An ultrasound nodule matching method, system, electronic device, and medium

Abstract

The application discloses an ultrasonic nodule matching method and system, electronic equipment and medium. It relates to the technical field of digital image processing. The method comprises judging whether the ROI image of the frame is processed for the first time; if yes, adding the ROI image of the frame to two image libraries; if no, processing the ROI image of the frame and the two image libraries by using a triple co-occurrence neural network to obtain two confidence sets; judging whether there are confidence values less than a first set threshold in the two confidence sets; if yes, classifying the nodule in the frame image into a target nodule and updating a nodule best cross-section image library and a nodule latest frame image library according to the ROI image of the frame; if no, adding the ROI image of the frame to the nodule best cross-section image library and the nodule latest frame image library. The application can improve the accuracy and robustness of the nodule matching result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital image processing, in particular to an ultrasound nodule matching method and system, an electronic device and a medium. BACKGROUND

[0002] Nodule refers to a lump formed by abnormal aggregation of cells, as a systemic disease, can affect any organ, such as thyroid, breast, liver, kidney, heart and lung, etc. Organ or gland, is a common clinical symptom, the signs and symptoms depend on the invaded organs. Clinically, it can be caused by many causes, such as inflammation, autoimmune reaction, neoplasm, and even cancer, etc. It can be single or multiple.

[0003] The commonly used diagnosis method at present is ultrasonic detection, which uses ultrasonic waves to scan the organs of the patient, and then the doctor makes artificial judgment on the ultrasonic image by naked eye, manually selects 1-2 cross-sectional images of each nodule, and finally manually classifies and includes in the diagnosis report. However, this method has the following disadvantages:

[0004] 1) The diagnosis efficiency is slow, and the doctor needs to select the best section while scanning the patient's affected area.

[0005] 2) When writing the diagnosis report, it is necessary to judge whether it is the same nodule by memory and experience, which is time-consuming and laborious, and easy to make mistakes.

[0006] 3) The workload of the doctor is large, and he needs to concentrate on each patient, which consumes a lot of physical and mental energy.

[0007] 4) The experience level of the doctor is required to be high, and the professional ultrasound doctor needs a long training period and high cost.

[0008] 5) The number of doctors who can meet the requirements is limited, and the medical resources are scarce and the cost is high.

[0009] Therefore, the prior art provides a traditional template matching algorithm, which uses a relatively simple basic algorithm to process the image to obtain the nodule matching result. However, because the algorithm is simple, it is difficult to process complex images such as medical images, resulting in low accuracy and poor robustness of the nodule matching result. SUMMARY

[0010] The purpose of the present application is to provide an ultrasound nodule matching method, system, electronic device and medium, which can improve the accuracy and robustness of the nodule matching result.

[0011] To achieve the above purpose, the present application provides the following solutions:

[0012] An ultrasound nodule matching method, comprising:

[0013] obtaining the ROI image of each frame in the ultrasound video and the nodule detection box coordinates of the ROI image of each frame;

[0014] For any ROI image of a frame, determining whether the ROI image of the frame is processed for the first time according to the ROI image of the frame and the nodule detection box coordinates of the ROI image of the frame to obtain a first determination result;

[0015] If the first determination result is yes, adding the ROI image of the frame to a nodule best cross-section image library and a nodule latest frame image library; the nodule best cross-section image library includes the ROI images of the best cross-sections of all nodules; and the nodule latest frame image library includes the ROI images of the latest frames of all nodules;

[0016] If the first determination result is no, processing the ROI image of the frame, the nodule latest frame image library and the nodule best cross-section image library by using a triple co-occurrence neural network to obtain a first confidence set and a second confidence set; the first confidence set includes the confidence of the ROI image of the frame and each image in the nodule latest frame image library; and the second confidence set includes the confidence of the ROI image of the frame and each image in the nodule best cross-section image library;

[0017] Determining whether there is a confidence less than a first set threshold in the first confidence set and whether there is a confidence less than the first set threshold in the second confidence set to obtain a second determination result;

[0018] If the second determination result is yes, classifying the nodule in the frame image into a target nodule and updating the nodule best cross-section image library and the nodule latest frame image library according to the ROI image of the frame; the target nodule is the nodule corresponding to the minimum confidence in the first confidence set and the second confidence set;

[0019] If the second determination result is no, adding the ROI image of the frame to the nodule best cross-section image library and the nodule latest frame image library.

