Cardiac ultrasonic strain calculation method and device, medium and program product

Abstract

The invention provides a cardiac ultrasonic strain calculation method and device, a medium and a program product, and belongs to the field of computer models and biomedical engineering.The method comprises the steps that a to-be-recognized ultrasonic cardiac image is obtained; performing key feature extraction on the echocardiography image by using a pre-trained echocardiography image type identification model; identifying different chamber view types of the echocardiography image according to the extracted key features; performing image segmentation by using a pre-trained segmentation network based on a diffusion model to obtain an image sequence containing a plurality of chamber views; identifying a target chamber region based on the image sequence, and calculating a target chamber area of the target chamber region for each time frame; constructing an area change sequence representing the area of the target chamber under different time frames; and calculating a target strain value of the target chamber region according to the area change sequence. According to the technical scheme, the accuracy and efficiency of the marking result and the strain calculation result of the cardiac ultrasound image are improved.

Description

Technical Field

[0001] This application belongs to the fields of computer models and biomedical engineering, and particularly relates to a method, device, medium, and program product for calculating cardiac ultrasound strain. Background Art

[0002] With the rapid development of imaging technology, the evaluation technology of cardiac ultrasound images has become increasingly mature. During the evaluation of cardiac ultrasound images, it is usually necessary to label the ventricular boundaries to measure and calculate the movement of the ventricles.

[0003] The related technologies mainly involve doctors manually labeling the ventricular boundaries in cardiac ultrasound images, using two-dimensional echocardiograms to measure the movement of the ventricles, and based on speckle tracking technology, analyzing the speckle patterns of the ventricular walls to track the movement of the myocardium during systole and diastole.

[0004] The main disadvantages of the related technologies are as follows:

[0005] First, the manual labeling method depends on the subjective experience and skills of the operator, resulting in large differences and errors in the labeling results. For complex ventricular structures, it is difficult to accurately capture the fine anatomical structures of the ventricles by the manual labeling method, leading to low accuracy in the strain measurement and calculation of the ventricles.

[0006] Second, in the case of cardiac pathological conditions, the ventricular morphology may change significantly, making the manual labeling operation more difficult, the labeling accuracy lower, and the labeling errors between different operators larger. Summary of the Invention

[0007] The purpose of this application is to provide a method, device, medium, and program product for calculating cardiac ultrasound strain, aiming to use a deep learning model for automatic classification and segmentation, improve the accuracy and efficiency of the labeling results and strain calculation results of cardiac ultrasound images, and reduce the manual labeling errors between different operators.

[0008] According to the first aspect of this application, a method for calculating cardiac ultrasound strain is provided, including:

[0009] Obtain an echocardiogram image to be recognized;

[0010] Extract key features from the echocardiogram image using a pre-trained echocardiogram image type recognition model;

[0011] Identify different chamber view types of the echocardiogram image according to the extracted key features;

[0012] Use a pre-trained segmentation network based on a diffusion model to perform image segmentation on the echocardiogram image, obtaining an image sequence containing several chamber views;

[0013] Identify the target chamber region based on the image sequence, and calculate the target chamber area of the target chamber region for each time frame;

[0014] Based on the target chamber areas of each time frame, construct an area change sequence representing the sizes of the target chamber areas at different time frames;

[0015] Calculate the target strain value of the target chamber region according to the area change sequence.

[0016] In an alternative embodiment, using a pre-trained segmentation network based on a diffusion model to perform image segmentation on the echocardiogram image, obtaining an image sequence containing several chamber views, includes:

[0017] Use the pre-trained segmentation network based on a diffusion model to perform image segmentation on each frame of the echocardiogram image, generating mask images of each chamber view in the two-chamber view, three-chamber view, and four-chamber view;

[0018] Generate an image sequence based on the mask images of each chamber view in the two-chamber view, three-chamber view, and four-chamber view.

