mapping relationship between different probe poses and scores, providing a basis for subsequent optimization search.

[0087] Among them, the initial proxy model can be implemented based on existing technologies such as Gaussian regression, random forest, neural network, etc., which is not limited here.

[0088] In the above step S102, the number of iterations of the artificial fish swarm algorithm and the number of artificial fish are calculated according to the data volume of the initial ultrasound image scoring data set. Specifically, the following steps S1021-S1022 are included:

[0089] S1021: According to the data volume of the initial ultrasound image scoring data set, the number of iterations of the artificial fish swarm algorithm is calculated based on the following formula 1:

[0090]

[0091] Where G is the number of iterations of the artificial fish swarm algorithm, G max is the preset maximum number of increased iterations, G base is the preset iteration base, S p is the data volume of the initial ultrasound image scoring dataset, S max To preset the maximum data volume, is the upward rounding function. max It can be set to 8, G base It can be set to 4 as an example.

[0092] S1022: According to the data volume of the initial ultrasound image scoring data set, the number of artificial fish of the artificial fish swarm algorithm is calculated based on the following formula 2:

[0093]

[0094] Where, F is the number of artificial fish in the artificial fish swarm algorithm, F max The maximum number of artificial fish to be added is F. base is the preset artificial fish base number, S p is the data volume of the initial ultrasound image scoring dataset, S max To preset the maximum data volume, is the upward rounding function. Among them, F max It can be set to 10, F base It can be set to 5 as an example.

[0095] In the embodiment of the present application, during the Bayesian optimal search, the training time of the proxy model is relatively short in the early stage of the iteration due to the small amount of data, and as the search progresses, the amount of data gradually increases in the later stage of the iteration, resulting in a significant increase in the training time of the proxy model. This fluctuation in training time will affect the duration of each iteration of the Bayesian optimal search, and thus affect the stability and efficiency of the algorithm.

[0096] To solve the above problem, the method implements dynamic adjustment through the above step S102, and uses the cosine function in formula 1 and formula 2 to make the number of iterations G of the artificial fish swarm algorithm and the number of artificial fish F increase with the amount of data S p The rise of G decreases slowly, making the number of iterations of the artificial fish swarm algorithm G base and G base +G max The number of artificial fish F is between F base and Fbase +F max Therefore, in the early stage of the Bayesian optimal search iteration, this method will increase the number of iterations of the artificial fish swarm algorithm and the number of artificial fish to extend the search time and make full use of the faster training speed at this time to conduct a deeper search. In the later stage of the Bayesian optimal search iteration, as the amount of data increases and the training time of the proxy model increases, the number of iterations of the artificial fish swarm algorithm and the number of artificial fish can be reduced accordingly to reduce the search time, thereby maintaining the relative stability of each round of iteration time and improving the stability and overall efficiency of the Bayesian optimal search.

[0097] In the above step S103, starting from the probe's initial position, the artificial fish school algorithm is used to find the next probe position based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish. Specifically, the following steps S1031-S1035 are included:

[0098] S1031: Initialize multiple artificial fishes according to the probe's initial posture and the number of artificial fishes, and each artificial fish corresponds to one posture.

[0099] Specifically, the search space may be determined according to the initial position and posture of the probe and a preset maximum step length, and a plurality of candidate points are randomly selected in the search space as the positions and postures corresponding to the plurality of artificial fish.

[0100] S1032: Calculate the sampling potential value of each artificial fish based on the initial proxy model and the acquisition function.

[0101] Specifically, the posture corresponding to each artificial fish is input into the initial proxy model to obtain the predicted score and variance, and the predicted score, variance and the posture corresponding to the artificial fish are input into the acquisition function to obtain the sampling potential value of the artificial fish.

[0102] The acquisition function may be set to expected improvement or upper confidence bound, etc., which is not limited here.

[0103] S1033: Control all artificial fish to execute the flocking behavior, tail-chasing behavior, foraging behavior and random behavior in the artificial fish school algorithm based on the corresponding sampling potential value, iteratively update all artificial fish until the number of iterations of the artificial fish school algorithm is reached, and obtain multiple updated artificial fish.

