Cardiac magnetic resonance examination method with high accuracy

Abstract

The invention discloses a cardiac magnetic resonance examination method with high accuracy, and relates to the technical field of medical treatment. According to the method, three algorithms of the SVM support vector machine, the CNN volume neural network and the XGboost extreme gradient lifting of an HCM patient are used as basic learners to assist accurate diagnosis of the HCM, and a multi-model fusion mixed training model is constructed under the cooperation of the SVM support vector machine, the CNN volume neural network and the XGboost extreme gradient lifting, so that the diagnosis accuracy of the model is improved; logistic regression is applied to multi-model fusion, the performance of the model is further optimized, and the generalization ability of the model is improved; for an HCM diagnosis task, a new training model is provided, and new technical support is provided for clinical diagnosis.

Description

Technical Field

[0001] The invention relates to the field of medical technology, and in particular to a cardiac magnetic resonance examination method with high accuracy. Background Art

[0002] Hypertrophic cardiomyopathy (HCM) is a common and potentially fatal hereditary cardiovascular disease. According to relevant statistics, the incidence of HCM in my country is no less than 1 / 200, and the disease is notable for its high sudden death rate and high heritability. Cardiac magnetic resonance imaging (CMR) has been widely recognized in the assessment of cardiac structure and function due to its non-invasiveness and versatility, and is currently the gold standard for non-invasive assessment of cardiac morphology.

[0003] However, current CMR technology still has certain limitations in diagnosis. For example, it is still insufficient in distinguishing HCM from hypertensive heart disease, or dilated heart disease from athlete's heart remodeling. For this reason, we propose a cardiac magnetic resonance examination method with high accuracy. Summary of the invention

[0004] The purpose of the present invention is to solve the problems mentioned in the above background technology and provide a cardiac magnetic resonance examination method with high accuracy.

[0005] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions:

[0006] A cardiac magnetic resonance examination method with high accuracy comprises the following steps:

[0007] Step 1: Build a machine learning model to convert digital images into quantitative indicators of various tissue characteristics and perform identification and analysis;

[0008] Step 2: Model evaluation, used to judge the learning model;

[0009] Step 3: extracting image features to extract the collected images;

[0010] Step 4: Image data processing: classifying and dividing the extracted image data;

[0011] Step 5: Build a database, construct an image database, and preset image features for comparison;

[0012] Step 6: feature screening: compare the processed image data with the image features in the database and perform screening;

[0013] Step 7: Statistical analysis: analyzing the data after screening;

[0014] Step 8: Summarize the data after statistical analysis and form a report.

[0015] Furthermore, the machine learning model constructed in step 1 is based on SVM support vector machine, CNN convolution neural network and XGboost extreme gradient boosting.

[0016] Furthermore, the step 1 constructs a machine learning model to convert digital images into quantitative indicators of various tissue characteristics, and performs identification and analysis.

[0017] Furthermore, the machine learning model constructed in step 1 is based on computer vision and image processing technology.

[0018] Furthermore, if the requirements are met in step 2, proceed to the next step; if not, return to the previous step.

[0019] Furthermore, the image features in the database built in step 5 include: left ventricular hypertrophy, myocardial fibrosis and left ventricular sclerosis.

[0020] Furthermore, step 5 also includes: definitions, symptoms, pathophysiology and diagnostic criteria of HCM, HHD, DCM and other related diseases.

[0021] Furthermore, the statistical analysis in step 7 is implemented based on SPSS statistics.

[0022] The beneficial effects of the present invention are as follows:

[0023] 1. The present invention uses three algorithms, namely, SVM support vector machine, CNN convolution neural network and XGboost extreme gradient boosting, as basic learners for HCM patients to assist in the accurate diagnosis of HCM, and constructs a multi-model fusion hybrid training model with the cooperation of SVM support vector machine, CNN convolution neural network and XGboost extreme gradient boosting to improve the diagnostic accuracy of the model.