[0019] In an alternative embodiment, identifying the target chamber region based on the image sequence, and calculating the target chamber area of the target chamber region for each time frame, includes:

[0020] Identify the target mask image of the target chamber region based on the image sequence;

[0021] For the target mask image of the target chamber region of each time frame, calculate the target chamber area of the target chamber region, and the calculation formula is as follows:

[0022] Area t =∑ i，j II(mask t [i, j]>0);

[0023] Where mask t [i, j] represents the pixel value of the target mask image at the i-th row and j-th column; mask t represents the target mask image of the target chamber region at the t-th time frame; Area t represents the target chamber area of the target chamber region;

[0024] II() represents an indicator function used to describe whether a certain condition is satisfied; mask t [i, j] > 0 is a conditional judgment. If the pixel value of the target mask image at the i-th row and j-th column is greater than 0, it is determined that the pixel at the i-th row and j-th column of the target mask image belongs to the target chamber area, Area t is valued as 1; if the pixel value of the target mask image at the i-th row and j-th column is not greater than 0, it is determined that the pixel at the i-th row and j-th column of the target mask image belongs to the background area, Area t is valued as 0;

[0025] ∑ i，j represents the summation calculation of the judgment results of the pixel positions (i, j) in the mask images of each chamber view; II(mask t [i, j] > 0) represents traversing all the pixel positions (i, j) in mask t . If the pixel value of this pixel position (i, j) is greater than 0, then Area t is valued as 1. If the pixel value of this pixel position (i, j) is not greater than 0, then Area t is valued as 0; ∑ i，j II(mask t [i, j] > 0) represents accumulating the judgment results of all pixel positions (i, j) to obtain the number of pixels in mask t whose judgment results are greater than 0.

[0026] In an alternative embodiment, calculating the target strain value of the target chamber area according to the area change sequence includes:

[0027] Taking the historical chamber area of the historical chamber area as a training sample and inputting it into an initial time series prediction LSTM model for model training until the LSTM model converges to obtain a trained LSTM model;

[0028] Inputting the area change sequence into the trained LSTM model and obtaining the target diastolic area and target systolic area of the target chamber area output by the trained LSTM model:

[0029] Calculating the target strain value of the target chamber area based on the target diastolic area and target systolic area of the target chamber area.

[0030] In an alternative embodiment, calculating the target strain value of the target chamber area based on the target diastolic area and target systolic area of the target chamber area includes:

[0031] Obtain the target diastolic area and target systolic area of the target chamber region under each chamber view;

[0032] Based on the target diastolic area and target systolic area of the target chamber region under each chamber view, calculate the strain value of the target chamber region under each chamber view; the calculation formula is as follows:

[0033]

[0034] where Strain represents the strain value of the target chamber region; Area max represents the target diastolic area of the target chamber region; Area min represents the target systolic area of the target chamber region;

[0035] Calculate the average value of the strain values of the target chamber region under each of the chamber views, and use the average value as the target strain value of the target chamber region; the calculation formula is as follows;

[0036]

[0037] where Average_Strain represents the target strain value of the target chamber region; Strain 2ch represents the first strain value of the target chamber region under the bi - chamber view; Strain 3ch represents the second strain value of the target chamber region under the three - chamber view; Strain 4ch represents the third strain value of the target chamber region under the four - chamber view.

[0038] In an alternative embodiment, obtaining the echocardiogram image to be recognized includes:

[0039] Obtain a cardiac ultrasound image as the echocardiogram image to be recognized;

[0040] Or,

[0041] Obtain a cardiac ultrasound image and a magnetic resonance imaging (MRI) image;

[0042] Fuse the cardiac ultrasound image and the magnetic resonance imaging (MRI) image to obtain a first multi - modal fusion image as the echocardiogram image to be recognized;

[0043] Or,

[0044] Obtain a cardiac ultrasound image and a computed tomography (CT) image;

[0045] Fuse the cardiac ultrasound image and the computed tomography (CT) image to obtain a second multimodal fusion image as the echocardiogram image to be recognized.

[0046] According to a second aspect of the present application, there is provided a computing device for cardiac ultrasound strain, comprising:

[0047] An acquisition unit configured to acquire an echocardiogram image to be recognized.

