A fully automated method and device for quantitative analysis of myocardial acoustic contrast imaging

Abstract

This application relates to a fully automated method and apparatus for quantitative analysis of myocardial acoustic contrast imaging. The method includes: acquiring myocardial acoustic contrast imaging images; segmenting the acquired myocardial acoustic contrast imaging images to obtain myocardial segmentation images; calculating the cardiac chamber area based on the myocardial segmentation images; automatically extracting the myocardial segmentation results at the end of cardiac systole using the periodic changes in cardiac chamber area; using the extracted end-systolic segmentation results as a mask to extract the corresponding position of each image as a target region for registration, thereby generating a one-to-one correspondence of registered image sequences; fitting a function model based on the pixel values ​​of each point in the registered image sequence; and generating a parametric image based on the registered image sequence and the fitted function model. Therefore, this application can perform fully automated quantitative analysis of myocardial acoustic contrast imaging.

Description

Technical Field

[0001] This application relates to the field of medical image processing, and in particular to a fully automated method and apparatus for quantitative analysis of myocardial acoustic contrast imaging. Background Technology

[0002] Cardiac imaging plays a crucial role in the prevention, diagnosis, and treatment of ischemic heart disease. Among these techniques, myocardial acoustic contrast imaging (MCE) can effectively help doctors assess myocardial perfusion.

[0003] The principle of myocardial acoustic contrast imaging is to use tiny bubbles in the contrast agent to enhance the echo of muscle tissue, thereby obtaining a clear view of the left ventricle and myocardial tissue to aid in the diagnosis of ischemic heart disease. After digitizing the myocardial acoustic contrast images, valuable clinical indicators can be extracted using image attributes (such as image grayscale values).

[0004] Studies have shown that during contrast-enhanced ultrasound imaging, the concentration of air bubbles within blood vessels increases over time and is correlated with image brightness (grayscale value). Under the premise of a stable contrast agent injection rate, the air bubble concentration reaches a stable peak after a period of time and no longer changes. This pattern can be represented by a single exponential model.

[0005] I(t)=A(1-e -βt )

[0006] Description. Where is the image grayscale value, A is the peak contrast intensity, which represents myocardial blood volume, and β is the contrast replenishment rate, which represents the average blood flow velocity. The product of A and β represents myocardial blood flow.

[0007] Existing quantitative analysis systems for myocardial ultrasound (such as Philips' QLAB quantitative software and the quantitative analysis system developed by Sun Fengrong et al.) all require manual definition of the ROI (Region of Interest). Some products even require fine-tuning of each frame of the image to achieve satisfactory results. This complex operation not only increases the learning cost for doctors but also significantly increases the time required for diagnosis. Summary of the Invention

[0008] Therefore, it is necessary to provide a fully automated method and device for quantitative analysis of myocardial acoustic contrast imaging to address the aforementioned technical problems.

[0009] This application discloses a fully automated quantitative analysis method for myocardial acoustic contrast imaging, comprising:

[0010] Acquiring myocardial acoustic contrast images;

[0011] The acquired acoustic contrast images of the myocardium are segmented to obtain segmented images of the myocardium;

[0012] The cardiac chamber area is calculated based on the segmented cardiac chamber images. The cardiac chamber area is periodically changed, and the segmented cardiac chamber area at the end of cardiac systole is automatically extracted. The extracted segmented cardiac chamber area at the end of cardiac systole is used as a mask to extract the corresponding position of each image as the target region for registration, so as to generate a one-to-one correspondence of registered image sequence.

[0013] A function model is fitted based on the pixel values ​​of each point in the registered image sequence, and a parametric image is generated based on the registered image sequence and the fitted function model.

[0014] Optionally, the segmentation of the acquired myocardial acoustic contrast images to obtain myocardial segmentation images specifically includes:

[0015] Based on the acquired myocardial acoustic contrast images, several myocardial acoustic contrast images were collected and formed into an initial image sequence;

[0016] The initial image sequence is input into a deep neural network model to obtain myocardial segmentation images;

[0017] The deep neural network model is trained based on the U-Net model; the training path of the deep neural network model includes a shrinking path and an expanding path.

