Artificial intelligence prediction method and device for pulmonary hypertension

Abstract

The present invention provides an artificial intelligence prediction method and device for pulmonary hypertension lesions. The artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery computed tomography (CTA) imaging includes the following steps: first, obtaining a patient's pulmonary artery CTA image; second, segmenting the patient's pulmonary artery CTA image using a convolutional neural network to obtain multiple substructure segmentation results; third, obtaining morphological parameters based on the structural segmentation results; fourth, constructing a binary classification model using machine learning based on the obtained morphological parameters; and fifth, using the binary classification model to predict whether the patient has pulmonary hypertension.

Description

Technical Field

[0001] The present invention relates to the medical field, and in particular to an artificial intelligence prediction method and device for pulmonary hypertension lesions based on pulmonary artery CTA imaging. Background Art

[0002] Pulmonary hypertension (PH) is a clinical and pathophysiological syndrome caused by multiple etiologies and mechanisms. Clinically, a mean pulmonary artery pressure >20 mmHg is considered PH. Pulmonary hypertension is a highly detrimental, high-risk disease with significant morbidity and mortality.

[0003] Early and accurate diagnosis and treatment of pulmonary hypertension play a vital role in its prognosis and improving survival rate.

[0004] The gold standard for diagnosing pulmonary hypertension is the invasive procedure of right cardiac catheterization, where a catheter is inserted directly into the pulmonary artery and a transducer is used to measure intraluminal pressure and hemodynamic data. However, right cardiac catheterization is an invasive procedure that requires anesthesia or local anesthesia, carries certain risks, can cause fear or physical discomfort in patients, and is relatively expensive. This limits its application and delays the early diagnosis and treatment of pulmonary hypertension. Developing a noninvasive and accurate diagnostic method is crucial for patients with pulmonary hypertension. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the above-mentioned defects in the existing technology and provide a non-invasive and accurate artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging that can accurately evaluate pulmonary hypertension lesions.

[0006] According to the present invention, an artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging is provided, comprising:

[0007] Step 1: Obtain the patient's pulmonary artery CTA image;

[0008] The second step: Use convolutional neural network to segment the patient's pulmonary artery CTA image and obtain multiple substructure segmentation results;

[0009] The third step: obtaining morphological parameters according to the structure segmentation results;

[0010] Step 4: Use machine learning to build a binary classification model based on the obtained morphological parameters;

[0011] Step 5: Use the binary classification model to predict whether the patient has pulmonary hypertension.

[0012] Preferably, the second step uses a convolutional neural network to segment the patient's pulmonary artery CTA image, thereby extracting the substructure segmentation results of the left atrium, left ventricle, right atrium, right ventricle, pulmonary trunk, left pulmonary artery, right pulmonary artery, superior vena cava, and inferior vena cava.

[0013] Preferably, the convolutional neural network adopts an improved 2D-Unet network, and introduces an attention mechanism based on the 2D-Unet network, using additive attention to introduce information of adjacent layer images.

[0014] Further preferably, in the attention mechanism, the image to be segmented and its underlying image are first additively fused to extract adjacent layer information, and then a feature map is obtained through a convolution operation. Subsequently, the attention weight matrix is obtained through a Sigmoid activation function, and finally the weight matrix is multiplied pixel by pixel with the image to be segmented to obtain a weighted image to be segmented, which is used as the input image of the 2D-Unet.

[0015] Preferably, the first step is to perform data preprocessing on the patient's pulmonary artery CTA image after acquiring the patient's pulmonary artery CTA image.

[0016] Preferably, the patient's pulmonary artery CTA image is acquired using an imaging device, so that the patient's pulmonary artery CTA image includes the structures of the patient's atrium, ventricle, pulmonary artery, and superior and inferior vena cava.

[0017] Preferably, the morphological parameters include: the ratio of the diameters of the right ventricle to the left ventricle, the diameter of the left pulmonary artery, the circumference of the left pulmonary artery, the area of the left pulmonary artery, the diameter of the right pulmonary artery, the circumference of the right pulmonary artery, the area of the right pulmonary artery, the long axis diameter of the superior vena cava, the circumference of the superior vena cava, the area of the superior vena cava, the long axis diameter of the inferior vena cava, the circumference of the inferior vena cava, and the area of the inferior vena cava.

