An ultrasonic intelligent analysis method, system, electronic device and storage medium

Abstract

This invention discloses an intelligent ultrasound analysis method, system, electronic device, and storage medium, relating to the field of ultrasound image processing technology. Based on ultrasound contrast imaging videos, this invention combines radiomics, time-intensity curves, and image development difference duty cycle information to extract high-throughput quantitative ultrasound image features from contrast imaging videos. Furthermore, it effectively integrates these features using feature engineering and machine learning classification methods, thereby improving the evaluation performance of the intelligent analysis technology.

Description

Technical Field

[0001] This invention relates to the field of ultrasound image processing technology, and in particular to an ultrasound intelligent analysis method, system, electronic device, and storage medium. Background Technology

[0002] Ultrasound imaging is the preferred imaging method for clinical screening and diagnosis of diseases. Contrast-enhanced ultrasound, in particular, can display the microvascular structure within tumors in real time, offering high spatiotemporal resolution and visually revealing the microcirculatory characteristics within the tumor. Currently, clinical examinations primarily use contrast-enhanced ultrasound to visualize the blood supply to lesions, combined with time-intensity curves for qualitative evaluation. This method heavily relies on the subjective evaluation of the examiner; inexperienced examiners have a high rate of misinterpretation when analyzing contrast-enhanced ultrasound videos, resulting in significant negative impacts.

[0003] Radiomics is an emerging field in image processing. Its goal is to extract and analyze large amounts of high-throughput quantitative image features from medical images, construct models to describe tumors and predict clinical phenotypes, and establish correlations with tissue disease progression, thereby assisting clinicians in diagnosis. However, few studies have yet been able to extract rich quantitative information from the dynamic imaging process of contrast-enhanced ultrasound. To truly reflect the qualitative information of biological tissue pathogenesis, high-throughput mining of quantitative features from contrast-enhanced ultrasound videos is of great significance, but current technologies lack relevant methods. Summary of the Invention

[0004] To address the aforementioned problems in the existing technology, this invention provides an ultrasonic intelligent analysis method, system, electronic device, and storage medium.

[0005] To achieve the above objectives, the present invention provides the following solution:

[0006] A method for constructing an ultrasonic intelligent analysis model, comprising:

[0007] Acquire ultrasound contrast imaging video data;

[0008] The ultrasound contrast imaging video data is segmented into regions of interest to obtain ultrasound contrast imaging images of interest; the region of interest is the target tissue region.

[0009] Quantitative radiomics features are extracted from the ultrasound contrast-enhanced images of interest to obtain a quantitative radiomics feature matrix;

[0010] Time-intensity curves were plotted based on different types of quantitative radiomics features in the aforementioned quantitative radiomics feature matrix;

[0011] Based on the time-intensity curve, various quantitative characteristic parameters are obtained to form a parameter feature matrix;

[0012] Acquire contrast-enhanced ultrasound images of the region of interest at time T0 and time T0+Ns, respectively.

[0013] Image difference duty cycle information is determined based on the ultrasound contrast image at time T0 and the ultrasound contrast image at time T0+Ns.

[0014] A new parameter feature matrix is ​​obtained by fusing the parameter feature matrix and the image difference duty cycle information;

[0015] The new parameter feature matrix is ​​subjected to feature processing to obtain processed features, and the processed features are input into a first classifier to obtain the optimal feature combination; the parameter conditions of the first classifier are the same as the parameter conditions of each of the new parameter feature matrices input.

[0016] The optimal feature combination is input into the second classifier to obtain the classification accuracy evaluation result.

[0017] Optionally, time-intensity curves are plotted based on different types of quantitative radiomics features in the quantitative radiomics feature matrix, specifically as follows:

[0018] Different types of quantitative radiomics features in the quantitative radiomics feature matrix are used as signal intensity information to plot time-intensity curves.