[0104] Specifically, all the artificial fish may be made to perform the herding behavior first, and then perform the tail-chasing behavior. After performing the herding behavior and the tail-chasing behavior, the sampling potential value of each artificial fish is calculated. If the sampling potential value of no artificial fish exceeds the maximum sampling potential value before the herding behavior and the tail-chasing behavior, that is, no better solution is found, all the artificial fish may continue to perform the foraging behavior. If still no better solution is found, random behavior is performed, thus completing one round of iteration of the artificial fish school algorithm, and re-executing the above-mentioned artificial fish school algorithm until the number of iterations of the artificial fish school algorithm is reached, thereby obtaining multiple updated artificial fish and the corresponding posture of each updated artificial fish.

[0105] Among them, the artificial fish school algorithm includes four basic behaviors: flocking behavior, tail-chasing behavior, foraging behavior and random behavior. Specifically including:

[0106] Swarming behavior: Artificial fish have a tendency to act collectively. Each artificial fish will decide whether to follow other fish based on the location and food concentration of other artificial fish nearby to maintain the concentration of the group and avoid dispersion, which helps to evenly distribute the fish in the entire search space to avoid local optimal traps. The corresponding position of the artificial fish after swarming behavior is calculated based on the following formula 3:

[0107]

[0108] Where, X next is the position of the artificial fish after the swarming behavior, X is the position of the artificial fish before the swarming behavior, and X c is the center position of the neighborhood artificial fish, S f is the preset maximum step size, and U(0,1) represents a random number in the range of 0-1. f It can be set to 0.3 as an example.

[0109] The center position X of the neighborhood artificial fish in the above formula 3 c Calculated by the following formula 4:

[0110]

[0111] Where, X c is the position corresponding to the center position of the neighborhood artificial fish, N f is the number of other artificial fish within the field of view of the artificial fish, is the position of the i-th artificial fish within the field of view of the artificial fish.

[0112] It should be noted that if the above-mentioned clustering behavior is to be performed, the artificial fish must meet the clustering conditions, which are shown in the following formula 5:

[0113]

[0114] Where δ is the crowding factor, Y is the sampling potential value of artificial fish, and N f is the number of other artificial fish within the field of view of the artificial fish, Y c is the position X corresponding to the center position of the neighborhood artificial fish c The corresponding sampling potential value is: Wherein, δ can be set to 0.2 as an example.

[0115] Tail-chasing behavior: In a school of artificial fish, if an artificial fish finds that the sampling potential value of other artificial fish is better, it may follow that artificial fish and use its exploration results to speed up the search speed. The corresponding position of the artificial fish after the tail-chasing behavior is calculated based on the following formula 6:

[0116]

[0117] In the formula, X next is the position of the artificial fish after the rear-end collision, X is the position of the artificial fish before the rear-end collision, and X o is the position of the artificial fish with the highest sampling potential value within the field of view of the artificial fish, S f is the preset maximum step size, and U(0,1) represents a random number in the range of 0-1. f It can be set to 0.3 as an example.

[0118] It should be noted that if the above-mentioned clustering behavior is to be performed, the artificial fish must meet the tail-chasing condition, which is shown in the following formula 7:

[0119]

[0120] Where δ is the crowding factor, Y is the sampling potential value of artificial fish, and N f is the number of other artificial fish within the field of view of the artificial fish, Y o is the highest sampling potential value within the field of view of the artificial fish. Wherein, δ can be set to 0.2 for example.

[0121] Foraging behavior: The artificial fish senses the current food concentration (i.e., the objective function value) within its field of view, and then determines the direction of movement based on the food concentration. If the food concentration ahead is high, it moves in that direction; otherwise, it randomly selects another direction to continue foraging. The corresponding position of the artificial fish after foraging behavior is calculated based on the following formula 8:

[0122]

[0123] In the formula, X next is the position of the artificial fish after foraging, X is the position of the artificial fish before foraging, and X j is the random pose within the visual range of the artificial fish, S fis the preset maximum step size, and U(0,1) represents a random number in the range of 0-1. f It can be set to 0.3 as an example.