[0048] A key feature extraction unit configured to extract key features from the echocardiogram image by using a pre-trained echocardiogram image type recognition model.

[0049] An identification unit configured to identify different chamber view types of the echocardiogram image according to the extracted key features.

[0050] A segmentation unit configured to perform image segmentation processing on the echocardiogram image by using a pre-trained segmentation network based on a diffusion model to obtain an image sequence including several chamber views.

[0051] An area calculation unit configured to identify a target chamber region based on the image sequence and calculate the target chamber area of the target chamber region for each time frame.

[0052] A construction unit configured to construct an area change sequence representing the sizes of the target chamber areas at different time frames based on the target chamber areas of each time frame.

[0053] A strain value calculation unit configured to calculate the target strain value of the target chamber region according to the area change sequence.

[0054] According to a third aspect of the present application, there is provided a computer device, which includes:

[0055] At least one processor; and,

[0056] A memory communicatively connected to the at least one processor; wherein,

[0057] The memory stores instructions executable by the at least one processor, enabling the at least one processor to execute the above-mentioned method for calculating cardiac ultrasound strain.

[0058] According to a fourth aspect of the present application, there is provided a computer-readable storage medium storing computer instructions for causing a computer to execute the above-mentioned method for calculating cardiac ultrasound strain.

[0059] According to a fifth aspect of the present application, there is provided a computer program product including computer instructions which, when executed by a processor, implement the above-described method for calculating cardiac ultrasound strain.

[0060] Compared with the related art, the technical solution of the present application has the following advantages:

[0061] Based on a deep learning model and automated image annotation and recognition technologies, the present application realizes full automation from echocardiogram image acquisition to strain value calculation, improving the image acquisition efficiency, image annotation efficiency, and strain value calculation efficiency. By automatically classifying and segmenting the acquired echocardiogram images based on the deep learning model and utilizing the powerful feature extraction ability of the deep learning model, the accuracy and efficiency of the annotation results and strain calculation results of cardiac ultrasound images are improved, and the manual annotation errors between different operators are reduced. It can be applied to processing complex cardiac ultrasound images.

[0062] Other features and advantages of the present application will be described in the following specification, and will be partially obvious from the specification, or will be understood by implementing the present application. The objectives and other advantages of the present application can be realized and obtained through the structures and processes pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] To more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the drawings required for use in the description of the embodiments or the related art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0064] Figure 1 It is a schematic flowchart of a method for calculating cardiac ultrasound strain provided by the present application;

[0065] Figure 2 It is a schematic structural diagram of a device for calculating cardiac ultrasound strain provided by the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] In order to make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0067] Based on the above analysis, the present application proposes a method, device, medium and program product for calculating cardiac ultrasound strain. By using a deep learning model to automatically classify and segment echocardiographic images, the dependence on the operator's experience is significantly reduced, and the influence of human factors on the image classification and segmentation results is minimized. The key features of echocardiographic images are accurately extracted and analyzed using deep learning technology to ensure the accuracy of calculating the target strain value in the target chamber region of echocardiographic images. The present application is particularly applied to processing relatively complex cardiac anatomical structures, and can significantly improve the image acquisition efficiency, image annotation efficiency and strain value calculation efficiency for complex cardiac ultrasound images, saving labor costs. Through a fully automated processing flow, the present application avoids the cumbersome manual operations of the operator, enabling the calculation of left ventricular strain to be efficiently and stably applied in clinical practice and improving the overall work efficiency.

[0068] See Figure 1 the flowchart of Figure 1 which is a schematic flowchart of a method for calculating cardiac ultrasound strain provided by the present application. According to the first aspect of the present application, a method for calculating cardiac ultrasound strain provided by the present application includes the following steps:

[0069] S101. Obtain an echocardiographic image to be recognized.