[0018] Preferably, the fourth step comprises: selecting predictive factors from the morphological parameters, and performing modeling using a machine learning algorithm based on the selected predictive factors to construct a binary classification model.

[0019] Further preferably, a chi-square test is used to select a predetermined number of the most relevant morphological parameters from the acquired morphological parameters as predictors; wherein the extracted morphological parameters are screened using the following chi-square test:

[0020] Where O is the observed frequency and E is the expected frequency;

[0021] According to χ 2 The p-value was obtained by calculating the degree of freedom, and the predetermined number of morphological parameters most related to pulmonary hypertension were selected as predictors according to the size of the p-value.

[0022] Preferably, according to the prediction results, the patients are divided into two categories: those with pulmonary hypertension and those without pulmonary hypertension.

[0023] According to the present invention, there is also provided an artificial intelligence prediction device for pulmonary hypertension lesions based on pulmonary artery CTA imaging, comprising:

[0024] CTA image acquisition unit: used to obtain CTA images of the patient's pulmonary artery;

[0025] Segmentation unit: used to segment the patient's pulmonary artery CTA images using a convolutional neural network to obtain multiple substructure segmentation results;

[0026] Morphological parameter acquisition unit: used to obtain morphological parameters according to the structure segmentation results;

[0027] Binary classification model unit: used to build a binary classification model using machine learning based on the acquired morphological parameters, and use the binary classification model to predict whether the patient has pulmonary hypertension.

[0028] Preferably, the segmentation unit uses a convolutional neural network to segment the patient's pulmonary artery CTA image, thereby extracting substructure segmentation results of the left atrium, left ventricle, right atrium, right ventricle, pulmonary trunk, left pulmonary artery, right pulmonary artery, superior vena cava, and inferior vena cava;

[0029] Preferably, the convolutional neural network adopts an improved 2D-Unet network, and introduces an attention mechanism based on the 2D-Unet network, and uses additive attention to introduce information of adjacent layer images; in the attention mechanism, the image to be segmented and its lower layer image are first additively fused to extract adjacent layer information, and then a feature map is obtained through a convolution operation, and then the attention weight matrix is obtained through a Sigmoid activation function, and finally the weight matrix is multiplied pixel by pixel with the image to be segmented to obtain a weighted image to be segmented, which is used as the input image of the 2D-Unet.

[0030] Preferably, an imaging device is used to obtain a pulmonary artery CTA image of the patient, so that the pulmonary artery CTA image of the patient includes the structures of the patient's atrium, ventricle, pulmonary artery, and superior and inferior vena cava; the morphological parameters include: the diameter ratio of the right ventricle to the left ventricle, the left pulmonary artery diameter, the left pulmonary artery circumference, the left pulmonary artery area, the right pulmonary artery diameter, the right pulmonary artery circumference, the right pulmonary artery area, the long-axis diameter of the superior vena cava, the circumference of the superior vena cava, the area of the superior vena cava, the long-axis diameter of the inferior vena cava, the circumference of the inferior vena cava, and the area of the inferior vena cava.

[0031] Preferably, the binary classification model unit selects a predictor from the morphological parameters, and uses a machine learning algorithm to perform modeling based on the selected predictor to construct a binary classification model;

[0032] Furthermore, a predetermined number of morphological parameters that are most relevant are selected from the obtained morphological parameters as predictors using a chi-square test; wherein the extracted morphological parameters are screened using the following chi-square test:

[0033] Where O is the observed frequency and E is the expected frequency;

[0034] According to χ 2 The p-value was obtained by calculating the degree of freedom, and the predetermined number of morphological parameters most related to pulmonary hypertension were selected as predictors according to the size of the p-value.

[0035] This invention uses a machine learning model to automatically predict the prevalence of pulmonary hypertension, employing artificial intelligence methods to achieve noninvasive diagnosis of pulmonary hypertension, reducing the financial and physical burden on patients. The artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging provided by this invention enables noninvasive, accurate, rapid, and precise assessment of pulmonary hypertension lesions. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] A more complete understanding of the present invention and its attendant advantages and features will be more readily appreciated by reference to the following detailed description taken in conjunction with the accompanying drawings, in which:

[0037] Figure 1 The flowchart of the artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging according to a preferred embodiment of the present invention is schematically shown.