[0020] Optionally, the triple co-occurrence neural network includes a first convolutional network, a second convolutional network, a third convolutional network, a first Euclidean distance calculation module, a second Euclidean distance calculation module, a first activation function and a second activation function; the weights of the first convolutional network, the second convolutional network and the third convolutional network are shared;

[0021] An output end of the first convolutional network and the second convolutional network is connected with an input end of the first Euclidean distance calculation module; an output end of the first Euclidean distance calculation module is connected with an input end of the first activation function; an output end of the second convolutional network and the third convolutional network is connected with an input end of the second Euclidean distance calculation module; an output end of the second Euclidean distance calculation module is connected with an input end of the second activation function; the first convolutional network is used for inputting the ROI image in the latest frame image library of the nodule; the second convolutional network is used for inputting the ROI image of the frame; and the third convolutional network is used for inputting the ROI image in the best section image library of the nodule.

[0022] Optionally, the updating of the best section image library of the nodule and the latest frame image library of the nodule according to the ROI image of the frame specifically comprises:

[0023] replacing the ROI image of the target nodule in the best section image library of the nodule with the ROI image of the frame;

[0024] replacing the ROI image of the target nodule in the latest frame image library of the nodule with the ROI image of the frame.

[0025] An ultrasonic nodule matching system, comprising:

[0026] an acquisition module, configured to acquire an ROI image of each frame in ultrasonic images and a nodule detection box coordinate of the ROI image of each frame;

[0027] a first judgment module, configured to, for the ROI image of any frame, judge whether the ROI image of the frame is processed for the first time according to the ROI image of the frame and the nodule detection box coordinate of the ROI image of the frame, to obtain a first judgment result;

[0028] a first update library module, configured to, if the first judgment result is yes, add the ROI image of the frame to a best section image library of a nodule and a latest frame image library of the nodule; the best section image library of the nodule comprises ROI images of best sections of all nodules; and the latest frame image library of the nodule comprises ROI images of latest frames of all nodules;

[0029] a confidence calculation module, configured to, if the first judgment result is no, process the ROI image of the frame, the latest frame image library of the nodule and the best section image library of the nodule by using a triple co-occurrence neural network to obtain a first confidence set and a second confidence set; the first confidence set comprises confidence degrees of the ROI image of the frame and each image in the latest frame image library of the nodule; and the second confidence set comprises confidence degrees of the ROI image of the frame and each image in the best section image library of the nodule;

[0030] a second determining module, configured to determine whether there is a confidence degree less than a first set threshold in the first confidence degree set and whether there is a confidence degree less than the first set threshold in the second confidence degree set, to obtain a second determination result;

[0031] a matching module, configured to, if the second determination result is yes, classify a nodule in the frame image into a target nodule and update a nodule best cross-section image library and a nodule latest frame image library according to the ROI image of the frame; the target nodule is a nodule corresponding to a minimum confidence degree in the first confidence degree set and the second confidence degree set;

[0032] a second updating library module, configured to, if the second determination result is no, add the ROI image of the frame to the nodule best cross-section image library and the nodule latest frame image library.

[0033] Optionally, the triplet co-occurrence neural network comprises a first convolutional network, a second convolutional network, a third convolutional network, a first Euclidean distance calculation module, a second Euclidean distance calculation module, a first activation function and a second activation function; weights of the first convolutional network, the second convolutional network and the third convolutional network are shared.

[0034] outputs of the first convolutional network and the second convolutional network are connected with an input end of the first Euclidean distance calculation module; an output end of the first Euclidean distance calculation module is connected with an input end of the first activation function; outputs of the second convolutional network and the third convolutional network are connected with an input end of the second Euclidean distance calculation module; an output end of the second Euclidean distance calculation module is connected with an input end of the second activation function; the first convolutional network is configured to input an ROI image in the nodule latest frame image library, the second convolutional network is configured to input the ROI image of the frame, and the third convolutional network is configured to input an ROI image in the nodule best cross-section image library.

[0035] Optionally, the matching module specifically comprises:

[0036] a nodule best cross-section image library updating unit, configured to replace an ROI image of the target nodule in the nodule best cross-section image library with the ROI image of the frame;

[0037] a nodule latest frame image library updating unit, configured to replace an ROI image of the target nodule in the nodule latest frame image library with the ROI image of the frame.

[0038] An electronic device comprises:

[0039] A memory for storing a computer program and a processor for running the computer program to cause the electronic device to perform the ultrasound nodule matching method according to the above.