[0070] In an alternative embodiment, obtaining an echocardiographic image to be recognized includes:

[0071] Obtain a cardiac ultrasound image as the echocardiographic image to be recognized;

[0072] Or,

[0073] Obtain a cardiac ultrasound image and a magnetic resonance imaging (MRI) image;

[0074] Fuse the cardiac ultrasound image and the magnetic resonance imaging (MRI) image to obtain a first multi-modal fusion image as the echocardiographic image to be recognized;

[0075] Or,

[0076] Obtain a cardiac ultrasound image and a computed tomography (CT) image;

[0077] Fuse the cardiac ultrasound image and the computed tomography (CT) image to obtain a second multi-modal fusion image as the echocardiographic image to be recognized.

[0078] The present application can separately obtain a cardiac ultrasound image as the echocardiographic image to be recognized for image segmentation and recognition, or can obtain multiple types of images for fusion processing to obtain a multi-modal fusion image, thereby improving the accuracy and comprehensiveness of cardiac ultrasound image strain calculation.

[0079] S102. Extract key features from the echocardiogram image using a pre-trained echocardiogram image type recognition model.

[0080] Optionally, use historical echocardiogram images with different chambers as training samples to train an initial deep learning model until the initial deep learning model can effectively extract key features from echocardiogram images of different chamber view types to be recognized, so as to accurately distinguish heart views of different chamber view types, and obtain a trained echocardiogram image type recognition model.

[0081] Optionally, the initial deep learning model can adopt ConvNext (Convolutional Neural Network), and the ConvNext model has strong feature extraction capabilities and is suitable for processing image classification tasks. The ConvNext model can effectively extract key features from the echocardiogram images to be recognized in this application to accurately distinguish heart views of different chamber view types.

[0082] It should be noted that in this application, ConvNext (Convolutional Neural Network) is exemplarily adopted, and users can also adopt other neural network models according to actual application requirements. The embodiments of the present disclosure do not limit this.

[0083] S103. Identify different chamber view types of the echocardiogram image according to the extracted key features.

[0084] Optionally, the different chamber view types of the recognized echocardiogram image include two-chamber view, three-chamber view, and four-chamber view. Among them, the two-chamber view usually refers to the view showing the left atrium and the left ventricle in the echocardiogram image, and the two-chamber view is usually used to observe the anatomical relationship between the atrium and the ventricle and evaluate the overall structure and function of the heart; the three-chamber view usually refers to the view showing the left atrium, the left ventricle, and the right ventricle in the echocardiogram image, and the three-chamber view is usually used to more detailedly evaluate the overall structure and function of the heart, including the movement of the valves and the hemodynamic state; the four-chamber view usually refers to the view showing the left atrium, the left ventricle, the right atrium, and the right ventricle in the echocardiogram image, and the four-chamber view is used to comprehensively evaluate the structure and function of the heart, especially in cases of valvular diseases, atrial fibrillation, etc., and can provide important clinical information.

[0085] S104. Perform image segmentation processing on the echocardiogram image using a pre-trained segmentation network based on a diffusion model to obtain an image sequence containing several chamber views.

[0086] This application uses a pre-trained segmentation network based on a diffusion model to perform segmentation processing on echocardiogram images, obtaining an image sequence containing two-chamber views, three-chamber views, and four-chamber views, so as to identify and extract the target chamber region. Taking the identification and extraction of the left ventricle region as an example, the automated segmentation and identification process achieves accurate extraction of the left ventricle boundary, reduces errors caused by manual annotation of images by operators, and thus improves the accuracy and efficiency of subsequent left ventricle strain calculation.

[0087] Optionally, historical echocardiogram images and corresponding left ventricle segmentation masks are used as training samples to train the initial segmentation network Medsegdiff based on the diffusion model until the initial segmentation network Medsegdiff based on the diffusion model converges, obtaining a trained Medsegdiff segmentation network.

[0088] It should be noted that this application takes the identification and extraction of the left ventricle region in echocardiogram images as an example for subsequent left ventricle area calculation and left ventricle strain calculation. Users can extract the corresponding chamber region in echocardiogram images according to actual needs and perform area calculation and strain calculation on the corresponding chamber region. This application does not make any restrictions on this.