[0019] Optionally, the image difference duty cycle information is determined based on the contrast-enhanced ultrasound image at time T0 and the contrast-enhanced ultrasound image at time T0+Ns, specifically including:

[0020] Acquire the first frame image of the ultrasound contrast imaging image at time T0 and the first frame image of the extracted ultrasound contrast imaging image at time T0+Ns;

[0021] The first frame of the ultrasound contrast imaging image at time T0 and the first frame of the extracted ultrasound contrast imaging image at time T0+Ns were subjected to grayscale and binarization processing, respectively, to obtain the binarized image of the ultrasound contrast imaging image at time T0 and the binarized image of the extracted ultrasound contrast imaging image at time T0+Ns.

[0022] The image difference duty cycle information is obtained based on the binarized difference between the binarized image of the ultrasound contrast imaging image at time T0 and the binarized image extracted from the ultrasound contrast imaging image at time T0+Ns.

[0023] Optionally, the classifier is one or more of the following: support vector machine, naive Bayes, decision tree, random forest, K-nearest neighbor, logistic regression, and discriminant analysis.

[0024] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0025] This invention is based on ultrasound contrast imaging videos, combined with radiomics and time-intensity curves and image development difference duty cycle information, to achieve high-throughput quantitative ultrasound image features extraction from contrast imaging videos, and uses feature engineering and machine learning classification methods to effectively integrate them, thereby improving the evaluation performance of intelligent analysis technology.

[0026] To implement the above-described method for constructing the ultrasonic intelligent analysis model, the present invention also provides the following implementation structure:

[0027] A system for constructing an ultrasonic intelligent analysis model, applied to the ultrasonic intelligent analysis model construction method provided above; the system includes:

[0028] The data acquisition module is used to acquire ultrasound contrast imaging video data;

[0029] The region segmentation module is used to segment the ultrasound contrast imaging video data into regions of interest to obtain ultrasound contrast imaging images of interest; the region of interest is the target tissue region.

[0030] The matrix extraction module is used to extract the quantitative radiomics features of the ultrasound contrast-enhanced images of interest to obtain a quantitative radiomics feature matrix.

[0031] The curve plotting module is used to plot time-intensity curves based on different types of quantitative radiomics features in the quantitative radiomics feature matrix.

[0032] A matrix forming module is used to obtain various quantitative feature parameters based on the time-intensity curve to form a parameter feature matrix;

[0033] The image acquisition module is used to acquire ultrasound contrast images of the region of interest at time T0 and time T0+Ns, respectively.

[0034] The information determination module is used to determine the image difference duty cycle information based on the ultrasound contrast image at time T0 and the ultrasound contrast image at time T0+Ns.

[0035] The feature fusion module is used to fuse the parameter feature matrix and the image difference duty cycle information to obtain a new parameter feature matrix;

[0036] The feature selection module is used to perform feature processing on the new parameter feature matrix to obtain processed features, and input the processed features into the first classifier to obtain the optimal feature combination; the parameter conditions of the first classifier are the same as the parameter conditions of each of the new parameter feature matrices input.

[0037] The classification evaluation module is used to input the optimal feature combination into the second classifier to obtain the classification accuracy evaluation result.

[0038] An electronic device, comprising:

[0039] Memory, used to store computer programs;

[0040] A processor, connected to the memory, is used to retrieve and execute the computer program to implement the method for constructing the ultrasound intelligent analysis model provided above.

[0041] A computer-readable storage medium for storing a computer program that implements the method for constructing the ultrasonic intelligent analysis model provided above.

[0042] Since the technical effects achieved by the above three implementation structures provided by the present invention are the same as those achieved by the construction method of the ultrasonic intelligent analysis model provided by the present invention, they will not be described again here. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart illustrating the method for constructing the ultrasonic intelligent analysis model provided by this invention;

[0045] Figure 2 This is a schematic diagram of the process for extracting fused radiomics and time-intensity curve features according to the present invention;

[0046] Figure 3 This is a data processing framework diagram for the automated system of radiomics and TIC provided by the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] The purpose of this invention is to provide an intelligent ultrasound analysis method, system, electronic device, and storage medium that can maximize the extraction of quantitative image information from ultrasound contrast-enhanced videos, thereby providing strong data support for clinicians' diagnoses.

