A medical ultrasonic image analysis system and method

Abstract

The present invention discloses a medical ultrasonic image analysis system and method, which relates to the technical field of medical image processing. The composition of the system includes: an ultrasonic image acquisition module, an image feature extraction module, a Bayesian network diagnosis model construction module, a probability inference calculation module, and a diagnosis result output module. The medical ultrasonic image analysis system of the present invention can comprehensively consider various features in the image, as well as the personal information and medical history of the patient by constructing a Bayesian network diagnosis model, so as to achieve a comprehensive and accurate diagnosis of diseases. It not only improves the accuracy and reliability of the diagnosis, but also greatly reduces the influence of the doctor's subjective judgment on the diagnosis result and reduces the risk of misdiagnosis and missed diagnosis. At the same time, due to the powerful inference ability and self-learning ability of the Bayesian network, this system can continuously optimize and improve the diagnosis model with the accumulation of clinical data, improving the accuracy and efficiency of the diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and specifically provides a medical ultrasonic image analysis system and method. Background Art

[0002] As one of the important means of modern medical diagnosis, medical ultrasonic imaging technology is widely used in clinical practice due to its non-invasive, real-time, low-cost and other characteristics. In the field of medical image analysis, especially when diagnosing ultrasonic images, doctors often need to comprehensively judge the type and severity of diseases based on various features in the images, as well as the patient's personal information and medical history. Therefore, it is particularly important to develop a system that can automatically and intelligently analyze medical ultrasonic images and accurately diagnose diseases.

[0003] However, some current medical ultrasonic image analysis systems using machine learning and artificial intelligence technologies have limitations in processing complex and variable ultrasonic images. They rely only on single image features for diagnosis, ignoring the correlation and complementarity between different features, resulting in insufficient accuracy and reliability of the diagnosis results. Moreover, the influence of patient information is not fully considered when constructing the diagnostic model, making the diagnosis results deviate from the actual situation, unable to provide comprehensive diagnostic reference information for doctors, and increasing the difficulty and risk of diagnostic decisions.

[0004] In summary, the existing technologies have limitations in the automatic analysis and uncertainty quantification of ultrasonic images. Therefore, it is necessary to develop a diagnostic model based on Bayesian network to provide a more comprehensive and accurate ultrasonic image analysis solution to meet the trend of modern medical development. Summary of the Invention

[0005] The purpose of the present invention is to make up for the deficiencies of the existing technologies, and provides a medical ultrasonic image analysis system and method. It can effectively integrate ultrasonic image features and patient information by constructing a diagnostic model based on Bayesian network, provide a more accurate probability of disease occurrence, and use confidence intervals and entropy values to evaluate the reliability of the results, so as to provide more comprehensive diagnostic reference information for doctors. This ability to quantify uncertainty enables doctors to better assess risks during the diagnosis process and make more reasonable treatment decisions.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: On the one hand, a medical ultrasonic image analysis system, the composition of which includes: an ultrasonic image acquisition module, an image feature extraction module, a Bayesian network diagnostic model construction module, a probability inference calculation module, and a diagnostic result output module;

[0007] The ultrasonic image acquisition module is connected to a medical ultrasonic imaging device, and is used to acquire medical ultrasonic images and preprocess the images to optimize the influence of interference factors on diagnosis;

[0008] The image feature extraction module extracts various features from the acquired ultrasonic images and uses the extracted features as input information for the Bayesian network diagnosis model. The features include edge gradient features, regional contrast features, and spectral features, where:

[0009] The extraction process of the edge gradient feature is as follows: Through the gray value at the pixel point in the image, use the Sobel operator to calculate the horizontal gradient and the vertical gradient , that is, the horizontal gradient operator is , the vertical gradient operator is , then , the edge gradient amplitude , for the entire image, the total number of image pixel points is , then the edge gradient feature value , where M and N are the number of rows and columns of the image, and this value is used to describe the overall change degree information of the tissue boundary in the image and input it into the Bayesian network diagnosis model;

[0010] The extraction process of the regional contrast feature is as follows: Divide the image into non-overlapping sub-regions , where , , set the average gray value of the sub-region as , the average gray value of the entire image is , then the contrast value of the sub-region and the entire image, the regional contrast feature , where is the weight of the sub-region , and this feature is used to reflect the gray difference situation between different regions and input it into the Bayesian network diagnosis model;