[0089] S105. Identify the target chamber region based on the image sequence, and calculate the target chamber area of the target chamber region for each time frame.

[0090] For each time frame, calculate the left ventricle area of the segmented left ventricle region based on the image sequence.

[0091] S106. Based on the target chamber area of each time frame, construct an area change sequence representing the size of the target chamber area at different time frames.

[0092] S107. Calculate the target strain value of the target chamber region according to the area change sequence.

[0093] Optionally, according to the area change sequence, use the trained LSTM model to capture the dynamic changes of the contraction and relaxation of the left ventricle, calculate the target diastolic area and target systolic area of the left ventricle, and calculate the target strain value of the left ventricle according to the target diastolic area and target systolic area of the left ventricle.

[0094] In an optional implementation, use a pre-trained segmentation network based on a diffusion model to perform image segmentation processing on echocardiogram images, obtaining an image sequence containing several chamber views, including:

[0095] Use a pre-trained segmentation network based on a diffusion model to perform image segmentation on each frame of echocardiogram images, generating mask images for each chamber view in the two-chamber view, three-chamber view, and four-chamber view;

[0096] Generate an image sequence based on the mask images of each chamber view in the two-chamber view, three-chamber view, and four-chamber view.

[0097] In an optional embodiment, identify the target chamber region based on the image sequence and calculate the target chamber area of the target chamber region for each time frame, including:

[0098] Identify the target mask image of the target chamber region based on the image sequence;

[0099] For the target mask image of the target chamber region in each time frame, calculate the target chamber area of the target chamber region. The calculation formula is as follows:

[0100] Area t =∑ i，j II(mask t [i, j]>0);

[0101] Where, mask t [i, j] represents the pixel value of the target mask image at the i-th row and j-th column; mask t represents the target mask image of the target chamber region in the t-th time frame; Area t represents the target chamber area of the target chamber region;

[0102] II() represents an indicator function used to describe whether a certain condition is met; mask t [i, j]>0 is a conditional judgment. If the pixel value of the target mask image at the i-th row and j-th column is greater than 0, it is determined that the pixel of the target mask image at the i-th row and j-th column belongs to the target chamber region, and the value of Area t is recorded as 1; if the pixel value of the target mask image at the i-th row and j-th column is not greater than 0, it is determined that the pixel of the target mask image at the i-th row and j-th column belongs to the background region, and the value of Area t is 0;

[0103] ∑ i，j represents the summation calculation of the judgment results of the pixel positions (i, j) in the mask images of each chamber view; II(mask t [i, j]>0) represents traversing all the pixel positions (i, j) in mask t . If the pixel value of this pixel position (i, j) is greater than 0, then Area tis denoted as 1. If the pixel value at the pixel position (i, j) is not greater than 0, then Area t is denoted as 0; ∑ i，j II(mask t [i, j] > 0) represents the accumulation of the judgment results for all pixel positions (i, j), obtaining the number of pixels in mask t where the judgment result is greater than 0.

[0104] In an alternative embodiment, according to the area change sequence, calculating the target strain value of the target chamber region includes:

[0105] Taking the historical chamber area of the historical chamber region as a training sample and inputting it into the initial time series prediction LSTM model for model training until the LSTM model converges, obtaining the trained LSTM model;

[0106] Inputting the area change sequence into the trained LSTM model and obtaining the target diastolic area and target systolic area of the target chamber region output by the trained LSTM model;

[0107] Based on the target diastolic area and target systolic area of the target chamber region, calculating the target strain value of the target chamber region.

[0108] Optionally, using the trained LSTM model to capture the dynamic changes of contraction and relaxation of the target chamber region based on the area change sequence. In this application, the target chamber region takes the left ventricular region as an example.