[0124] The random pose X of the artificial fish within the visual range in the above formula 8 is j Calculated by the following formula 9:

[0125] X j =X+V f *U(0,1) Formula 9

[0126] Where X is the position of the artificial fish before foraging, X j is the random pose within the visual range of the artificial fish, V f is the visual range of the artificial fish, and U(0,1) represents a random number in the range of 0-1. f It can be set to 0.5 as an example.

[0127] Random behavior: If the above three behaviors do not produce a better solution, the artificial fish can randomly select a direction to explore, increasing the diversity of the search space and avoiding falling into the local optimum. The corresponding position of the artificial fish after random behavior is calculated based on the following formula 10:

[0128] X next =X+V f *U(0,1) Formula 10

[0129] In the formula, X next is the position of the artificial fish after random behavior, X is the position of the artificial fish before random behavior, V f is the visual range of the artificial fish, and U(0,1) represents a random number in the range of 0-1. f It can be set to 0.5 as an example.

[0130] S1034: Calculate the sampling potential value of each updated artificial fish according to the initial agent model and the acquisition function.

[0131] Specifically, the posture corresponding to each updated artificial fish is input into the initial proxy model to obtain the predicted score and variance, and the predicted score, variance and the posture corresponding to the updated artificial fish are input into the acquisition function to obtain the sampling potential value of the updated artificial fish.

[0132] S1035: The posture corresponding to the updated artificial fish with the largest updated sampling potential value is used as the next probe posture.

[0133] In the embodiment of the present application, in the Bayesian optimal search, the optimization algorithm needs to be executed once in each iteration. If the optimization algorithm used is inefficient, the iteration cycle of the Bayesian optimal search algorithm will be significantly extended, which will cause problems such as slow robot movement and disordered movement rhythm, interfere with the normal operation of other programs and greatly reduce the overall scanning efficiency. The gradient ascent method commonly used in the prior art is prone to fall into the local optimal solution. The artificial fish swarm algorithm adopted in this method has a fast convergence speed and strong global search ability. It can effectively explore the search space and avoid falling into the local optimum, and the parallel operation is applicable to various computing environments to improve the solution efficiency. And the algorithm is applicable to multi-objective, continuous and discrete optimization and other problems, showing a high degree of robustness and scalability, and is applicable to the global optimization problems with high real-time requirements in this method.

[0134] In the above step S104, the probe of the ultrasonic robot is controlled to move to the next probe posture, and a new ultrasonic image is acquired in real time, and the initial probe posture data set and the initial ultrasonic image scoring data set are updated according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set. Specifically, the following steps S1041-S1042 are included:

[0135] S1041: Control the probe of the ultrasound robot to move to the next probe posture, and collect multiple new ultrasound images and the probe posture corresponding to each new ultrasound image in real time, and determine the score corresponding to each new ultrasound image.

[0136] The method of determining the corresponding score according to the new ultrasound image has been described in detail in the above step S107 and will not be repeated here.

[0137] S1042: Add the probe poses and scores corresponding to the multiple new ultrasound images to the initial probe pose dataset and the initial ultrasound image score dataset, and perform data sparse downsampling to obtain an updated probe pose dataset and an updated ultrasound image score dataset. Specifically, the following steps S10421-S10423 are included:

[0138] S10421: Calculate the L2 norm of the probe pose corresponding to the last probe pose in the initial probe pose data set and the first new ultrasound image, and determine whether the L2 norm is greater than the preset sampling interval: if so, execute step S10422; if not, execute step S10423.

[0139] The first new ultrasound image refers to the new ultrasound image acquired earliest among the multiple new ultrasound images.

[0140] S10422: Add the probe pose corresponding to the first new ultrasound image to the initial probe pose data set, and add the score corresponding to the first new ultrasound image to the initial ultrasound image score data set.

[0141] S10423: Re-execute the above process of calculating the L2 norm based on the subsequent new ultrasound image until there is no subsequent new ultrasound image, and obtain an updated probe pose data set and an updated ultrasound image score data set.