[0024] 2. The present invention applies logistic regression to multi-model fusion, further optimizes the performance of the model and improves the generalization ability of the model.

[0025] 3. This invention proposes a new training model for HCM diagnosis tasks, providing new technical support for clinical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a work flow chart of the present invention;

[0027] Figure 2 It is a working block diagram of constructing a machine learning model in the present invention;

[0028] Figure 3 It is a working block diagram of library building in the present invention. DETAILED DESCRIPTION

[0029] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0030] See also Figure 1 - Figure 3 The present invention provides a cardiac magnetic resonance examination method with high accuracy, comprising the following steps:

[0031] Step 1: Build a machine learning model to convert digital images into quantitative indicators of multiple tissue characteristics, and perform identification and analysis; in order to achieve further development in the field of HCM diagnosis.

[0032] Step 2: Model evaluation is used to evaluate the learning model; it can ensure the accuracy of the model operation and the effectiveness of the inspection.

[0033] Step 3: extracting image features to extract the collected images;

[0034] Step 4: Image data processing: classifying and dividing the extracted image data;

[0035] Step 5: Build a database, construct an image database, and preset image features for comparison;

[0036] Step 6: feature screening: compare the processed image data with the image features in the database and perform screening;

[0037] Step 7: Statistical analysis: analyzing the data after screening;

[0038] Step 8: Summarize the data after statistical analysis and form a report.

[0039] In this embodiment, preferably, the machine learning model constructed in step 1 is based on SVM support vector machine, CNN convolution neural network and XGboost extreme gradient boosting; wherein, SVM support vector machine can process high-dimensional data, has strong generalization ability, is suitable for small sample data, can handle nonlinear problems, has good robustness and interpretability, etc.; in summary, SVM performs well in processing high-dimensional data, small samples, nonlinear problems, and providing robust and interpretable classification results, and is suitable for a wide range of machine learning tasks. CNN convolutional neural network can extract local features in images. These features are invariant to the position of the image, which means that CNN can uniformly process features at different positions, thereby improving the adaptability to different scales and angles in the image, automatically learning and detecting important features in the input data without manual feature engineering, which makes CNN more efficient and accurate in processing complex tasks. The filters (or kernels) of the convolutional layer slide on different parts of the input image, sharing weights and biases. This sharing mechanism significantly reduces the number of network parameters while maintaining high recognition accuracy. The layers of CNN are only partially connected, which enhances the sparsity and parallelism of the network, reduces the number of training parameters, and helps prevent overfitting. By performing pooling operations of different sizes on the original image, CNN can effectively reduce the dimension of the feature vector, reduce the amount of calculation, and help extract the main features while maintaining image transformations (such as CNN has the advantages of robustness to translation, rotation, etc. Deep CNN learns the feature hierarchy and abstracts higher-level feature representations layer by layer, so as to perform image recognition and classification tasks more effectively. CNN performs well in processing large-scale image data, can extract useful information from large amounts of data, and improve the generalization ability of the model through training. Due to the use of efficient underlying implementation methods such as matrix multiplication, CNN has high computational efficiency when processing image data. CNN achieves positional translation invariance of features in the image through the design of shared weights and biases, local receptive fields, and pooling layers, that is, no matter where the features appear in the image, they can be recognized. CNN can directly input the original image for training, avoiding the complex preprocessing steps in traditional image processing and simplifying the complexity of model development. In summary, these advantages of CNN give it a broad application prospect in the field of image recognition and processing, and it has become one of the core technologies in this field.XGboost extreme gradient boosting has made many improvements in algorithm optimization, making it faster in training speed than traditional gradient boosting tree (GBDT). By integrating multiple weak learners (usually decision trees) to build a strong learner, XGBoost can improve the prediction accuracy of the model. XGBoost allows users to customize the loss function, so as to flexibly adapt to different machine learning tasks. XGBoost adds regularization terms in the cost function to control the complexity of the model, reduce the variance of the model, and prevent overfitting. XGBoost supports parallel computing, including parallelism and block structure at the feature granularity, which can accelerate the training process and is suitable for processing large-scale data sets. XGBoost has been optimized to effectively process data containing missing values. The distributed version of XGBoost supports running on various distributed environments such as Kubernetes, Hadoop, SGE, MPI, Dask, etc., and has good portability. In summary, XGBoost has been widely used and recognized in the field of machine learning, especially in structured data, for its high efficiency, flexibility and powerful performance. A multi-model fusion hybrid training model was constructed with the cooperation of SVM support vector machine, CNN convolutional neural network and XGboost extreme gradient boosting to assist in the accurate diagnosis of HCM and improve the diagnostic accuracy of the model.