[0049] Existing intelligent analysis methods based on ultrasound contrast imaging videos only utilize information about the time-dependent changes in acoustic intensity. Analysis methods using time-intensity curves can semi-automatically extract quantitative information from regions of interest (ROIs) in contrast imaging videos. However, this approach requires human intervention and extracts limited features. On the other hand, radiomics aims to extract and analyze a large number of high-throughput image features from medical images, offering advantages such as automation, high repeatability, and accuracy. Therefore, this invention, based on ROIs in ultrasound contrast imaging videos, first utilizes radiomics to extract high-throughput omics features from the image, then generates a time-intensity curve for each feature, extracts multiple feature parameters from these curves, and finally effectively fuses them using feature engineering and machine learning algorithms to generate a highly accurate intelligent algorithm.

[0050] There is currently a Chinese paper with a similar idea to ours, titled "The Value of Texture Analysis Combined with TIC Curve in Differential Diagnosis of Invasive Ductal Carcinoma and Fibroadenoma of the Breast," but it differs from our invention in at least the following ways: 1. Wang Liang used MRI images, while our invention uses contrast-enhanced ultrasound images; 2. Wang Liang's paper only uses texture features in its radiomics approach, while our invention's radiomics approach includes semantic features, texture features, and other statistical features; 3. The method used in Wang Liang's paper to combine TIC and MRI texture parameter analysis is not clearly defined and may be a simple additive relationship, while our invention's method is a relationship of radiomics features × TIC curve feature parameters. This is the essential difference.

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0052] like Figure 1 As shown, the present invention provides a method for constructing an ultrasonic intelligent analysis model, comprising:

[0053] Step 100: Acquire ultrasound contrast imaging video data. In practical applications, ultrasound contrast imaging video data of the target tissue can be acquired from an ultrasound instrument.

[0054] Step 101: Perform region of interest segmentation on the ultrasound contrast-enhanced video data to obtain ultrasound contrast-enhanced images of interest. The region of interest (ROI) is the target tissue region.

[0055] The target tissue needs to be located by an experienced physician observing the patient's ultrasound video to pinpoint the lesion's location. The region of interest (ROI) segmentation in the ultrasound contrast-enhanced image can be performed using an interactive manual method. This method involves the physician observing the patient's data and extracting the ROI one by one, then annotating the contrast-enhanced video with the relevant ROIs. Redundant text and borders are removed using image coordinate cropping, resulting in a sequence of ultrasound video sequences with ROIs of the same size and aligned with the physician's annotations.

[0056] Step 102: Extract quantitative radiomics features from the contrast-enhanced ultrasound images of interest to obtain a quantitative radiomics feature matrix. This step mainly uses the corresponding contrast-enhanced ultrasound images within the region of interest as signal sources to establish a multi-parameter radiomics model. Quantitative radiomics features are extracted from the ultrasound radiofrequency echo signals within the region of interest. The quantitative radiomics features can be divided into two groups: (1) semantic features, such as features describing the shape and size of the lesion area; (2) agnostic features, including first-order, second-order, and higher-order statistics obtained through statistical analysis methods. Among them, second-order statistical features can also be called texture features.

[0057] Based on step 102, M quantitative radiomics features can be obtained in each frame of ultrasound contrast imaging image. For N frames of ultrasound contrast imaging video, N*M quantitative radiomics feature matrices can be obtained.

[0058] Step 103: Plot time-intensity curves based on different types of quantitative radiomics features in the quantitative radiomics feature matrix.

[0059] Based on the above description, after obtaining the quantitative radiomics feature matrix corresponding to N frames of ultrasound contrast-enhanced video, a time-intensity curve analysis method is introduced to further extract the time-intensity curves of different types of quantitative radiomics features, obtain high-throughput quantitative radiomics time dimension information, and explore the rich biological essence information hidden behind the images.

[0060] The time-intensity curve analysis method reflects the signal intensity of the analysis area depicted in each frame of an ultrasound contrast-enhanced video as it changes over time. It can be used to quantitatively observe the enhancement of ultrasound contrast agents in that area and is a commonly used quantitative analysis method for ultrasound contrast-enhanced videos.