[0040] A computer readable storage medium storing a computer program, which, when executed by a processor, implements the ultrasound nodule matching method described above.

[0041] According to the specific embodiments of the present application, the following technical effects are disclosed: Compared with the twin network, the triple symbiotic neural network has an additional template comparison branch, and compared with the pairwise comparison of the twin network, the template comparison can provide a stable reference to improve the accuracy and robustness of the nodule matching result. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0043] Figure 1 The ultrasound nodule matching method flow chart provided for the embodiments of the present application;

[0044] Figure 2 The ultrasound nodule matching method principle diagram provided for the embodiments of the present application;

[0045] Figure 3 The specific flow chart for using the triple symbiotic neural network to analyze and compare the ROI image of the current frame and the latest ROI of each nodule in the historical record one by one, and classifying the comparison results after obtaining the comparison results;

[0046] Figure 4 The black border filling result image;

[0047] Figure 5 The flow chart for processing using the triple symbiotic neural network. DETAILED DESCRIPTION

[0048] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0049] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0050] The present application provides an ultrasound nodule matching method, as shown in the embodiment, the method comprises: Figure 1

[0051] Step 101: obtaining the ROI image of each frame in the ultrasound image and the nodule detection box coordinates of the ROI image of each frame.

[0052] Step 102: for the ROI image of any frame, judging whether the ROI image of the frame is processed for the first time according to the ROI image of the frame and the nodule detection box coordinates of the ROI image of the frame, and obtaining a first judgment result.

[0053] Step 103: if the first judgment result is yes, adding the ROI image of the frame to the nodule best cross-section image library and the nodule latest frame image library; the nodule best cross-section image library includes the ROI image of the best cross-section of all nodules; the nodule latest frame image library includes the ROI image of the latest frame of all nodules.

[0054] Step 104: if the first judgment result is no, processing the ROI image of the frame, the nodule latest frame image library and the nodule best cross-section image library by using a triple co-occurrence neural network to obtain a first confidence set and a second confidence set; the first confidence set includes the confidence of the ROI image of the frame and each image in the nodule latest frame image library; the second confidence set includes the confidence of the ROI image of the frame and each image in the nodule best cross-section image library.

[0055] Step 105: judging whether there is a confidence less than a first set threshold in the first confidence set and whether there is a confidence less than the first set threshold in the second confidence set, and obtaining a second judgment result.

[0056] Step 106: if the second judgment result is yes, classifying the nodule in the frame image into a target nodule and updating the nodule best cross-section image library and the nodule latest frame image library according to the ROI image of the frame; the target nodule is the nodule corresponding to the minimum confidence in the first confidence set and the second confidence set.

[0057] Step 107: if the second judgment result is no, adding the ROI image of the frame to the nodule best cross-section image library and the nodule latest frame image library.

[0058] ​In practical applications, the three-way symbiotic neural network includes: a first convolutional network, a second convolutional network, a third convolutional network, a first Euclidean distance calculation module, a second Euclidean distance calculation module, a first activation function, and a second activation function; the weights of the first convolutional network, the second convolutional network, and the third convolutional network are shared.

[0059] The outputs of the first and second convolutional networks are connected to the input of the first Euclidean distance calculation module; the output of the first Euclidean distance calculation module is connected to the input of the first activation function; the outputs of the second and third convolutional networks are connected to the input of the second Euclidean distance calculation module; the output of the second Euclidean distance calculation module is connected to the input of the second activation function; the first convolutional network is used to input the ROI image from the latest frame image library of the nodule, the second convolutional network is used to input the ROI image of the frame, and the third convolutional network is used to input the ROI image from the best cross-section image library of the nodule.

[0060] In practical applications, updating the nodule optimal cross-sectional image library and the latest nodule frame image library based on the ROI image of the frame specifically includes:

[0061] Replace the ROI image of the target nodule in the nodule best section image library with the ROI image of the frame.

[0062] Replace the ROI image of the target nodule in the latest frame image library of the nodule with the ROI image of the frame.

[0063] This invention provides a more specific embodiment to describe the above method in detail: such as Figure 2 As shown, this embodiment of the invention analyzes ultrasound images based on the output of the ultrasound imaging nodule dynamic localization software, using the best cross-section of the nodule in the image as a matching template. Based on a triplet network, it analyzes and categorizes nodules appearing in different images. Images of the same nodule with different cross-sections in different images are categorized together, while cross-section images of different nodules are categorized separately. Specifically, this includes:

[0064] (1) The ultrasound scan images are transmitted to a computer with ultrasound imaging nodule dynamic positioning software installed via a data acquisition card.