[0038] Figure 2 The figure schematically shows an attention mechanism algorithm diagram of an artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging according to a preferred embodiment of the present invention.

[0039] Figure 3 The figure schematically shows a segmentation algorithm diagram of the artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging according to a preferred embodiment of the present invention.

[0040] Figure 4 The figure schematically shows an artificial intelligence prediction device for pulmonary hypertension lesions based on pulmonary artery CTA imaging according to a preferred embodiment of the present invention.

[0041] It should be noted that the accompanying drawings are intended to illustrate the present invention, not to limit it. Note that the accompanying drawings showing structures may not be drawn to scale. Furthermore, in the accompanying drawings, identical or similar elements are labeled with identical or similar reference numerals. DETAILED DESCRIPTION

[0042] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0043] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0044] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0045] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0046] <Method Example>

[0047] Figure 1 The flowchart of the artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging according to a preferred embodiment of the present invention is schematically shown.

[0048] like Figure 1 As shown, the artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging according to a preferred embodiment of the present invention includes:

[0049] Step 1 S1: Obtaining a pulmonary artery CTA image of the patient;

[0050] Preferably, the first step S1 is to perform data preprocessing on the patient's pulmonary artery CTA image after acquiring the patient's pulmonary artery CTA image.

[0051] Specifically, an imaging device is used to obtain a pulmonary artery CTA image of the patient, which includes structures such as the patient's atrium, ventricle, pulmonary artery, and superior and inferior vena cava.

[0052] Furthermore, data preprocessing operations such as resampling and normalization can be performed on the obtained patient's pulmonary artery CTA images to make the images more suitable for subsequent segmentation operations.

[0053] Resampling can make the image pixel spacing consistent and reduce the impact of different data. Interpolation algorithms such as the nearest neighbor interpolation algorithm, bilinear interpolation algorithm, and cubic spline interpolation algorithm can be selected for resampling.

[0054] Normalization can correspond the CT value to between 0 and 1 to prevent the segmentation effect from being affected by large differences in CT values. The normalization formula is:

[0055]

[0056] Wherein, Xnew represents the normalized result, CT i represents the CT value of the i-th time, CT imin represents the minimum CT value obtained, and CT imax represents the maximum CT value obtained.

[0057] Step 2 S2: Use a convolutional neural network to segment the patient's pulmonary artery CTA image and obtain multiple substructure segmentation results;

[0058] Specifically, for example, the second step uses a convolutional neural network to segment the patient's pulmonary artery CTA image, thereby extracting the substructure segmentation results of the left atrium, left ventricle, right atrium, right ventricle, pulmonary trunk, left pulmonary artery, right pulmonary artery, superior vena cava, and inferior vena cava.

[0059] Preferably, the convolutional neural network adopts an improved 2D-Unet network, and introduces an attention mechanism based on the 2D-Unet network, using additive attention to introduce information of adjacent layer images, thereby improving segmentation accuracy.

[0060] The algorithm diagram of the attention mechanism is as follows Figure 2 As shown in the figure, the adjacent layer information is first extracted by additively fusing the image to be segmented and its underlying image, and then the feature map is obtained by convolution operation. Then, the attention weight matrix is obtained by the Sigmoi d activation function. Finally, the matrix is multiplied pixel by pixel with the image to be segmented to obtain the weighted image to be segmented, which is the image input to 2D-Unet. The processed image increases the weight of the target area to be segmented, and increases the model's attention to the target area.

[0061] The attention module is introduced before the 2D-Unet network. The attention module fuses the information of adjacent layers and extracts the weight matrix, superimposes weights on the image to be segmented, strengthens the weight of the segmentation target, and uses the weighted image as the input image of the 2D-Unet network to improve the segmentation network's attention to the target area. The schematic diagram of the algorithm is shown below. Figure 3 shown.