[0011] The extraction process of the spectral feature is as follows: Perform a two-dimensional discrete Fourier transform 2D-DFT on the ultrasonic image, that is , where and are the number of rows and columns of the image respectively, , , the spectral feature includes the peak frequency , that is, the frequency coordinate corresponding to the maximum spectral amplitude , and the bandwidth​ , the corresponding spectral energy function is , calculate the total energy , accumulate the energy starting from the low frequency. When the accumulated energy reaches , the corresponding frequency range is the bandwidth , this spectral feature is used to analyze the relationship between the frequency characteristics of the ultrasonic reflection signal and the tissue characteristics and is used as the input of the model. Denote the spectral feature value as ;

[0012] The Bayesian network diagnosis model construction module is used to construct a Bayesian network diagnosis model, determine the nodes of the network and the probability relationship between the nodes, set the extracted image features and patient information as network nodes, and determine the conditional probability distribution between the nodes according to the clinical data to obtain the disease nodes, thereby constructing a complete Bayesian network structure. At the same time, an uncertainty quantification index, namely the confidence interval and entropy value, is introduced into the model to quantify the uncertainty degree of the probability result. Among them, the specific construction process of the Bayesian network diagnosis model is as follows:

[0013] Determine the node type: According to the edge gradient feature value , regional contrast feature , spectral feature value , and the age and gender information of the patient, respectively as the image feature node and the patient information node, to obtain the basic node set in the Bayesian network as , where the image feature nodes include the edge gradient feature node , regional contrast feature node , spectral feature node , the patient information nodes include the age node , gender node ;

[0014] Determine the probability relationship between the nodes: Collect clinical case data, including the ultrasonic image data of the patient, the corresponding calculation results of the image features, the age and gender information of the patient, and the finally diagnosed disease type. Determine the probability distribution between the nodes through the conditional probability distribution, that is, for the disease node and the basic node set , calculate the conditional probability distribution , that is, consider the disease and a combination of a set of feature values and patient information values , count the number of times the disease and this feature combination appear simultaneously in the dataset , and the number of times this feature combination appears alone , then the conditional probability , all the image feature nodes and patient information nodes in the basic node set are associated with the disease nodes through this conditional probability method ;

[0015] Construct a Bayesian network structure: According to the determined nodes and the probability relationships between the nodes, construct a Bayesian network structure. Each node in the network represents a random variable, and the directed edges between the nodes represent the dependence relationships between the variables;

[0016] Introduce uncertainty quantification indicators: Introduce uncertainty quantification indicators in the Bayesian network diagnosis model, including confidence intervals and entropy values, to quantify the uncertainty degree of the probability results;

[0017] The probability inference calculation module inputs the feature data obtained by the image feature extraction module into the Bayesian network diagnosis model, calculates the occurrence probability of the disease through the Bayesian network structure. During the calculation process, the values of the uncertainty quantification indicators are updated in real time. As new data is input and probability inference proceeds, the indicators of the confidence interval and entropy value are dynamically adjusted;

[0018] The diagnosis result output module presents the occurrence probability of the disease obtained by the probability inference calculation module to the doctor in the form of a probability value list, and at the same time provides corresponding diagnostic suggestions to assist the doctor in making a final diagnostic decision.

[0019] Furthermore, during the process of constructing the Bayesian network diagnosis model by the Bayesian network diagnosis model construction module, for the edge gradient feature node of the image feature node and the disease node The association determination process is as follows: Perform edge gradient feature extraction on the clinical ultrasound image samples to obtain the edge gradient feature value set of each sample image , where is the edge gradient feature value of the th sample image. For each disease type , statistically analyze the distribution of the edge gradient feature values in the sample images where the disease exists, that is, the edge gradient feature value subset corresponding to the disease is , calculate the conditional probability distribution between the disease and the edge gradient feature node as: , where represents the number of samples with the disease and the edge gradient feature value of , represents the total number of samples with the edge gradient feature value of .

[0020] Furthermore, in the process of constructing the Bayesian network diagnosis model, for the regional contrast feature node of the image feature node and the disease node the association determination process is as follows: Calculate the regional contrast feature value for the clinical ultrasound image sample to obtain the regional contrast feature value set For the disease determine its corresponding regional contrast feature value subset as Calculate the conditional probability distribution of the disease and the regional contrast feature node as: wherein, represents the number of samples with the disease and the regional contrast feature value of , represents the total number of samples with the regional contrast feature value of .

[0021] Furthermore, in the process of constructing the Bayesian network diagnosis model, for the spectral feature node of the image feature node and the disease node the association determination process is as follows: Obtain the spectral feature value from the spectral analysis of the clinical ultrasound image sample, including the peak frequency and the bandwidth to form the spectral feature value set For the disease determine its corresponding spectral feature value subset Calculate the conditional probability distribution of the disease and the spectral feature node as: where the count function is used to count the number of samples of the combination of the disease and the spectral feature value.