[0061] Step 104: Obtain various quantitative feature parameters based on the time-intensity curve to form a parameter feature matrix. These quantitative feature parameters extracted from the time-intensity curve can be categorized into three types: time-related, velocity-related, and intensity-related. Commonly used parameters include: maximum intensity (IMAX), time to peak (TTP), mean transit time (MTT), the slope of the mean transit time curve, and the area under the curve (AUC). Specifically, IMAX characterizes the maximum intensity of contrast enhancement within the ROI, TTP represents the time it takes for contrast enhancement to reach maximum intensity, MTT represents the time from the start of contrast enhancement to a decrease to half of the maximum intensity, and AUC represents the area under the time-intensity curve.

[0062] Based on the description of steps 103 and 104 above, this invention uses the N*M quantitative radiomics feature matrix corresponding to N frames of ultrasound contrast imaging video, takes M quantitative radiomics parameters as the signal intensity information corresponding to the image, and then plots the corresponding time-intensity curve (TIC) from each quantitative radiomics matrix to obtain the feature parameters on the G time-intensity curves.

[0063] Therefore, after steps 103 and 104, a total of M time-intensity curves are obtained from the quantitative radiomics feature matrix, thereby generating N*M*G high-throughput parameter feature matrices. Here, G represents the number of feature parameters on each time-intensity curve.

[0064] Step 105: Acquire the ultrasound contrast-enhanced image of the region of interest at time T0 and time T0+Ns, respectively. Time T0 is the start of contrast enhancement, and time T0+Ns is the time when the contrast enhancement is at its brightest.

[0065] Step 106: Determine the image difference duty cycle information based on the ultrasound contrast image at time T0 and the ultrasound contrast image at time T0+Ns.

[0066] In this step, it is necessary to determine the binarized difference between the images extracted at these two time points in order to extract the image difference duty cycle information. For example, firstly, the first frame images of the ultrasound contrast imaging image at time T0 and the ultrasound contrast imaging image at the brightest point (i.e., time T0+Ns) are read out and converted into grayscale images and binarized images respectively. After obtaining the binarized images corresponding to the image at time T0 and the brightest point T0+Ns, a subtraction operation is performed on each pixel of the two images to obtain the image after the pixel value difference, which is the binarized difference image.

[0067] Step 107: Fuse the parameter feature matrix and image difference duty cycle information to obtain a new parameter feature matrix.

[0068] Step 108: Perform feature processing on the new parameter feature matrix to obtain processed features, and input the processed features into the first classifier to obtain the optimal feature combination. The parameter conditions of the first classifier are the same as the parameter conditions of each new input parameter feature matrix.

[0069] In this step, feature reduction algorithms can be used to remove irrelevant features, avoiding the curse of dimensionality. Examples include classic principal component analysis, non-zero matrix factorization, and information gain algorithms.

[0070] After feature dimensionality reduction, the data needs to be processed through feature selection. The purpose is to select a subset of features and their combinations that are beneficial to the learning algorithm, thereby improving the accuracy of the established model combining radiomics and time-intensity curves. An iterative approach is used, inputting each combination into a classifier with the same parameters to obtain the optimal feature combination. Commonly used classifiers include Support Vector Machine (SVM), Native Bayesian (NB), Decision Tree (DA), Random Forest (RDF), K-Nearest Neighbor (KNN), Logistic Regression (LR), and Discriminant Analysis (DA).

[0071] Step 109: Input the optimal feature combination into the second classifier to obtain the classification accuracy evaluation result. The selected optimal feature combination is the optimal combination for the corresponding model output result. In practical applications, the area under the highest receiver operating characteristic curve (ROC) is usually selected as the final evaluation result of the model. The second classifier used here is a different classifier from the first classifier.

[0072] The selection process for vascular dynamic features based on radiomics is as follows: Figure 2 As shown. When the method for constructing the ultrasound intelligent analysis model provided by this invention is integrated into an automated system for radiomics and TIC, the implementation process is as follows. Figure 3 As shown.