[0062] The third step S3: obtaining morphological parameters according to the structure segmentation result;

[0063] For example, the morphological parameters include: the ratio of the diameters of the right ventricle to the left ventricle, the diameter of the left pulmonary artery, the circumference of the left pulmonary artery, the area of the left pulmonary artery, the diameter of the right pulmonary artery, the circumference of the right pulmonary artery, the area of the right pulmonary artery, the long-axis diameter of the superior vena cava, the circumference of the superior vena cava, the area of the superior vena cava, the long-axis diameter of the inferior vena cava, the circumference of the inferior vena cava, the area of the inferior vena cava, etc. Furthermore, for example, the nine substructures obtained by segmentation can be fitted to calculate the morphological parameters.

[0064] Step 4 S4: constructing a binary classification model using machine learning based on the acquired morphological parameters;

[0065] Preferably, the fourth step S4 comprises: selecting a predictor from the morphological parameters, and performing modeling using a machine learning algorithm according to the selected predictor to construct a binary classification model.

[0066] Further preferably, a chi-square test is used to select a predetermined number (eg, 7) of the most relevant morphological parameters from the acquired morphological parameters as predictors.

[0067] Specifically, for example, to simplify the classification model and reduce computational costs, the following chi-square test can be used to screen the extracted morphological parameters:

[0068]

[0069] Where O is the observed frequency and E is the expected frequency.

[0070] According to χ 2 and degrees of freedom to obtain a p-value (the p-value is a reference for judging the results of hypothesis testing), and a predetermined number (for example, 7) of morphological parameters most relevant to pulmonary hypertension are selected as predictors based on the p-value.

[0071] In addition, for example, a random forest algorithm can be used for modeling to construct a classifier that implements binary classification.

[0072] Step 5 S5: using a binary classification model to predict whether the patient suffers from pulmonary hypertension.

[0073] For example, based on the prediction results, patients are divided into two categories: those with pulmonary hypertension and those without pulmonary hypertension.

[0074] For example, the diagnostic criteria for pulmonary hypertension are: mean pulmonary artery pressure > 20 mmHg measured by right cardiac catheterization at rest at sea level.

[0075] Using the supervised learning method, a large amount of data was used to train the binary classification model (the input of the classifier was seven screened predictors), and the output results divided the patients into two categories: "suffering from pulmonary hypertension" and "not suffering from pulmonary hypertension".

[0076] This invention can use pulmonary artery CTA images to predict the patient's pulmonary hypertension, enabling non-invasive diagnosis of pulmonary hypertension and reducing the patient's financial and physical burden. By using artificial intelligence methods to predict pulmonary hypertension lesions, this method achieves non-invasive diagnosis of pulmonary hypertension compared to traditional right heart catheterization methods.

[0077] In addition, by using appropriate preprocessing and data enhancement methods, CT images can be made more suitable for segmentation and the accuracy of segmentation can be further improved.

[0078] Moreover, by using convolutional neural networks to automatically segment pulmonary artery CTA images and automatically extract morphological parameters, compared with manual processing, it can reduce the workload of doctors, reduce subjectivity in the processing process, and improve the accuracy of predictions.

[0079] Moreover, by using an improved convolutional neural network, the information of adjacent layers can be fused to increase the segmentation model's attention to the segmentation target, thereby improving the segmentation accuracy.

[0080] This invention uses a machine learning model to automatically predict the prevalence of pulmonary hypertension, employing artificial intelligence methods to achieve noninvasive diagnosis of pulmonary hypertension, reducing the financial and physical burden on patients. The artificial intelligence prediction method for pulmonary hypertension lesions based on pulmonary artery CTA imaging provided by this invention enables noninvasive, accurate, rapid, and precise assessment of pulmonary hypertension lesions.

[0081] <Equipment Example>

[0082] Figure 4 The figure schematically shows an artificial intelligence prediction device for pulmonary hypertension lesions based on pulmonary artery CTA imaging according to a preferred embodiment of the present invention.

[0083] like Figure 4 As shown, according to a preferred embodiment of the present invention, an artificial intelligence prediction device for pulmonary hypertension lesions based on pulmonary artery CTA imaging includes:

[0084] CTA image acquisition unit 10: acquires a CTA image of the patient's pulmonary artery;

[0085] Preferably, the CTA image acquisition unit 10 performs data preprocessing on the patient's pulmonary artery CTA image after acquiring the patient's pulmonary artery CTA image.