[0065] (2) The ultrasound imaging nodule dynamic positioning software analyzes each frame of the ultrasound image, and saves the coordinates of the nodule detection frame and the corresponding frame image, along with the best section of each nodule in the historical record.

[0066] (3) Using triple symbiotic neural network, taking the ROI of the best cross-section of each nodule in the history record as a template, the ROI image of the current frame and the ROI of the latest frame of each nodule in the history record are analyzed and compared one by one, and the comparison results are classified.

[0067] In practical application, as shown in Figure 3 , using triple symbiotic neural network, taking the ROI of the best cross-section of each nodule in the history record as a template, the ROI image of the current frame and the ROI of the latest frame of each nodule in the history record are analyzed and compared one by one, and the comparison results are classified. Specifically, it includes:

[0068] (1) After obtaining the nodule detection box coordinates of the current frame and the image of the corresponding frame, the first step is to determine whether it is the first input. If it is the first input, a new nodule is added and classified as No. 1, and the information is saved. If not, it will enter the next step of analysis and comparison.

[0069] (2) According to the nodule detection box coordinates of the current frame and the image of the corresponding frame, the ROI is cropped, and the size is adjusted to 64*64*3 by black edge padding method under the premise of maintaining the original ROI aspect ratio, as shown in Figure 4 .

[0070] (3) Then the processed ROI image of the latest frame (the ROI of the latest frame of each nodule in the history record) Figure 5 is ROI(p), the ROI image of the current frame, Figure 5 is ROI(n), the best cross-section ROI image (the ROI of the best cross-section of each nodule in the history record), Figure 5 is ROI(t), which is compared one by one through triple symbiotic neural network, and the network structure is as shown in Figure 5 .

[0071] The input is respectively ROI(p), ROI(n) and ROI(t), which is respectively obtained through 3 EfficientNetV2B0 (convolutional network) sharing weights to obtain 3 feature vectors P, N and T. Then, according to the formula , the Euclidean distances d(P, N) and d(N, T) of P and N, and N and T are respectively calculated, wherein x i represents the i-th element of the input X vector, y i represents the i-th element of the input Y vector, when calculating d(P, N) x i represents the i-th element of the P vector, y i represents the i-th element of the input N vector, when calculating d(N, T) x idenotes the i-th element of the N vector, y i denotes the i-th element of the T vector.

[0072] (4) The triplet co-occurrence neural network inputs d(P, N) and d(N, T) into x in the Sigmoid function to obtain two confidences. The higher the confidence, the greater the difference in comparison; the lower the confidence, the smaller the difference in comparison, and the higher the similarity. The formula of the Sigmoid function is

[0073] (5) If there are two confidences less than 0.5 in the comparison results of the current frame nodule and all nodules, the ROI image of the current frame is classified into the nodule with the minimum confidence; if there is no confidence less than 0.5 in the comparison results, a new nodule is created, that is, the ROI image of the current frame is classified as the n+1th nodule.

[0074] The embodiment of the application also provides an ultrasound nodule matching system corresponding to the above method, comprising:

[0075] The acquisition module is configured to acquire the ROI image of each frame in the ultrasound image and the nodule detection box coordinates of the ROI image of each frame.

[0076] The first judgment module is configured to, for any ROI image of a frame, judge whether the ROI image of the frame is processed for the first time according to the ROI image of the frame and the nodule detection box coordinates of the ROI image of the frame, and obtain a first judgment result.

[0077] The first update library module is configured to, if the first judgment result is yes, add the ROI image of the frame to a nodule best cross-sectional image library and a nodule latest frame image library; the nodule best cross-sectional image library includes the ROI image of the best cross section of all nodules; and the nodule latest frame image library includes the ROI image of the latest frame of all nodules.

[0078] The confidence calculation module is configured to, if the first judgment result is no, process the ROI image of the frame, the nodule latest frame image library and the nodule best cross-sectional image library by using a triplet co-occurrence neural network to obtain a first confidence set and a second confidence set; the first confidence set includes the confidence of the ROI image of the frame and each image in the nodule latest frame image library; and the second confidence set includes the confidence of the ROI image of the frame and each image in the nodule best cross-sectional image library.

[0079] The second judgment module is configured to judge whether there is a confidence less than a first set threshold in the first confidence set and whether there is a confidence less than the first set threshold in the second confidence set, and obtain a second judgment result.