[0109] In an alternative embodiment, based on the target diastolic area and target systolic area of the target chamber region, calculating the target strain value of the target chamber region includes:

[0110] Obtaining the target diastolic area and target systolic area of the target chamber region under each chamber view;

[0111] Based on the target diastolic area and target systolic area of the target chamber region under each chamber view, calculating the strain value of the target chamber region under each chamber view; The calculation formula is as follows:

[0112]

[0113] where Strain represents the strain value of the target chamber region; Area max represents the target diastolic area of the target chamber region; Area min represents the target systolic area of the target chamber region; The area change sequence is expressed as Area s = [Area 0 ,Area 1, …, Area N , where Area s represents the area change sequence, and Area M represents the area of the target chamber region in different time frames. The value range of M is (0, 1, …, N), and (0, 1, …, N) represents each time frame;

[0114] Calculate the average value of the strain values of the target chamber region in each chamber view, and use the average value as the target strain value of the target chamber region; The calculation formula is as follows;

[0115]

[0116] where Average_Strain represents the target strain value of the target chamber region; Strain 2ch represents the first strain value of the target chamber region in the two-chamber view; Strain 3ch represents the second strain value of the target chamber region in the three-chamber view; Strain 4ch represents the third strain value of the target chamber region in the four-chamber view.

[0117] Optionally, for each frame of echocardiogram image, this application uses a pre-trained segmentation network based on a diffusion model to perform segmentation processing to obtain mask images of each chamber view in the two-chamber view, three-chamber view, and four-chamber view, and generate an image sequence. And identify the target mask image of the left ventricle from the image sequence, and calculate the left ventricular area of the left ventricular region for the target mask image of the left ventricle in each time frame.

[0118] Construct an area change sequence based on the left ventricular area of each time frame. Input the area change sequence into the trained LSTM model to capture the dynamic changes of the contraction and relaxation of the left ventricle, so as to obtain the left ventricular diastolic area and left ventricular systolic area of the left ventricular region output by the trained LSTM model. Calculate the left ventricular diastolic area and left ventricular systolic area of the left ventricular region in the two-chamber view, three-chamber view, and four-chamber view respectively, where the diastolic area refers to the maximum area and the systolic area refers to the minimum area.

[0119] Based on the left ventricular diastolic area and left ventricular systolic area of the left ventricular region in the two-chamber view, calculate the first strain value of the left ventricular region in the two-chamber view; Based on the left ventricular diastolic area and left ventricular systolic area of the left ventricular region in the three-chamber view, calculate the second strain value of the left ventricular region in the three-chamber view; Based on the left ventricular diastolic area and left ventricular systolic area of the left ventricular region in the four-chamber view, calculate the third strain value of the left ventricular region in the four-chamber view. Finally, calculate the average value of the first strain value, the second strain value, and the third strain value as the final target strain value of the left ventricular region.

[0120] Optionally, the present application can also generate a visualization chart based on the target strain value of the left ventricular region, display the strain change trend at different time points through the visualization chart, and can also generate the average strain value of the left ventricular region, facilitating clinical analysis and decision-making by doctors.

[0121] Optionally, the present application can also continuously optimize the model through an online learning mechanism to adapt to different echocardiogram images and improve applicability, and by establishing a user feedback mechanism, continuously improve the segmentation and classification models to improve the performance and practicality of the models for segmenting and classifying echocardiogram images.

[0122] Based on a deep learning model and automated image annotation and recognition techniques, the present application realizes full automation from echocardiogram image acquisition to strain value calculation, improves the image acquisition efficiency, image annotation efficiency, and strain value calculation efficiency. Based on the deep learning model, the acquired echocardiogram images are automatically classified and segmented. Utilizing the powerful feature extraction ability of the deep learning model, the accuracy and efficiency of the annotation results and strain calculation results of cardiac ultrasound images are improved, and the manual annotation errors between different operators are reduced, and it can be applied to processing complex cardiac ultrasound images.