[0142] In order to facilitate those skilled in the art to understand the present solution, the specific implementation process of S1042 provided in the embodiment of the present application is described more clearly and completely below: The probe postures corresponding to the multiple new ultrasound images are represented as P = {p1, p 2, …,p i ,...,p n}, the corresponding score is E = {e1,e 2, …,e i ,...,e n}, the initial probe pose dataset is represented as P x ={...,p0}, the initial ultrasound image scoring dataset is denoted as E x ={...,e0}. Among them, p i represents the probe posture corresponding to the i-th new ultrasound image, e i represents the score corresponding to the i-th new ultrasound image, p0 represents the last probe pose in the initial probe pose dataset, and e0 represents the last score in the initial ultrasound image scoring dataset. Set the current probe pose p current =p0, and then perform the following calculation:

[0143] d i =‖p i -p current ‖2 for i=1,2,…,n

[0144] if d i >τ P x =P x ∪{p i}E x =E x ∪{e i} current =p i

[0145] Among them, d i is the L2 norm of the two probe postures, and τ is the preset sampling interval, which can be exemplarily set to 0.03.

[0146] In an embodiment of the present application, after executing the above step S104 to obtain the updated probe pose data set and the updated ultrasound image scoring data set, it is also necessary to execute step S109 to perform data time distance filtering on the updated probe pose data set and the updated ultrasound image scoring data set.

[0147] Specifically, the preset maximum number of probe poses and scores in the updated probe pose data set and the updated ultrasound image score data set, i.e., the latest preset maximum number of data, can be retained, and other data with a long time distance can be deleted. The preset maximum number can be exemplarily set to 200.

[0148] The specific algorithm is:

[0149]

[0150] Among them, S max is the preset maximum number, S P is the data volume for updating the probe pose dataset or updating the ultrasound image scoring dataset, and k is the index. max It can be set to 200 as an example.

[0151] In the embodiment of the present application, when the probe collects data, if all data are sampled, the data set will be too large, the proxy model training time will be long, and the scanning efficiency will be reduced. At the same time, the human body will inevitably move and deform due to breathing, and the condition of the target organ may change over time. The longer the time distance, the lower the credibility of the data. Therefore, in the above step S104, this method can effectively reduce the size of the data set, reduce data redundancy and calculation amount, thereby improving processing efficiency, and filter out invalid data with a long time interval to ensure the timeliness and relevance of the data. It can reduce noise interference and improve the accuracy and efficiency of the search.

[0152] In the above step S105, the initial proxy model is updated based on the updated probe pose dataset and the updated ultrasound image scoring dataset to obtain an updated proxy model.

[0153] Specifically, the hyperparameters of the initial proxy model may be re-optimized based on the updated probe pose dataset and the updated ultrasound image scoring dataset to obtain an updated proxy model.

[0154] In the above step S106, it is determined whether the preset stop condition is met: if so, the optimal probe posture is obtained, and the probe is controlled to acquire the optimal ultrasound image of the target organ at the optimal probe posture; if not, the above steps S102-S105 are re-executed.

[0155] The preset stop condition may be that a maximum number of iterations is reached, the score of the ultrasound image reaches a preset score threshold, etc.

[0156] In order to facilitate those skilled in the art to understand the present solution, the specific implementation process of the human organ autonomous scanning method based on Bayesian optimal search provided in the embodiment of the present application is described more clearly and completely below: Figure 3As shown in the figure, the target organ is the carotid artery as an example to illustrate the overall framework of the method:

[0157] The upper left part of the figure shows the process of obtaining the initial ultrasound image scoring dataset and the initial probe posture dataset corresponding to step S107;

[0158] The right side of the figure is a schematic diagram of the steps of the Bayesian optimal search described in steps S102-S106 of the present method, which specifically includes: starting the iterative process of the Bayesian optimal search based on the initial ultrasound image scoring data set and the initial probe pose data set, and the preset stopping condition is whether the maximum number of iterations is reached. If so, the optimal probe pose is obtained, if not, the proxy model is updated, the acquisition function is optimized, the next probe pose is output, the movement of the ultrasound robot is controlled, the image score and correspondence of the acquisition movement process, the probe pose, and the image evaluation score data set and the probe pose data set are updated.

[0159] The lower left corner of the figure is a schematic diagram of controlling the probe to acquire the optimal ultrasound image at the optimal probe posture after the optimal probe posture is obtained.