[0022] Furthermore, in the process of constructing the Bayesian network diagnosis model, for the age node of the patient information node and the disease node the association determination process is as follows: Collect the patient age information and disease diagnosis results in the clinical case data, divide the age into different intervals, and count the incidence number of each disease in each age interval and the total number of samples in each age interval, and calculate the disease and the age node The conditional probability distribution is as follows: , where represents the th interval in the age range.

[0023] Furthermore, in the process of constructing the Bayesian network diagnosis model by the Bayesian network diagnosis model construction module, for the gender node of the patient information node and the disease node , the process of determining the association is as follows: Count the number of occurrences of each disease among male and female patients in the clinical case data and , as well as the total sample sizes of male and female patients and , calculate the conditional probability distribution of the disease and the gender node as:

[0024] .

[0025] Furthermore, the Bayesian network diagnosis model construction module introduces quantification indicators of uncertainty confidence intervals and entropy values in the process of constructing the Bayesian network diagnosis model, where:

[0026] For the confidence interval: For the disease occurrence probability , calculate the sample mean and the sample standard deviation , and use the unbiased estimation formula in the normal approximation method , where is the estimated value of the disease occurrence probability of the rd sample, is the sample size, and at the confidence level of , obtain the quantile by looking up the standard normal distribution table, then the confidence interval is , and this confidence interval can intuitively show the range of the estimated disease occurrence probability value;

[0027] For the entropy value: According to the disease occurrence probability distribution , is the occurrence probability of the disease , calculate the entropy value as: , where Let \(N\) be the total number of disease types. A large entropy value indicates a dispersed probability distribution and high uncertainty; a small entropy value indicates that the probability is concentrated on one of the disease types and low uncertainty. The entropy value is used to quantify the uncertainty of the disease occurrence probability result, providing more comprehensive diagnostic reference information for doctors to help them evaluate the credibility of the diagnostic results.

[0028] Furthermore, the probability inference calculation module adopts the Bayesian inference formula , and dynamically adjusts the indicators of the confidence interval and the entropy value, where is the disease node, is the evidence node set, that is, the image feature and patient information nodes, is the probability of the evidence appearing under the condition that the disease occurs, is the prior probability of the disease , is the posterior probability of the disease under the evidence . During the iterative calculation process, after each new data input, the parameters of and are updated according to the structure and conditional probability distribution of the Bayesian network, and the posterior probability is recalculated. At the same time, the confidence interval and the entropy value are updated, that is, when new image feature data and patient information are added, the new conditional probability distribution is calculated according to the new evidence combination, and then the estimated value of the disease occurrence probability and its confidence interval and entropy value are updated. Set the new evidence node to be added, calculate , update to be , and then recalculate . For the confidence interval update, and are recalculated according to the new sample data, and then the new confidence interval is obtained. The entropy value update is composed of the new posterior probability to form a new probability distribution , and is recalculated.

[0029] On the other hand, a medical ultrasonic image analysis method, the specific steps of which are as follows:

[0030] S1. Start the ultrasonic image acquisition module, establish a connection with the ultrasonic imaging device, receive and store the collected ultrasonic image data in real time, and record the metadata of the acquisition time and patient information;

[0031] S2. Transmit the collected ultrasonic images to the image feature extraction module, extract edge gradient features, regional contrast features, and spectral features from the images, store the processed feature data in the feature database, and establish an associated index with the patient information;

[0032] S3. Construct a Bayesian network diagnosis model, determine the node types and quantities of the Bayesian network, collect clinical case data, calculate the conditional probability distribution between nodes based on the clinical data, and integrate a calculation model of uncertainty quantification indicators in the model so that the model can synchronously output uncertainty quantification results while calculating the disease occurrence probability;

[0033] S4. Read the image feature data and basic patient information of the current patient from the feature database, input the data into the constructed Bayesian network diagnosis model, start the probability inference calculation process, perform iterative calculations, and output the disease occurrence probability results and corresponding uncertainty quantification values;

[0034] S5. Transmit the calculated disease occurrence probabilities and their uncertainty quantification information to the diagnosis result output module to generate a visual diagnosis result report.