[0123] See Figure 2 , Figure 2 which is a schematic structural diagram of a calculation device for cardiac ultrasound strain provided by the present application. According to the second aspect of the present application, a calculation device for cardiac ultrasound strain is provided, including: an acquisition unit 21 configured to acquire an echocardiogram image to be recognized; a key feature extraction unit 22 configured to extract key features from the echocardiogram image by using a pre-trained echocardiogram image type recognition model; an identification unit 23 configured to identify different chamber view types of the echocardiogram image according to the extracted key features; a segmentation unit 24 configured to perform image segmentation processing on the echocardiogram image by using a pre-trained segmentation network based on a diffusion model to obtain an image sequence including several chamber views; an area calculation unit 25 configured to identify a target chamber area based on the image sequence and calculate the target chamber area of the target chamber area for each time frame; a construction unit 26 configured to construct an area change sequence representing the size of the target chamber area at different time frames based on the target chamber area of each time frame; a strain value calculation unit 27 configured to calculate the target strain value of the target chamber area according to the area change sequence.

[0124] The above device can be implemented by a method for calculating cardiac ultrasound strain provided in the embodiments of the first aspect. The specific implementation manner can be referred to the description in the embodiments of the first aspect and will not be elaborated here.

[0125] According to a third aspect of the present application, a computer device is provided. The computer device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-described method for calculating cardiac ultrasound strain.

[0126] According to a fourth aspect of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores computer instructions for causing a computer to execute the above-described method for calculating cardiac ultrasound strain.

[0127] According to a fifth aspect of the present application, a computer program product is provided, including computer instructions that implement the above-described method for calculating cardiac ultrasound strain when executed by a processor.

[0128] It can be understood that the model structures, names, and parameters described in the above embodiments are only examples. Those skilled in the art can also make easily conceivable combinations and adjustments to the structural features of the above-mentioned multiple embodiments according to the usage requirements, and should not limit the concept of the present application to the specific details of the above examples.

[0129] Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for calculating cardiac ultrasonic strain, characterized in that: include: acquiring an ultrasound cardiac image to be identified; Extracting key features of the ultrasonic cardiac image using a pre-trained ultrasonic cardiac image type recognition model; identifying different chamber view types of the ultrasound cardiac image based on the extracted key features; Performing image segmentation processing on the ultrasound cardiac image using a pre-trained diffusion model-based segmentation network to obtain an image sequence including a plurality of chamber views; identifying a target chamber region based on the image sequence, and calculating a target chamber area of ​​the target chamber region for each time frame; Based on the target chamber area in each time frame, constructing an area change sequence representing the size of the target chamber area in different time frames; According to the area change sequence, a target strain value of the target chamber area is calculated.

2. The method for calculating cardiac ultrasonic strain according to claim 1, characterized in that: The ultrasound cardiac image is segmented using a pre-trained diffusion model-based segmentation network to obtain an image sequence containing several chamber views, including: Performing image segmentation processing on each frame of the ultrasound cardiac image using the pre-trained diffusion model-based segmentation network to generate a mask image of each chamber view in a dual-chamber view, a three-chamber view, and a four-chamber view; An image sequence is generated based on the mask image of each chamber view among the dual-chamber view, the three-chamber view and the four-chamber view.

3. The method for calculating cardiac ultrasonic strain according to claim 2, characterized in that: Identifying a target chamber region based on the image sequence, and calculating a target chamber area of ​​the target chamber region for each time frame, comprising: identifying a target mask image of the target chamber region based on the image sequence; For the target mask image of the target chamber area in each time frame, the target chamber area of ​​the target chamber area is calculated, and the calculation formula is as follows: Area t =∑ i，j II(mask t [i,j]>0); Among them, mask t [i, j] represents the pixel value of the target mask image in the i-th row and j-th column; mask t The target mask image representing the target chamber area at the t-th time frame; Area t a target chamber area representing the target chamber area; II() indicates an indicator function, which is used to describe whether a certain condition is met; mask t [i, j]>0 is a conditional judgment. If the pixel value of the target mask image in the i-th row and the j-th column is greater than 0, it is determined that the pixel of the target mask image in the i-th row and the j-th column belongs to the target cavity area. t The value of is 1; if the pixel value of the target mask image in the i-th row and the j-th column is not greater than 0, it is determined that the pixel of the target mask image in the i-th row and the j-th column belongs to the background area, Area t The value of is 0; ∑ i，j represents the summation of the judgment results of the pixel position (i, j) in the mask image of each chamber view; II (mask t [i, j]>0) indicates that the mask t All pixel positions (i, j) in Area are traversed. If the pixel value of the pixel position (i, j) is greater than 0, then Area t The value of is 1. If the pixel value of the pixel position (i, j) is not greater than 0, then Area t The value of is recorded as 0; i，j II(mask t [i, j]>0) means that the judgment results of all pixel positions (i, j) are accumulated to obtain mask t The number of pixels whose judgment result is greater than 0.