[0086] Specifically, an imaging device is used to obtain a pulmonary artery CTA image of the patient, which includes structures such as the patient's atrium, ventricle, pulmonary artery, and superior and inferior vena cava.

[0087] Furthermore, data preprocessing operations such as resampling and normalization can be performed on the obtained patient's pulmonary artery CTA images to make the images more suitable for subsequent segmentation operations.

[0088] Resampling can make the image pixel spacing consistent and reduce the impact of different data. Interpolation algorithms such as the nearest neighbor interpolation algorithm, bilinear interpolation algorithm, and cubic spline interpolation algorithm can be selected for resampling.

[0089] Normalization can correspond the CT value to between 0 and 1 to prevent the segmentation effect from being affected by large differences in CT values. The normalization formula is:

[0090]

[0091] Wherein, Xnew represents the normalized result, CTi represents the CT value of the i-th time, CTimin represents the minimum CT value obtained, and CTimax represents the maximum CT value obtained.

[0092] Segmentation unit 20: Use convolutional neural network to segment the patient's pulmonary artery CTA image and obtain multiple substructure segmentation results;

[0093] Specifically, for example, the segmentation unit 20 uses a convolutional neural network to segment the patient's pulmonary artery CTA image, thereby extracting the substructure segmentation results of the left atrium, left ventricle, right atrium, right ventricle, pulmonary trunk, left pulmonary artery, right pulmonary artery, superior vena cava, and inferior vena cava.

[0094] Preferably, the convolutional neural network adopts an improved 2D-Unet network, and introduces an attention mechanism based on the 2D-Unet network, using additive attention to introduce information of adjacent layer images, thereby improving segmentation accuracy.

[0095] The algorithm diagram of the attention mechanism is as follows Figure 2As shown in the figure, the adjacent layer information is first extracted by additively fusing the image to be segmented and its underlying image, and then the feature map is obtained by convolution operation. Then, the attention weight matrix is obtained by the Sigmoid activation function. Finally, the matrix is multiplied pixel by pixel with the image to be segmented to obtain the weighted image to be segmented, which is the image input to 2D-Unet. The processed image increases the weight of the target area to be segmented, and increases the model's attention to the target area.

[0096] The attention module is introduced before the 2D-Unet network. The attention module fuses the information of adjacent layers and extracts the weight matrix, superimposes weights on the image to be segmented, strengthens the weight of the segmentation target, and uses the weighted image as the input image of the 2D-Unet network to improve the segmentation network's attention to the target area. The schematic diagram of the algorithm is shown below. Figure 3 shown.

[0097] Morphological parameter acquisition unit 30: acquires morphological parameters according to the structure segmentation result;

[0098] For example, the morphological parameters include: the ratio of the diameters of the right ventricle to the left ventricle, the diameter of the left pulmonary artery, the circumference of the left pulmonary artery, the area of the left pulmonary artery, the diameter of the right pulmonary artery, the circumference of the right pulmonary artery, the area of the right pulmonary artery, the long-axis diameter of the superior vena cava, the circumference of the superior vena cava, the area of the superior vena cava, the long-axis diameter of the inferior vena cava, the circumference of the inferior vena cava, the area of the inferior vena cava, etc. Furthermore, for example, the nine substructures obtained by segmentation can be fitted to calculate the morphological parameters.

[0099] The binary classification model unit 40 is used to construct a binary classification model using machine learning according to the acquired morphological parameters, and use the binary classification model to predict whether the patient suffers from pulmonary hypertension.

[0100] Preferably, the binary classification model unit 40 selects prediction factors from the morphological parameters, and uses a machine learning algorithm to perform modeling based on the selected prediction factors to construct a binary classification model.

[0101] Further preferably, a chi-square test is used to select a predetermined number (eg, 7) of the most relevant morphological parameters from the acquired morphological parameters as predictors.

[0102] Specifically, for example, to simplify the classification model and reduce computational costs, the following chi-square test can be used to screen the extracted morphological parameters:

[0103]

[0104] Where O is the observed frequency and E is the expected frequency.