[0160] Embodiment 2

[0161] Based on the same inventive concept, the embodiment of the present invention also provides a human organ autonomous scanning device based on Bayesian optimal search, referring to Figure 4 As shown, the device comprises:

[0162] A first training module 101 is used to train an initial proxy model according to an acquired initial probe posture dataset and an initial ultrasound image scoring dataset;

[0163] A first calculation module 102 is used to calculate the number of iterations of the artificial fish swarm algorithm and the number of artificial fish according to the data volume of the initial ultrasound image scoring data set;

[0164] A first optimization module 103 is used to start from the probe's initial posture and find the next probe posture by using the artificial fish school algorithm based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish;

[0165] A first updating module 104 is used to control the probe of the ultrasonic robot to move to the next probe posture, acquire a new ultrasonic image in real time, and update the initial probe posture data set and the initial ultrasonic image scoring data set according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set;

[0166] A second updating module 105 is used to update the initial proxy model based on the updated probe pose dataset and the updated ultrasound image scoring dataset to obtain an updated proxy model;

[0167] The first judgment module 106 is used to judge whether the preset stop condition is met: if so, the optimal probe posture is obtained, and the probe is controlled to acquire the optimal ultrasound image of the target organ at the optimal probe posture; if not, the above-mentioned first calculation module is re-executed.

[0168] Embodiment 3

[0169] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the method for autonomous scanning of human organs based on Bayesian optimal search as described in the above-mentioned embodiment 1 is implemented.

[0170] Embodiment 4

[0171] Based on the same inventive concept, an embodiment of the present invention further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the autonomous scanning method of human organs based on Bayesian optimal search as described in the above-mentioned embodiment 1.

[0172] Embodiment 5

[0173] Based on the same inventive concept, an embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory. When the processor executes the computer program, it implements the autonomous scanning method of human organs based on Bayesian optimal search as described in the above embodiment 1.

[0174] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0175] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0176] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0177] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0178] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for autonomous scanning of human organs based on Bayesian optimal search, characterized in that: include: An initial proxy model is trained based on the acquired initial probe pose dataset and initial ultrasound image scoring dataset; Calculating the number of iterations of the artificial fish swarm algorithm and the number of artificial fish according to the data volume of the initial ultrasound image scoring data set; Starting from the probe's initial posture, the artificial fish school algorithm is used to find the next probe posture based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish; Controlling the probe of the ultrasonic robot to move to the next probe posture, and acquiring a new ultrasonic image in real time, and updating the initial probe posture data set and the initial ultrasonic image scoring data set according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set; Update the initial proxy model based on the updated probe pose dataset and the updated ultrasound image scoring dataset to obtain an updated proxy model; The above-mentioned Bayesian optimal search process is re-executed until a preset stop condition is reached to obtain an optimal probe posture, and the probe is controlled to acquire an optimal ultrasound image of the target organ at the optimal probe posture.

2. The method according to claim 1, characterized in that The step of calculating the number of iterations of the artificial fish swarm algorithm and the number of artificial fish according to the data volume of the initial ultrasound image scoring data set comprises: According to the data volume of the initial ultrasound image scoring data set, the number of iterations of the artificial fish swarm algorithm is calculated based on the following formula: Where G is the number of iterations of the artificial fish swarm algorithm, G max is the preset maximum number of increased iterations, G max is the preset iteration base, S p is the data volume of the initial ultrasound image scoring dataset, S max To preset the maximum data volume, is the ceiling rounding function; According to the data volume of the initial ultrasound image scoring data set, the number of artificial fish of the artificial fish swarm algorithm is calculated based on the following formula: Where, F is the number of artificial fish in the artificial fish swarm algorithm, F max The maximum number of artificial fish to be added is preset, F base is the preset artificial fish base number, S p is the data volume of the initial ultrasound image scoring dataset, S max To preset the maximum data volume, is the ceiling function.

3. The method according to claim 1, characterized in that The control of the probe of the ultrasonic robot to move to the next probe posture, and acquiring a new ultrasonic image in real time, and updating the initial probe posture data set and the initial ultrasonic image scoring data set according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set, including: Controlling the probe of the ultrasonic robot to move to the next probe posture, and collecting multiple new ultrasonic images and the probe posture corresponding to each new ultrasonic image in real time, and determining the score corresponding to each new ultrasonic image; The probe poses and scores corresponding to the multiple new ultrasound images are added to the initial probe pose dataset and the initial ultrasound image score dataset respectively, and data sparse downsampling is performed at the same time to obtain an updated probe pose dataset and an updated ultrasound image score dataset.