[0035] Compared with the prior art, the medical ultrasonic image analysis system and method have the following beneficial effects:

[0036] First, by constructing a Bayesian network diagnosis model, the medical ultrasonic image analysis system of the present invention can comprehensively consider various features in the image, as well as the personal information and medical history of the patient, so as to achieve a comprehensive and accurate diagnosis of diseases. It not only improves the accuracy and reliability of the diagnosis, but also greatly reduces the influence of doctors' subjective judgments on the diagnosis results, reduces the risks of misdiagnosis and missed diagnosis. At the same time, due to the powerful reasoning ability and self-learning ability of the Bayesian network, this system can continuously optimize and improve the diagnosis model with the accumulation of clinical data, improving the accuracy and efficiency of the diagnosis.

[0037] Second, the present invention also introduces uncertainty quantification indicators, providing more comprehensive diagnostic reference information for doctors. By calculating uncertainty quantification indicators such as confidence intervals and entropy values, the present invention can quantify the uncertainty degree of the diagnosis results, helping doctors more accurately evaluate the severity of the disease and its possible development trends, enhancing the objectivity and scientific nature of the diagnosis results, and providing strong support for doctors to develop personalized treatment plans.

[0038] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be learned from the practice of the present invention. Brief Description of the Drawings

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0040] Figure 1 It is an operation flowchart of a medical ultrasonic image analysis system;

[0041] Figure 2 It is a node association process diagram of an image feature extraction module in a medical ultrasonic image analysis system. Specific implementation manners

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0043] Embodiment 1

[0044] This embodiment demonstrates the application of the medical ultrasonic image analysis system of the present invention in the diagnosis of cardiovascular diseases. Through ultrasonic image acquisition, feature extraction, construction and analysis of a Bayesian network diagnosis model, a diagnosis result and suggestions are finally output, providing strong support for the accurate diagnosis of cardiovascular diseases.

[0045] In specific implementation, first, the ultrasonic image acquisition module is connected to a cardiovascular ultrasonic imaging device to acquire ultrasonic images of the patient's heart. During the acquisition process, according to the patient's body position and heart condition, the probe frequency, scanning depth, and gain parameters are adjusted to obtain clear ultrasonic image data. The acquired images are preprocessed to remove noise interference, enhance the contrast and clarity of the images, optimize the image quality, and transmit the images to the image feature extraction module.

[0046] Then, the image feature extraction module extracts edge gradient features from the ultrasonic images. By applying the Sobel operator to the ultrasonic images, the edge gradient amplitude of each pixel point is calculated, and then statistical analysis is performed on the edge gradient amplitude of the entire image, that is, the mean and variance statistics are calculated as the values of this node. This node aims to capture the degree of drastic change and overall features of the tissue boundaries in the image. Because diseased tissues often have differences in boundary morphology from normal tissues, their edges are more blurred, irregular, or have higher gradient changes. Therefore, for the pixel points , calculate the horizontal gradient using the Sobel operator (where is the horizontal gradient operator, is the gray value at pixel ), and the vertical gradient (where is the vertical gradient operator), and calculate the edge gradient magnitude For the entire image (assuming the total number of pixels is , the number of rows of the image is , and the number of columns is ), the edge gradient eigenvalue is used to describe the overall change degree information of the cardiac tissue boundary, and is input into the Bayesian network diagnosis model. Then, extract the regional contrast feature: divide the image into multiple sub-regions, calculate the gray contrast between each sub-region and the entire image respectively, and sum them after assigning weights according to the importance of the sub-regions to obtain this eigenvalue. This node reflects the difference in gray levels between different regions in the image. Lesion regions may show different contrasts from the surrounding normal tissues due to changes in tissue density and structure, which helps to locate and identify the lesion site and its scope. The specific process is as follows: divide the cardiac ultrasound image into non-overlapping sub-regions , , calculate the average gray value of sub-region and the average gray value of the entire image. The contrast value between sub-region and the entire image, and the regional contrast feature (where is the weight of sub-region (i.e., the frequency coordinate corresponding to the maximum spectral amplitude ) and the bandwidth . First, calculate the spectral energy function , the total energy . Cumulatively sum the energy from the low frequency. The frequency range corresponding to when the cumulative energy reaches is the bandwidth , spectral eigenvalue It is used to analyze the relationship between the frequency characteristics of ultrasonic reflection signals and the characteristics of cardiac tissues and is input into the Bayesian network diagnosis model.