[0073] Based on the above description, another variation of the present invention can be based on ultrasound contrast imaging video. First, high-throughput texture parameter features in the region of interest are extracted using radiomics image analysis methods. At the same time, time-intensity curve parameters of the ultrasound contrast imaging video are extracted. Then, the two types of parameters are combined and feature filtering and feature fusion are performed to generate a highly accurate intelligent algorithm.

[0074] Compared with the prior art, the beneficial effects of this invention are as follows:

[0075] (1) This invention establishes a multi-parameter model suitable for ultrasound contrast imaging videos, taking into account the advantages of combining time-intensity curves, radiomics, and the difference duty cycle information of image development, and extracting high-throughput radiomics features of target tissues from them.

[0076] (2) The present invention uses feature engineering methods to reduce the dimensionality, select and combine high-throughput features optimally, and uses machine learning algorithms to effectively fuse multiple candidate parameters in order to maximize the mining of quantitative image information from ultrasound contrast video.

[0077] Furthermore, to implement the above-mentioned method for constructing the ultrasonic intelligent analysis model, the present invention also provides the following implementation structure:

[0078] An ultrasonic intelligent analysis model construction system is provided, applied to the ultrasonic intelligent analysis model construction method described above. The system includes:

[0079] The data acquisition module is used to acquire ultrasound contrast video data.

[0080] The region segmentation module is used to segment the ultrasound contrast-enhanced video data into regions of interest (ROIs) to obtain ultrasound contrast-enhanced images of interest. The ROI is the target tissue region.

[0081] The matrix extraction module is used to extract quantitative radiomics features from ultrasound contrast images of interest, and obtain a quantitative radiomics feature matrix.

[0082] The curve plotting module is used to plot time-intensity curves based on different types of quantitative radiomics features in the quantitative radiomics feature matrix.

[0083] The matrix forming module is used to obtain various quantitative characteristic parameters based on the time-intensity curve to form a parametric feature matrix.

[0084] The image acquisition module is used to acquire ultrasound contrast images of the region of interest at time T0 and time T0+Ns, respectively.

[0085] The information determination module is used to determine the image difference duty cycle information based on the ultrasound contrast image at time T0 and the ultrasound contrast image at time T0+Ns.

[0086] The feature fusion module is used to fuse the parameter feature matrix and image difference duty cycle information to obtain a new parameter feature matrix.

[0087] The feature selection module processes the new parameter feature matrix to obtain processed features, which are then input into the first classifier to obtain the optimal feature combination. The parameter conditions of the first classifier are the same as those of each new input parameter feature matrix.

[0088] The classification evaluation module is used to input the optimal feature combination into the second classifier to obtain the classification accuracy evaluation result.

[0089] An electronic device, comprising:

[0090] Memory is used to store computer programs.

[0091] A processor, connected to a memory, is used to retrieve and execute computer programs to implement the method for constructing the ultrasound intelligent analysis model provided above.