4. The method according to claim 3, characterized in that The updated probe posture dataset, updated ultrasound image scoring dataset and multiple new ultrasound images are sorted by acquisition time; The step of adding the probe poses and scores corresponding to the multiple new ultrasound images to the initial probe pose dataset and the initial ultrasound image score dataset respectively, and performing data sparse downsampling at the same time to obtain an updated probe pose dataset and an updated ultrasound image score dataset, comprises: Calculate the L2 norm of the probe pose corresponding to the last probe pose in the initial probe pose data set and the first new ultrasound image, and determine whether the L2 norm is greater than a preset sampling interval: If yes, then adding the probe pose corresponding to the first new ultrasound image to the initial probe pose data set, and adding the score corresponding to the first new ultrasound image to the initial ultrasound image score data set; If not, the above process of calculating the L2 norm is re-executed based on the subsequent new ultrasound images until there are no unprocessed new ultrasound images, thereby obtaining an updated probe pose data set and an updated ultrasound image scoring data set.

5. The method according to claim 3, characterized in that: The data in the updated probe posture dataset and the updated ultrasound image scoring dataset are sorted by acquisition time; After obtaining the updated probe pose dataset and the updated ultrasound image scoring dataset, it also includes: A preset maximum number of probe poses and scores that are sorted later in the updated probe pose dataset and the updated ultrasound image score dataset are retained.

6. The method according to claim 1, characterized in that Starting from the probe starting position, based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish, the artificial fish school algorithm is used to find the next probe position, including: Initializing a plurality of artificial fishes according to the starting position of the probe and the number of the artificial fishes; wherein each of the artificial fishes corresponds to a position; Calculating a sampling potential value of each artificial fish according to the initial proxy model and the acquisition function; Controlling all the artificial fish to perform the flocking behavior, tail-chasing behavior, foraging behavior and random behavior in the artificial fish school algorithm based on the corresponding sampling potential values, iteratively updating all the artificial fish until the iteration number of the artificial fish school algorithm is reached, and obtaining a plurality of updated artificial fish; Calculating a sampling potential value of each updated artificial fish according to the initial proxy model and the acquisition function; The posture corresponding to the updated artificial fish with the largest update sampling potential value is used as the next probe posture.

7. A human organ autonomous scanning device based on Bayesian optimal search, characterized in that: include: A first training module is used to train an initial proxy model according to the acquired initial probe posture dataset and initial ultrasound image scoring dataset; A first calculation module, configured to calculate the number of iterations of the artificial fish swarm algorithm and the number of artificial fish according to the data volume of the initial ultrasound image scoring data set; A first optimization module is used to start from the probe's initial posture and find the next probe posture by using the artificial fish school algorithm based on the initial proxy model, the preset acquisition function, the number of iterations of the artificial fish school algorithm and the number of artificial fish; A first updating module is used to control the probe of the ultrasonic robot to move to the next probe posture, and acquire a new ultrasonic image in real time, and update the initial probe posture data set and the initial ultrasonic image scoring data set according to the new ultrasonic image to obtain an updated probe posture data set and an updated ultrasonic image scoring data set; A second updating module is used to update the initial proxy model based on the updated probe pose dataset and the updated ultrasound image scoring dataset to obtain an updated proxy model; The first judgment module is used to judge whether the preset stop condition is met: if so, the optimal probe posture is obtained, and the probe is controlled to acquire the optimal ultrasound image of the target organ at the optimal probe posture; if not, the above-mentioned first calculation module is re-executed.

8. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by a processor, the method for autonomous scanning of human organs based on Bayesian optimal search as described in any one of claims 1 to 6 is implemented.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the method for autonomous scanning of human organs based on Bayesian optimal search as described in any one of claims 1 to 6 is implemented.

10. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the autonomous scanning method of human organs based on Bayesian optimal search as described in any one of claims 1-6.

Images