[0047] Subsequently, the Bayesian network diagnosis model construction module is used to construct a Bayesian network diagnosis model, and its construction process is as follows: Determine the node type: According to the extracted edge gradient eigenvalue , regional contrast feature , spectral eigenvalue as well as the age and gender information of the patient, determine the nodes in the Bayesian network. The image feature nodes include edge gradient feature nodes , regional contrast feature nodes , spectral feature nodes , and the patient information nodes include age nodes , gender nodes , and the basic node set is ; Determine the probability relationship between nodes: Collect a large amount of clinical cardiovascular disease case data, including the ultrasonic image data of patients, the corresponding calculation results of image features, age, gender information, and the finally diagnosed disease types. For the disease node and the basic node set , calculate the conditional probability distribution . For example, for the disease and a combination of a set of eigenvalue and patient information values , the conditional probability (where represents the number of times that the disease and the eigenvalue combination appear simultaneously in the dataset represents the number of times that the eigenvalue combination appears). Specifically, the association between each node and the disease node is determined as follows: The edge gradient feature node is associated with the disease node. Edge gradient feature extraction is performed on the clinical cardiovascular ultrasonic image samples to obtain an edge gradient eigenvalue set . For the disease , count its corresponding edge gradient eigenvalue subset . The conditional probability distribution of the disease and the edge gradient feature node ; The regional contrast feature node is associated with the disease node. Calculate the regional contrast eigenvalue of the clinical ultrasonic image samples to obtain a regional contrast eigenvalue set . For the disease , determine its corresponding regional contrast eigenvalue subset ,disease With the regional contrast feature node The conditional probability distribution of ; Regional contrast feature nodes are associated with disease nodes to calculate regional contrast feature values ​​of clinical ultrasound image samples , and obtain the regional contrast feature value set , for diseases , determine the corresponding regional contrast feature value subset ,disease With the regional contrast feature node The conditional probability distribution of ; Spectral feature nodes are associated with disease nodes, and spectral feature values ​​are obtained from spectral analysis of clinical ultrasound image samples , for diseases , determine the corresponding subset of spectral eigenvalues ,disease With the Spectrum Feature Node The conditional probability distribution of ; The age node is associated with the disease node, collecting the patient age information and disease diagnosis results in the clinical case data, and dividing the age into different intervals , statistics of each disease in each age range The number of cases And the total number of samples in each age range ,disease With age node The conditional probability distribution of ; Gender nodes are associated with disease nodes, and the number of male and female patients with each disease in clinical case data is counted The number of cases and , and the total sample size of male and female patients and ,disease With gender node The probability of occurrence of the conditional probability distribution ; Construct a Bayesian network structure: Construct a Bayesian network structure based on the determined probabilistic relationship between nodes. Each node in the network represents a random variable, and the directed edges between nodes represent the dependency relationship between variables; Introduce uncertainty quantification indicators: Introduce confidence intervals and entropy values ​​in the Bayesian network diagnostic model. For confidence intervals, calculate the sample mean for the probability of disease occurrence and the sample standard deviation ( It is The estimated probability of disease occurrence for samples is is the sample size), at a confidence level of Under this, the quantile is obtained by looking up the standard normal distribution table , and the confidence interval is . For the entropy value, according to the disease occurrence probability distribution ( is the occurrence probability of disease ), the entropy value .

[0048] Next, the probability inference calculation module inputs the feature data (edge gradient feature value , regional contrast feature , spectral feature value ) obtained by the image feature extraction module and the patient information (age, sex) into the Bayesian network diagnosis model, and uses the Bayesian inference formula (where is the disease node, is the set of evidence nodes, is the probability of the evidence appearing under the condition that the disease occurs, is the prior probability of the disease , is the posterior probability of the disease under the evidence ) to perform iterative calculations. After each new data input, update the and parameters according to the structure and conditional probability distribution of the Bayesian network, recalculate the posterior probability , and at the same time update the confidence interval and entropy value. For example, when a new evidence node is added, calculate , update to , and then recalculate . For the confidence interval update, recalculate and according to the new sample data, and then obtain the new confidence interval; the entropy value update is composed of the new posterior probability to form a new probability distribution , and recalculate .

[0049] Finally, the diagnostic result output module presents the probabilities of various diseases and the corresponding uncertainty quantification information obtained by the probability reasoning calculation module to the doctor in an intuitive and clear manner, and combines the disease occurrence probability and the uncertainty quantification information to intelligently generate detailed diagnostic suggestions. For diseases with high probability and low uncertainty, it is recommended that the doctor further conduct targeted examinations or formulate treatment plans; for cases with low probability but high uncertainty, it prompts the doctor to conduct reexaminations or make comprehensive judgments in combination with other diagnostic methods; for complex situations where the probabilities of multiple diseases are similar and the uncertainty is large, it provides multi-dimensional analysis ideas and recommendations for the direction of further examinations to assist the doctor in making a scientific and reasonable final diagnostic decision.