[0040] In this embodiment, preferably, step 1 constructs a machine learning model to convert digital images into quantitative indicators of various tissue characteristics, and performs identification and analysis, in order to achieve further development in the field of HCM diagnosis.

[0041] In this embodiment, preferably, the machine learning model constructed in step 1 is based on computer vision and image processing technology, which can extract image features and related data processing.

[0042] In this embodiment, preferably, if the requirements are met in step 2, proceed to the next step, if not, return to the previous step.

[0043] In this embodiment, preferably, the image features in the database built in step 5 include: left ventricular hypertrophy, myocardial fibrosis and left ventricular sclerosis; accurate division can be achieved, ensuring the accuracy of the examination.

[0044] In this embodiment, preferably, step 5 also includes: definitions, symptoms, pathophysiology and diagnostic criteria of HCM, HHD, DCM and other related diseases; and the ability to provide standards for the results of the examination, that is, to ensure the accuracy of the examination.

[0045] In this embodiment, preferably, the statistical analysis in step 7 is implemented based on SPSS statistics. All operations of SPSS are performed in a unified window mode. Users only need to select variables, set statistical parameters, and output results. They do not need to have an in-depth understanding of statistical expertise and can complete professional statistical analysis using default settings; SPSS provides a wealth of statistical methods, including descriptive statistics, mean comparison, general linear model, correlation analysis, regression analysis, log-linear model, cluster analysis, data simplification, survival analysis, time series analysis, multiple responses, etc., which can meet various data analysis needs; SPSS's interface design is concise and easy to understand, and the operation is relatively simple, suitable for beginners and professionals. Its Windows-style interface is extremely friendly, and the data view is similar to the Excel layout, which is convenient for users to quickly get started; SPSS provides a wealth of charts and image display tools, which are convenient for data analysis presentation and report writing, making the results more intuitive and easy to understand; SPSS supports large-scale data set processing and management, can cope with complex data analysis needs, and supports direct access to Excel. l and other data sources; SPSS has powerful programming capabilities and supports secondary development. Users can write scripts or plug-ins to expand its functions as needed; SPSS is applicable to various fields of natural sciences, technical sciences, and social sciences, and is widely used in economics, mathematics, statistics, logistics management, biology, psychology and other fields; SPSS has a high degree of recognition in the international academic community. Many scientific researchers and international conferences use SPSS as a data analysis tool, and its analysis results usually do not require detailed explanation of the algorithm in international academic exchanges; SPSS supports multiple operating systems such as Wi ndows and Mac OS X, making it convenient for users to use on different platforms; SPSS's output results are not only clear and beautiful, but also support transfer to multiple formats, which is convenient for data sharing and report writing; In summary, SPSS statistics occupies an important position among many statistical software with its simple operation, powerful functions and user-friendly interface, and is suitable for users from different backgrounds to process and analyze data.