[0105] According to χ 2and degrees of freedom to obtain a p-value (the p-value is a reference for judging the results of hypothesis testing), and a predetermined number (for example, 7) of morphological parameters most relevant to pulmonary hypertension are selected as predictors based on the p-value.

[0106] In addition, for example, a random forest algorithm can be used for modeling to construct a classifier that implements binary classification.

[0107] For example, based on the prediction results, patients are divided into two categories: those with pulmonary hypertension and those without pulmonary hypertension.

[0108] For example, the diagnostic criteria for pulmonary hypertension are: mean pulmonary artery pressure > 20 mmHg measured by right cardiac catheterization at rest at sea level.

[0109] Using the supervised learning method, a large amount of data was used to train the binary classification model (the input of the classifier was seven screened predictors), and the output results divided the patients into two categories: "suffering from pulmonary hypertension" and "not suffering from pulmonary hypertension".

[0110] This invention can use pulmonary artery CTA images to predict the patient's pulmonary hypertension, enabling non-invasive diagnosis of pulmonary hypertension and reducing the patient's financial and physical burden. By using artificial intelligence methods to predict pulmonary hypertension lesions, this method achieves non-invasive diagnosis of pulmonary hypertension compared to traditional right heart catheterization methods.

[0111] In addition, by using appropriate preprocessing and data enhancement methods, CT images can be made more suitable for segmentation and the accuracy of segmentation can be further improved.

[0112] Moreover, by using convolutional neural networks to automatically segment pulmonary artery CTA images and automatically extract morphological parameters, compared with manual processing, it can reduce the workload of doctors, reduce subjectivity in the processing process, and improve the accuracy of predictions.

[0113] Moreover, by using an improved convolutional neural network, the information of adjacent layers can be fused to increase the segmentation model's attention to the segmentation target, thereby improving the segmentation accuracy.

[0114] In addition, it should be noted that, unless otherwise specified, the terms "first", "second", "third", etc. in the specification are only used to distinguish the various components, elements, steps, etc. in the specification, and are not used to indicate the logical relationship or sequential relationship between the various components, elements, steps, etc.

[0115] It is understood that although the present invention has been disclosed above with reference to preferred embodiments, the above embodiments are not intended to limit the present invention. For any person skilled in the art, without departing from the scope of the technical solution of the present invention, the technical content disclosed above can be used to make many possible changes and modifications to the technical solution of the present invention, or to modify it into an equivalent embodiment with equivalent changes. Therefore, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of protection of the technical solution of the present invention.

Claims

1. An artificial intelligence prediction method for pulmonary hypertension based on pulmonary artery CTA imaging, characterized by include: Step 1: Obtain the patient's pulmonary artery CTA image; The second step: Use convolutional neural network to segment the patient's pulmonary artery CTA image and obtain multiple substructure segmentation results; The third step: obtaining morphological parameters according to the structure segmentation results; Step 4: Use machine learning to build a binary classification model based on the obtained morphological parameters; Step 5: Use the binary classification model to predict whether the patient has pulmonary hypertension; The convolutional neural network uses an improved 2D-Unet network and introduces an attention mechanism based on the 2D-Unet network. The information of adjacent layer images is introduced by additive attention. In the attention mechanism, the image to be segmented and its underlying image are first additively fused to extract the information of adjacent layers. Then, a feature map is obtained through convolution operation. Then, the attention weight matrix is obtained through Sigmoid activation function. Finally, the weight matrix is multiplied pixel by pixel with the image to be segmented to obtain the weighted image to be segmented, which is used as the input image of 2D-Unet. Moreover, the imaging device is used to obtain the patient's pulmonary artery CTA image, so that the patient's pulmonary artery CTA image includes the structures of the patient's atrium, ventricle, pulmonary artery and superior and inferior vena cava; the morphological parameters include: the diameter ratio of the right ventricle to the left ventricle, the left pulmonary artery diameter, the left pulmonary artery circumference, the left pulmonary artery area, the right pulmonary artery diameter, the right pulmonary artery circumference, the right pulmonary artery area, the long-axis diameter of the superior vena cava, the circumference of the superior vena cava, the area of the superior vena cava, the long-axis diameter of the inferior vena cava, the circumference of the inferior vena cava, and the area of the inferior vena cava.