[0050] In summary, this embodiment elaborates in detail the application process of the present invention in the diagnosis of cardiovascular diseases. Through the links of image acquisition, feature extraction, Bayesian network model construction and analysis, and diagnostic result output, it fully demonstrates the effectiveness and accuracy of the system. This system can comprehensively consider various image features and patient information, accurately calculate the probability of disease occurrence, and provide a more comprehensive diagnostic basis for doctors through uncertainty quantification indicators, effectively assisting doctors in making scientific and reasonable diagnostic decisions and improving the diagnostic level and medical quality of cardiovascular diseases.

[0051] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed claims.

Claims

1. A medical ultrasonic image analysis system, characterized in that: The system consists of: ultrasound image acquisition module, image feature extraction module, Bayesian network diagnosis model construction module, probability reasoning calculation module and diagnosis result output module; The ultrasonic image acquisition module is connected to the medical ultrasonic imaging device and is used to acquire medical ultrasonic images and pre-process the images to optimize the influence of interference factors on diagnosis; The image feature extraction module extracts multiple features from the collected ultrasound image and uses the extracted features as input information of the Bayesian network diagnosis model, wherein the features include edge gradient features, regional contrast features, and spectrum features; The Bayesian network diagnostic model construction module is used to construct a Bayesian network diagnostic model, determine the probability relationship between the nodes of the network, set the extracted image features and patient information as network nodes, and determine the conditional probability distribution between the nodes according to the clinical data to obtain the disease nodes, thereby constructing a complete Bayesian network structure. At the same time, uncertainty quantification indicators, namely confidence intervals and entropy values, are introduced into the model to quantify the degree of uncertainty of the probability results. The specific construction process of the Bayesian network diagnostic model is as follows: Determine the node type: According to the edge gradient feature value EG, regional contrast feature RC, spectrum feature value SF, and patient age and sex information obtained by the image feature extraction module, they are used as image feature nodes and patient information nodes respectively, and the basic node set in the Bayesian network is N = {n1, n2, ..., n k }, where the image feature nodes include edge gradient feature nodes n eg , regional contrast feature node n rc , spectrum feature node n sf , patient information nodes include age nodes n age , gender node n sex ; Determine the probability relationship between nodes: Collect clinical case data, including the patient's ultrasound image data, the corresponding image feature calculation results, the patient's age, gender information, and the final confirmed disease type, and determine the probability distribution between nodes through conditional probability distribution, that is, for the disease node D and the basic node set N = {n1, n2, ..., n k }, calculate the conditional probability distribution P(D|n1,n2,…,n k ), that is, considering the disease d i and a combination of a set of feature values ​​and patient information values ​​(e1, e2, ..., e k ), count the co-occurrence of diseases in the data set d i The number of times it is combined with this feature count(d i , e1, e2, …, e k ), and the number of times this feature combination appears alone count(e1, e2, …, e k ), then the conditional probability All image feature nodes and patient information nodes in the basic node set N are associated with disease nodes through this conditional probability method; Constructing a Bayesian network structure: Based on the determined probabilistic relationship between nodes, a Bayesian network structure is constructed. Each node in the network represents a random variable, and the directed edges between nodes represent the dependency relationship between variables. Introducing uncertainty quantification indicators: Introducing uncertainty quantification indicators into the Bayesian network diagnosis model, including confidence intervals and entropy values, to quantify the degree of uncertainty of probability results; The probability reasoning calculation module inputs the feature data obtained by the image feature extraction module into the Bayesian network diagnosis model, calculates the probability of disease occurrence through the Bayesian network structure, and updates the value of the uncertainty quantification index in real time during the calculation process. With the input of new data and the progress of probability reasoning, the confidence interval and entropy value indicators are dynamically adjusted; The diagnosis result output module presents the disease occurrence probability obtained by the probability reasoning calculation module to the doctor in the form of a probability numerical list, and provides corresponding diagnosis suggestions to assist the doctor in making the final diagnosis decision.