[0046] The working principle and use process of the present invention:

[0047] Step 1: Build a machine learning model to convert digital images into quantitative indicators of multiple tissue characteristics, and perform identification and analysis; in order to achieve further development in the diagnosis of HCM. The machine learning model is based on SVM support vector machine, CNN convolution neural network and XGboost extreme gradient boosting; among them, SVM performs well in processing high-dimensional data, small samples, nonlinear problems and providing robust and interpretable classification results, and is suitable for a wide range of machine learning tasks; CNN has broad application prospects in the field of image recognition and processing, and has become one of the core technologies in this field; XGBoost has been widely used and recognized in the field of machine learning, especially in structured data, for its high efficiency, flexibility and powerful performance; with the cooperation of SVM support vector machine, CNN convolution neural network and XGboost extreme gradient boosting, a multi-model fusion hybrid training model is constructed to assist the accurate diagnosis of HCM and improve the diagnostic accuracy of the model. Convert digital images into quantitative indicators of multiple tissue characteristics, and perform identification and analysis, in order to achieve further development in the diagnosis of HCM; in addition, the machine learning model is based on computer vision and image processing technology, which can extract image features and related data processing.

[0048] Step 2: Model evaluation is used to evaluate the learning model, which can ensure the accuracy of the model and the effect of the inspection. If the requirements are met, proceed to the next step; if not, return to the previous step.

[0049] Step 3: Extract image features to extract the collected images.

[0050] Step 4: Image data processing: classify and divide the extracted image data.

[0051] Step 5: Building a database, constructing an image database, and presetting image features for comparison; in this embodiment, preferably, the image features in the database built in step 5 include: left ventricular hypertrophy, myocardial fibrosis, and left ventricular sclerosis; accurate division can be achieved to ensure the accuracy of the examination. It also includes: the definition, symptoms, pathophysiology and diagnostic criteria of related diseases such as HCM, HHD, DCM, etc.; the results of the examination can provide standards, that is, the accuracy of the examination can be guaranteed.

[0052] Step 6: feature screening: compare the processed image data with the image features in the database and perform screening;

[0053] Step 7: Statistical analysis: analyze the data after screening; it is implemented based on SPSS statistics. SPSS statistics occupies an important position among many statistical software with its simple operation, powerful functions and user-friendly interface. It is suitable for users with different backgrounds to process and analyze data and can ensure the accuracy of the inspection.

[0054] Step 8: Summarize the data after statistical analysis and form a report.

[0055] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A cardiac magnetic resonance examination method with high accuracy, characterized in that: The steps include: Step 1: Build a machine learning model to convert digital images into quantitative indicators of various tissue characteristics and perform identification and analysis; Step 2: Model evaluation, used to judge the learning model; Step 3: extracting image features to extract the collected images; Step 4: Image data processing: classifying and dividing the extracted image data; Step 5: Build a database, construct an image database, and preset image features for comparison; Step 6: feature screening: compare the processed image data with the image features in the database and perform screening; Step 7: Statistical analysis: analyzing the data after screening; Step 8: Summarize the data after statistical analysis and form a report.

2. A cardiac magnetic resonance examination method with high accuracy according to claim 1, characterized in that: The machine learning model constructed in step 1 is based on SVM support vector machine, CNN convolution neural network and XGboost extreme gradient boosting.

3. A cardiac magnetic resonance examination method with high accuracy according to claim 1, characterized in that: The step 1 constructs a machine learning model to convert digital images into quantitative indicators of various tissue characteristics, and performs identification and analysis.

4. A cardiac magnetic resonance examination method with high accuracy according to claim 1, characterized in that: The machine learning model constructed in step 1 is based on computer vision and image processing technology.

5. The method for cardiac magnetic resonance examination with high accuracy according to claim 1, characterized in that: If the requirements are met in step 2, proceed to the next step; if not, return to the previous step.

6. A cardiac magnetic resonance examination method with high accuracy according to claim 1, characterized in that: The image features in the database built in step 5 include: left ventricular hypertrophy, myocardial fibrosis and left ventricular sclerosis.

7. A cardiac magnetic resonance examination method with high accuracy according to claim 1, characterized in that: The step 5 also includes: the definition, symptoms, pathophysiology and diagnostic criteria of HCM, HHD, DCM and other related diseases.

8. A cardiac magnetic resonance examination method with high accuracy according to claim 1, characterized in that: The statistical analysis in step 7 is implemented based on SPSS statistics.

Images