[0092] Furthermore, when the aforementioned computer program is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0093] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0094] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for constructing an ultrasonic intelligent analysis model, characterized in that, The method comprises the following steps: acquiring ultrasound contrast video data; segmenting the ultrasound contrast video data to obtain ultrasound contrast interest images; the interest region is a target tissue region; extracting quantitative image genomics features of the ultrasound contrast interest images to obtain a quantitative image genomics feature matrix; drawing time-intensity curves based on different types of quantitative image genomics features in the quantitative image genomics feature matrix; obtaining multiple quantitative feature parameters based on the time-intensity curves to form a parameter feature matrix; acquiring ultrasound contrast images at T0 and T0+Ns of the interest region respectively; determining image difference duty cycle information based on the ultrasound contrast images at T0 and T0+Ns, comprising: acquiring the first frame image of the ultrasound contrast image at T0 and the first frame image of the ultrasound contrast image extraction image at T0+Ns; performing grayscale and binaryzation processing on the first frame image of the ultrasound contrast image at T0 and the first frame image of the ultrasound contrast image extraction image at T0+Ns respectively to obtain the binaryzation image of the ultrasound contrast image at T0 and the binaryzation image of the ultrasound contrast image extraction image at T0+Ns; obtaining the image difference duty cycle information based on the binaryzation difference between the binaryzation image of the ultrasound contrast image at T0 and the binaryzation image of the ultrasound contrast image extraction image at T0+Ns; T0 is the beginning of development, and T0+Ns is the brightest time of development; fusing the parameter feature matrix and the image difference duty cycle information to obtain a new parameter feature matrix; removing irrelevant features from the new parameter feature matrix by using a feature dimension reduction algorithm, performing feature processing after feature dimension reduction to obtain processed features, and inputting the processed features into a first classifier to obtain an optimal feature combination; the parameter conditions of the first classifier are the same as the parameter conditions of each input new parameter feature matrix; inputting the optimal feature combination into a second classifier to obtain a classification accuracy evaluation result. 2.The method of claim 1, wherein, Drawing time-intensity curves based on different types of quantitative image genomics features in the quantitative image genomics feature matrix, specifically: regarding different types of quantitative image genomics features in the quantitative image genomics feature matrix as signal intensity information to draw time-intensity curves. 3.The method of claim 1, wherein, The classifier is one or more of support vector machines, naive Bayes method, decision tree, random forest, K nearest neighbor method, logistic regression and discriminant analysis.

4. A system for constructing an ultrasound intelligent analysis model, characterized by, The system is applied to the construction method of the ultrasound intelligent analysis model according to any one of claims 1-3, and the system comprises: a data acquisition module configured to acquire ultrasound contrast video data; a region segmentation module configured to segment the ultrasound contrast video data to obtain ultrasound contrast interest images; the interest region is a target tissue region; a matrix extraction module configured to extract quantitative image genomics features of the ultrasound contrast interest images to obtain a quantitative image genomics feature matrix; a curve drawing module configured to draw time-intensity curves based on different types of quantitative image genomics features in the quantitative image genomics feature matrix; a matrix forming module configured to obtain a plurality of quantitative characteristic parameters based on the time-intensity curve to form a parameter characteristic matrix; an image acquisition module configured to acquire an ultrasound contrast image at a time T0 and an ultrasound contrast image at a time T0+Ns of the region of interest respectively; an information determination module configured to determine image difference duty cycle information based on the ultrasound contrast image at the time T0 and the ultrasound contrast image at the time T0+Ns, including: acquiring a first frame image of the ultrasound contrast image at the time T0 and a first frame image of the ultrasound contrast image at the time T0+Ns; performing grayscale and binaryzation processing on the first frame image of the ultrasound contrast image at the time T0 and the first frame image of the ultrasound contrast image at the time T0+Ns respectively to obtain a binaryzation image of the ultrasound contrast image at the time T0 and a binaryzation image of the ultrasound contrast image at the time T0+Ns; obtaining the image difference duty cycle information based on a binaryzation difference between the binaryzation image of the ultrasound contrast image at the time T0 and the binaryzation image of the ultrasound contrast image at the time T0+Ns; the time T0 is a developing start time, and the time T0+Ns is a time when the developing is brightest; a feature fusion module configured to fuse the parameter characteristic matrix and the image difference duty cycle information to obtain a new parameter characteristic matrix; a feature selection module configured to remove irrelevant features from the new parameter characteristic matrix using a feature dimension reduction algorithm, perform feature processing after feature dimension reduction, obtain processed features, and input the processed features into a first classifier to obtain an optimal feature combination; a parameter condition of the first classifier is the same as a parameter condition of each of the new parameter characteristic matrices inputted; a classification evaluation module configured to input the optimal feature combination into a second classifier to obtain a classification accuracy evaluation result.

5. An electronic device, comprising: comprising: a memory configured to store a computer program; a processor connected with the memory and configured to call and execute the computer program to implement the method for constructing an ultrasound intelligent analysis model according to any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, a computer program for storing the method for constructing an ultrasound intelligent analysis model according to any one of claims 1-3.

Images