2. A medical ultrasonic image analysis system according to claim 1, characterized in that: In the process of constructing the Bayesian network diagnosis model, the Bayesian network diagnosis model construction module determines the association between the edge gradient feature node of the image feature node and the disease node D by extracting the edge gradient feature of the clinical ultrasound image sample, and obtaining the edge gradient feature value set EG of each sample image = {G1, G2, ..., G n }, where G i is the edge gradient feature value of the i-th sample image, for each disease type d j , statistics the distribution of edge gradient eigenvalues ​​in sample images where the disease exists, that is, disease d j The corresponding edge gradient eigenvalue subset is {G j1 , G j2 , …, G jm }, calculate disease d j With edge gradient feature node n eg The conditional probability distribution P(d j |n eg )for: Among them, count(d j ,G j ) indicates disease j And the edge gradient eigenvalue is G j The number of samples, count(G j ) indicates that the edge gradient eigenvalue is G j The total number of samples.

3. A medical ultrasonic image analysis system according to claim 1, characterized in that: The Bayesian network diagnosis model building module, in the process of building the Bayesian network diagnosis model, for the regional contrast feature node n of the image feature node rc The process of determining the association with the disease node D is as follows: the regional contrast eigenvalue C is calculated for the clinical ultrasound image sample, and the regional contrast eigenvalue set {C1, C2, ..., C n }, for disease d j , determine the corresponding regional contrast feature value subset as {C j1 , C j2 , …, C jm }, calculate disease d j With regional contrast feature node n rc The conditional probability distribution P(d j |n rc )for: Among them, cont(d j , C j ) indicates disease j And the regional contrast characteristic value is C j The number of samples, count(C j ) indicates that the regional contrast feature value is C j The total number of samples.

4. The medical ultrasonic image analysis system according to claim 1, characterized in that: The Bayesian network diagnosis model building module, in the process of building the Bayesian network diagnosis model, for the spectrum feature node n of the image feature node sf The process of determining the association with the disease node D is as follows: obtaining the spectrum feature value F from the spectrum analysis of the clinical ultrasound image sample, including the peak frequency (u p , v p ) and bandwidth B, forming a spectrum feature value set SF = {(u p1 , v p1 , B1), (u p2 , v p2 , B2),…,(u pn , v pm , B n )}, for diseases j , determine the corresponding spectral eigenvalue subset {(u pj1 , v pj1 , B j1 ), (u pj2 , v pj2 , B j2 ),…,(u pjm , v pjm , B jm )}, calculate the disease d j With spectrum feature node n sf The conditional probability distribution P(d j |n sf )for: The count function is used to count the number of samples that combine disease and spectral feature values.

5. The medical ultrasonic image analysis system according to claim 1, characterized in that: The Bayesian network diagnosis model building module, in the process of building the Bayesian network diagnosis model, for the age node n of the patient information node age The process of determining the association with the disease node D is as follows: collect the patient age information and disease diagnosis results in the clinical case data, divide the age into different intervals, and count the number of each disease d in each age interval. j The number of cases count(d j ,age i ) and the total number of samples in each age range count(age i ), calculate disease d j With age node n age The conditional probability distribution P(d j |n age )for: Among them, age i represents the i-th interval in the age interval.

6. The medical ultrasonic image analysis system according to claim 1, characterized in that: The Bayesian network diagnosis model building module, in the process of building the Bayesian network diagnosis model, for the gender node n of the patient information node sex The process of determining the association with disease node D is as follows: Count the number of male and female patients with each disease d in the clinical case data j The number of cases count(d j , sex m ) and count(d j , sex f ), and the total number of male and female patients count(sex m ) and count(sex f ), calculate disease d j With gender node n sex The conditional probability distribution P(d j |n sex )for:

7. The medical ultrasonic image analysis system according to claim 1, characterized in that: The Bayesian network diagnostic model construction module introduces quantitative indicators of uncertainty confidence interval and entropy value in the process of constructing the Bayesian network diagnostic model, where: For the confidence interval: For the probability of disease occurrence P(D), calculate the sample mean and the sample standard deviation σ P(D) , using the unbiased estimation formula in the normal approximation method Where P(Di) is the estimated probability of disease occurrence of the i-th sample, N is the number of samples, and the quantile z is obtained by looking up the standard normal distribution table at a confidence level of 1-α α / 2 , then the confidence interval is This confidence interval can intuitively show the range of the estimated probability of disease occurrence; For entropy value: According to the probability distribution of disease occurrence P = {p1, p2, ..., p m }, p i For disease i The probability of occurrence, the calculated entropy value is: Among them, m is the total number of disease types. A large entropy value indicates a dispersed probability distribution with high uncertainty. A small entropy value indicates that the probability is concentrated on one disease type with low uncertainty. The entropy value is used to quantify the degree of uncertainty in the probability of disease occurrence, providing doctors with more comprehensive diagnostic reference information and helping them evaluate the credibility of the diagnostic results.