2. The artificial intelligence prediction method for pulmonary hypertension based on pulmonary artery CTA imaging according to claim 1, characterized in that: The second step uses a convolutional neural network to segment the patient's pulmonary artery CTA image, thereby extracting the substructure segmentation results of the left atrium, left ventricle, right atrium, right ventricle, pulmonary trunk, left pulmonary artery, right pulmonary artery, superior vena cava, and inferior vena cava.

3. The artificial intelligence prediction method for pulmonary hypertension based on pulmonary artery CTA imaging according to claim 1 or 2, characterized in that: The fourth step comprises: selecting predictors from the morphological parameters, and using a machine learning algorithm to perform modeling based on the selected predictors to construct a binary classification model; and selecting a predetermined number of morphological parameters that are most relevant from the obtained morphological parameters as predictors using a chi-square test; wherein the extracted morphological parameters are screened using the following chi-square test: , where O is the observed frequency and E is the expected frequency; according to The p-value was obtained by calculating the degree of freedom, and the predetermined number of morphological parameters most related to pulmonary hypertension were selected as predictors according to the size of the p-value.

4. An artificial intelligence prediction device for pulmonary hypertension based on pulmonary artery CTA imaging, characterized by include: CTA image acquisition unit: used to obtain CTA images of the patient's pulmonary artery; Segmentation unit: used to segment the patient's pulmonary artery CTA images using a convolutional neural network to obtain multiple substructure segmentation results; Morphological parameter acquisition unit: used to obtain morphological parameters according to the structure segmentation results; Binary classification model unit: used to construct a binary classification model using machine learning based on the acquired morphological parameters, and use the binary classification model to predict whether the patient has pulmonary hypertension; The convolutional neural network uses an improved 2D-Unet network and introduces an attention mechanism based on the 2D-Unet network. The information of adjacent layer images is introduced by additive attention. In the attention mechanism, the image to be segmented and its underlying image are first additively fused to extract the information of adjacent layers. Then, a feature map is obtained through convolution operation. Then, the attention weight matrix is obtained through Sigmoid activation function. Finally, the weight matrix is multiplied pixel by pixel with the image to be segmented to obtain the weighted image to be segmented, which is used as the input image of 2D-Unet. Moreover, the imaging device is used to obtain the patient's pulmonary artery CTA image, so that the patient's pulmonary artery CTA image includes the structures of the patient's atrium, ventricle, pulmonary artery and superior and inferior vena cava; the morphological parameters include: the diameter ratio of the right ventricle to the left ventricle, the left pulmonary artery diameter, the left pulmonary artery circumference, the left pulmonary artery area, the right pulmonary artery diameter, the right pulmonary artery circumference, the right pulmonary artery area, the long-axis diameter of the superior vena cava, the circumference of the superior vena cava, the area of the superior vena cava, the long-axis diameter of the inferior vena cava, the circumference of the inferior vena cava, and the area of the inferior vena cava.

5. The artificial intelligence prediction device for pulmonary hypertension based on pulmonary artery CTA imaging according to claim 4, characterized in that: The segmentation unit uses a convolutional neural network to segment the patient's pulmonary artery CTA images, thereby extracting the substructure segmentation results of the left atrium, left ventricle, right atrium, right ventricle, pulmonary trunk, left pulmonary artery, right pulmonary artery, superior vena cava, and inferior vena cava.

6. The artificial intelligence prediction device for pulmonary hypertension based on pulmonary artery CTA imaging according to claim 4 or 5, characterized in that: The binary classification model unit selects a predictor from the morphological parameters, and uses a machine learning algorithm to perform modeling based on the selected predictor to construct a binary classification model; Furthermore, a predetermined number of morphological parameters that are most relevant are selected from the obtained morphological parameters as predictors using a chi-square test; wherein the extracted morphological parameters are screened using the following chi-square test: , where O is the observed frequency and E is the expected frequency; according to The p-value was obtained by calculating the degree of freedom, and the predetermined number of morphological parameters most related to pulmonary hypertension were selected as predictors according to the size of the p-value.

Images