4. The method for calculating cardiac ultrasonic strain according to claim 3, characterized in that: Calculating a target strain value of the target chamber region according to the area change sequence includes: The historical cavity area of ​​the historical cavity area is used as a training sample, and input into the initial time series prediction LSTM model for model training until the LSTM model converges to obtain a trained LSTM model; Inputting the area change sequence into the trained LSTM model, and obtaining the target diastolic area and the target systolic area of ​​the target chamber area output by the trained LSTM model; A target strain value of the target chamber region is calculated based on the target diastolic area and the target systolic area of ​​the target chamber region.

5. The method for calculating cardiac ultrasonic strain according to claim 4, characterized in that: Calculating a target strain value of the target chamber region based on a target diastolic area and a target systolic area of ​​the target chamber region includes: Acquire a target diastolic area and a target systolic area of ​​the target chamber region in each chamber view; Based on the target diastolic area and the target systolic area of ​​the target chamber area in each chamber view, the strain value of the target chamber area in each chamber view is calculated; the calculation formula is as follows: Wherein, Strain represents the strain value of the target chamber area; Area max represents the target diastolic area of ​​the target chamber area; Area min a target systolic area representing the target chamber region; Calculate the average value of the strain value of the target chamber area under each of the chamber views, and use the average value as the target strain value of the target chamber area; the calculation formula is as follows; Among them, Average _ Strain represents the target strain value of the target chamber area; 2ch represents the first strain value of the target chamber area in the dual-chamber view; Strain 3ch represents the second strain value of the target chamber area under the three-chamber view; Strain 4ch Represents the third strain value of the target chamber area in the four-chamber view.

6. The method for calculating cardiac ultrasonic strain according to claim 5, characterized in that: Acquire an echocardiogram to be identified, including: Acquiring a cardiac ultrasound image as the ultrasound cardiogram image to be identified; or, Acquire cardiac ultrasound images and magnetic resonance imaging (MRI) images; fusing the cardiac ultrasound image and the magnetic resonance imaging (MRI) image to obtain a first multimodal fusion image as the ultrasonic cardiogram to be identified; or, Obtain cardiac ultrasound images and computed tomography (CT) images; The cardiac ultrasound image and the computerized tomography (CT) image are fused to obtain a second multimodal fused image as the ultrasonic cardiogram to be identified.

7. A device for calculating cardiac ultrasonic strain, characterized in that: include: an acquisition unit configured to acquire an ultrasound cardiology image to be identified; A key feature extraction unit is configured to extract key features from the ultrasonic cardiac image using a pre-trained ultrasonic cardiac image type recognition model; an identification unit configured to identify different chamber view types of the ultrasound cardiac image according to the extracted key features; A segmentation unit is configured to perform image segmentation processing on the ultrasound cardiac image using a pre-trained diffusion model-based segmentation network to obtain an image sequence including a plurality of chamber views; an area calculation unit configured to identify a target chamber region based on the image sequence, and calculate a target chamber area of ​​the target chamber region for each time frame; A construction unit, configured to construct an area change sequence representing the size of the target chamber area in different time frames based on the target chamber area in each time frame; The strain value calculation unit is configured to calculate a target strain value of the target chamber area according to the area change sequence.

8. A computer device, characterized in that: The computer device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, so that the at least one processor can execute the method for calculating cardiac ultrasonic strain according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method for calculating cardiac ultrasonic strain as described in any one of claims 1 to 6.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the method for calculating cardiac ultrasonic strain according to any one of claims 1 to 6 is implemented.

Images