8. The medical ultrasonic image analysis system according to claim 1, characterized in that: The probability reasoning calculation module adopts the Bayesian reasoning formula Dynamically adjust the confidence interval and entropy value indicators, where D is the disease node, E is the evidence node set, that is, the image feature and patient information nodes, P(E|D) is the probability of evidence E appearing under the condition of disease D, P(D) is the prior probability of disease D, and P(D|E) is the posterior probability of disease D under evidence E. In the iterative calculation process, after each new data is input, the parameters of P(E|D) and P(D) are updated according to the structure and conditional probability distribution of the Bayesian network, the posterior probability P(D|E) is recalculated, and the confidence interval and entropy value are updated at the same time. That is, when new image feature data and patient information are added, the new conditional probability distribution is calculated according to the new evidence combination, and then the estimated value of the disease probability and its confidence interval and entropy value are updated, and a new evidence node E is set. new Add, calculate P(E new |D), update P(E|D) to P(E, E new |D)=P(E|D)P(E new |D) and then recalculate P(D|E, E new ), for confidence interval updates, recalculate based on new sample data and σ P(D) , and then get a new confidence interval, and the entropy value is updated by the new posterior probability P(D|E, E new ) form a new probability distribution P = {p′1, p′2, …, p′ m }, recalculate 9. The medical ultrasonic image analysis system according to claim 1, characterized in that: The process of extracting edge gradient features in the image feature extraction module is as follows: the gray value I(x, y) at the pixel point (x, y) in the image f(x, y) is used to calculate the horizontal gradient G using the Sobel operator. x (x,y) and the vertical gradient G y (x,y), that is, the horizontal gradient operator is The vertical gradient operator is but Edge gradient magnitude For the entire image, the total number of image pixels is N, then the edge gradient eigenvalue Where M and N are the number of rows and columns of the image, which is used to describe the overall degree of change of the tissue boundary in the image and input into the Bayesian network diagnosis model; The extraction process of the regional contrast feature is as follows: the image is divided into m×n non-overlapping sub-regions R ij , where j = 1, 2, ..., m, j = 1, 2, ..., n, set the sub-region R ij The average gray value is μ ij , the average gray value of the entire image is μ, then the sub-region R ij The contrast value C with the entire image ij =|μ ij -μ|, regional contrast feature where w ij Sub-region R ij The weight of , which is used to reflect the grayscale differences between different regions and input into the Bayesian network diagnosis model; The process of extracting the spectrum features is as follows: performing a two-dimensional discrete Fourier transform 2D-DFT on the ultrasound image, that is, Where M and N are the number of rows and columns of the image, u = 0, 1, ..., M-1, v = 0, 1, ..., N-1, and the spectral characteristics include the peak frequency (u p , v p ), that is, the frequency coordinate corresponding to the maximum spectrum amplitude |F(u, v)|, and the bandwidth B. The corresponding spectrum energy function is E(u, v) = |F(y, v)| 2 , calculate the total energy Start accumulating energy from low frequency, and when the accumulated energy reaches 0.9E total The frequency range corresponding to the time is the bandwidth B. The spectrum feature is used to analyze the relationship between the frequency characteristics of the ultrasonic reflection signal and the tissue characteristics and is used as the model input. The spectrum feature value is SF = (u p ,v p ,B).

10. A medical ultrasonic image analysis method, applicable to a medical ultrasonic image analysis system according to any one of claims 1 to 9, characterized in that: The specific steps of this method are: S1. Start the ultrasound image acquisition module, establish a connection with the ultrasound imaging device, receive and store the acquired ultrasound image data in real time, and record the acquisition time and metadata of the patient information; S2, transmitting the collected ultrasound image to the image feature extraction module, performing edge gradient feature, regional contrast feature, and spectrum feature extraction on the image, storing the processed feature data in the feature database, and establishing an associated index with the patient information; S3. Construct a Bayesian network diagnostic model, determine the node type and number of the Bayesian network, collect clinical case data, calculate the conditional probability distribution between nodes based on the clinical data, and integrate the calculation model of uncertainty quantification indicators into the model so that the model can simultaneously output uncertainty quantification results while calculating the probability of disease occurrence; S4, reading the image feature data and basic information of the current patient from the feature database, inputting the data into the constructed Bayesian network diagnosis model, starting the probability reasoning calculation process, performing iterative calculation, and outputting the disease occurrence probability result and the corresponding uncertainty quantification value; S5. The calculated probability of occurrence of various diseases and their uncertainty quantitative information are transmitted to the diagnosis result output module to generate a visual diagnosis result report.

Images