[0080] The matching module is configured to, if the second determination result is yes, classify the nodule in the frame image into a target nodule, and update a nodule best cross-section image library and a nodule latest frame image library according to the ROI image of the frame; the target nodule is a nodule corresponding to the minimum confidence in the first confidence set and the second confidence set.

[0081] The second updating library module is configured to, if the second determination result is no, add the ROI image of the frame to the nodule best cross-section image library and the nodule latest frame image library.

[0082] In actual application, the triplet co-occurrence neural network comprises a first convolutional network, a second convolutional network, a third convolutional network, a first Euclidean distance calculation module, a second Euclidean distance calculation module, a first activation function and a second activation function; the first convolutional network, the second convolutional network and the third convolutional network share weights.

[0083] The output ends of the first convolutional network and the second convolutional network are connected with the input end of the first Euclidean distance calculation module; the output end of the first Euclidean distance calculation module is connected with the input end of the first activation function; the output ends of the second convolutional network and the third convolutional network are connected with the input end of the second Euclidean distance calculation module; the output end of the second Euclidean distance calculation module is connected with the input end of the second activation function; the first convolutional network is configured to input the ROI image in the nodule latest frame image library, the second convolutional network is configured to input the ROI image of the frame, and the third convolutional network is configured to input the ROI image in the nodule best cross-section image library.

[0084] In actual application, the matching module specifically comprises:

[0085] The nodule best cross-section image library updating unit is configured to replace the ROI image of the target nodule in the nodule best cross-section image library with the ROI image of the frame.

[0086] The nodule latest frame image library updating unit is configured to replace the ROI image of the target nodule in the nodule latest frame image library with the ROI image of the frame.

[0087] The embodiment of the present application further provides an electronic device comprising:

[0088] The memory is configured to store a computer program, and the processor is configured to run the computer program to enable the electronic device to execute the ultrasound nodule matching method.

[0089] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the ultrasound nodule matching method.

[0090] Compared with the prior art, the present application has the following advantages:

[0091] The present application is based on a triple symbiotic neural network, analyzes and classifies the nodules appearing on different pictures, classifies the images of different sections of the same nodule on different pictures together, and classifies the section images of different nodules separately, so as to quickly analyze the nodules and include them in the report, greatly improve the diagnosis and treatment efficiency, and reduce the pressure of doctors.

[0092] The present application can maintain the original nodule proportion, avoid distortion, and more effectively extract the similarity features of the images by adjusting the size through the black edge filling method.

[0093] The shared weights of the triple symbiotic neural network can ensure that the vector features of the two images extracted by the network are kept in a unified standard, and the triple symbiotic neural network has one more template comparison branch than the twin network, which provides a stable reference and higher comparison accuracy compared with the pairwise comparison of the twin network.

[0094] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts of each embodiment can be referred to each other.

[0095] The principles and implementation modes of the present application are described by applying specific examples in this paper, and the above embodiment description is only used to help understand the method of the present application and its core idea; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for ultrasound nodule matching, characterized in that, include: Obtain the ROI image of each frame in the ultrasound image and the coordinates of the nodule detection box in each frame of the ROI image; For any frame of ROI image, determine whether the ROI image of the frame is being processed for the first time based on the ROI image of the frame and the coordinates of the nodule detection box of the ROI image of the frame, and obtain the first determination result; If the first determination result is yes, then the ROI image of the frame is added to the nodule optimal section image library and the nodule latest frame image library; the nodule optimal section image library includes the ROI images of the optimal sections of all nodules; the nodule latest frame image library includes the ROI images of the latest frames of all nodules; If the first judgment result is negative, then a three-linked symbiotic neural network is used to process the ROI image of the frame, the latest frame image library of nodules, and the best cross-section image library of nodules to obtain a first confidence set and a second confidence set; the first confidence set includes the confidence of the ROI image of the frame and each image in the latest frame image library of nodules; The second confidence set includes the confidence scores of the ROI images of the frame and each image in the nodule best section image library; The three-way symbiotic neural network includes: a first convolutional network, a second convolutional network, a third convolutional network, a first Euclidean distance calculation module, a second Euclidean distance calculation module, a first activation function, and a second activation function; the weights of the first convolutional network, the second convolutional network, and the third convolutional network are shared; The outputs of the first and second convolutional networks are connected to the input of the first Euclidean distance calculation module; the output of the first Euclidean distance calculation module is connected to the input of the first activation function; the outputs of the second and third convolutional networks are connected to the input of the second Euclidean distance calculation module; the output of the second Euclidean distance calculation module is connected to the input of the second activation function; the first convolutional network is used to input the ROI image from the latest frame image library of the nodule, the second convolutional network is used to input the ROI image of the frame, and the third convolutional network is used to input the ROI image from the best cross-section image library of the nodule; Determine whether there is a confidence level in the first confidence level set that is less than a first set threshold, and whether there is a confidence level in the second confidence level set that is less than the first set threshold, to obtain a second determination result; If the second determination result is yes, then the nodules in the frame image are classified into target nodules, and the optimal cross-sectional image library of nodules and the latest frame image library of nodules are updated according to the ROI image of the frame; the target nodule is the nodule corresponding to the minimum confidence in the first confidence set and the second confidence set; If the second determination result is negative, then the ROI image of the frame is added to the nodule best section image library and the nodule latest frame image library.

2. The ultrasound nodule matching method according to claim 1, characterized in that, The step of updating the nodule optimal cross-sectional image library and the latest nodule frame image library based on the ROI image of the frame specifically includes: Replace the ROI image of the target nodule in the nodule best section image library with the ROI image of the frame; Replace the ROI image of the target nodule in the latest frame image library of the nodule with the ROI image of the frame.

3. An ultrasound nodule matching system, characterized in that, include: The acquisition module is used to acquire the ROI image of each frame in the ultrasound image and the coordinates of the nodule detection box of each frame of the ROI image. The first judgment module is used to determine whether the ROI image of any frame is being processed for the first time based on the ROI image of the frame and the coordinates of the nodule detection box of the ROI image of the frame, and to obtain a first judgment result. The first update library module is used to add the ROI image of the frame to the nodule best section image library and the nodule latest frame image library if the first determination result is yes; the nodule best section image library includes the ROI images of the best sections of all nodules; the nodule latest frame image library includes the ROI images of the latest frames of all nodules. The confidence calculation module is used to process the ROI image of the frame, the latest frame image library of nodules, and the best cross-section image library of nodules using a three-linked symbiotic neural network if the first judgment result is negative, to obtain a first confidence set and a second confidence set; the first confidence set includes the confidence of the ROI image of the frame and each image in the latest frame image library of nodules. The second confidence set includes the confidence scores of the ROI images of the frame and each image in the nodule best section image library; The three-way symbiotic neural network includes: a first convolutional network, a second convolutional network, a third convolutional network, a first Euclidean distance calculation module, a second Euclidean distance calculation module, a first activation function, and a second activation function; the weights of the first convolutional network, the second convolutional network, and the third convolutional network are shared; The outputs of the first and second convolutional networks are connected to the input of the first Euclidean distance calculation module; the output of the first Euclidean distance calculation module is connected to the input of the first activation function; the outputs of the second and third convolutional networks are connected to the input of the second Euclidean distance calculation module; the output of the second Euclidean distance calculation module is connected to the input of the second activation function; the first convolutional network is used to input the ROI image from the latest frame image library of the nodule, the second convolutional network is used to input the ROI image of the frame, and the third convolutional network is used to input the ROI image from the best cross-section image library of the nodule; The second judgment module is used to determine whether there is a confidence level less than the first set of confidence levels in the first confidence level set, and whether there is a confidence level less than the first set of confidence levels in the second confidence level set, and to obtain a second judgment result; The matching module is used to classify the nodules in the frame image into target nodules and update the best cross-sectional image library of nodules and the latest frame image library of nodules according to the ROI image of the frame if the second judgment result is yes; the target nodule is the nodule corresponding to the minimum confidence in the first confidence set and the second confidence set. The second update library module is used to add the ROI image of the frame to the nodule best section image library and the nodule latest frame image library if the second judgment result is negative.

4. The ultrasound nodule matching system according to claim 3, characterized in that, The matching module specifically includes: The nodule optimal section image library update unit is used to replace the ROI image of the target nodule in the nodule optimal section image library with the ROI image of the frame; The nodule latest frame image library update unit is used to replace the ROI image of the target nodule in the nodule latest frame image library with the ROI image of the frame.

5. An electronic device, characterized in that, include: A memory and a processor, the memory for storing a computer program, the processor for running the computer program to cause the electronic device to perform the ultrasound nodule matching method according to any one of claims 1 to 2.

6. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the ultrasound nodule matching method as described in any one of claims 1 to 2.

Images