[0018] Optionally, the shrinking path includes 9 encoding blocks, each encoding block contains two 3×3 convolutional layers, and each convolutional layer is followed by a batch normalization layer and a ReLU activation layer. The first 8 encoding blocks are downsampled using a 2×2 max pooling layer. The feature maps of corresponding blocks in the shrinking and expanding paths are skip-connected according to the channel dimension.

[0019] Optionally, the expansion path includes 8 decoding blocks, each decoding block contains two 3×3 convolutional layers, each convolutional layer is followed by a batch normalization layer and a ReLU activation layer, and each decoding block is preceded by a 2×2 deconvolutional layer for upsampling.

[0020] Optionally, during each iteration of the training process, each myocardial ultrasound contrast image in the initial image sequence is transformed N times to obtain N amplified image sequences. Each transformation includes random rotation, random translation, random cropping, and random scaling performed sequentially. The initial image sequence and the amplified image sequence together form the training image set.

[0021] Optionally, a sequence of registered images with one-to-one correspondence between points is generated, specifically including:

[0022] For each segmented myocardial image, calculate the convex hull image of the myocardium;

[0023] The myocardial segmentation results at the end of cardiac systole are extracted using the convex hull region of the myocardium in each image;

[0024] Using the myocardial segmentation results at the end of cardiac systole as the target image, the remaining myocardial segmentation results are registered to generate a registered image sequence.

[0025] Optionally, generating a registered image sequence with one-to-one correspondence between points further includes:

[0026] The initial cardiac chamber region image is obtained by subtracting the segmented myocardium image from the convex hull image of the myocardium.

[0027] Morphological filtering was used to process the image to obtain an ideal cardiac chamber region image.

[0028] The area of ​​the heart chambers was calculated using images of each heart chamber region, and the curves showing the changes in the heart chamber area were recorded and generated.

[0029] Find the minimum value in the curve of change of heart chamber area, record the index of each minimum value and calculate the number of interval frames, and take the mode of all interval frames as the cardiac cycle.

[0030] All end-systolic images from the myocardial segmentation image sequence and all end-systolic images from the original MCE image sequence were extracted using the cardiac cycle.

[0031] Using the image with the smallest central cavity area in the end-contraction segmentation image sequence as a mask, images within the corresponding mask region are extracted from all images in the original end-contraction image sequence to generate a mask image sequence;

[0032] After using the masked image sequence as the target region, the large deformation differential homeomorphism method is used to register all images in the end-systolic segmented image sequence to ensure that the pixels in the myocardial region of each image correspond one-to-one. The registered image sequence is then saved and transferred to the quantitative calculation module for parameter calculation.

[0033] Optionally, a function model is fitted based on the pixel values ​​of each point in the registered image sequence, and a parameter image is generated based on the registered image sequence and the fitted function model, specifically including:

[0034] Extract the grayscale value of each point in the registered image sequence;

[0035] The unknown parameters in the calculation function model are fitted using the extracted gray values;

[0036] Time-intensity curves and parameter graphs were plotted based on the mathematical model.

[0037] Alternatively, the function model is: I(t) = A(1-e -βt );

[0038] Where I is the image grayscale value, A is the peak contrast intensity, and β is the contrast replenishment rate.

[0039] This application also provides a fully automated quantitative analysis system for myocardial acoustic contrast imaging, comprising:

[0040] Image acquisition module, used to acquire myocardial acoustic contrast images;

[0041] The image segmentation module is used to segment the acquired myocardial acoustic contrast images to obtain myocardial segmentation images;

[0042] The image processing module is used to calculate the cardiac chamber area based on the myocardial segmentation image, automatically extract the myocardial segmentation results at the end of cardiac systole by utilizing the periodic change characteristics of the cardiac chamber area, and use the extracted end-systole segmentation results as a mask to extract the corresponding position of each image as the target region for registration, so as to generate a one-to-one correspondence of registered image sequence.

[0043] The quantitative calculation module is used to fit a function model based on the pixel values ​​of each point in the registered image sequence, and to generate a parametric image based on the registered image sequence and the fitted function model.

[0044] The fully automated quantitative analysis method for myocardial acoustic contrast imaging of this application has at least the following advantages:

[0045] This application utilizes a deep neural network for automatic segmentation of myocardial acoustic contrast imaging sequences, enabling the automatic acquisition of segmented myocardial image sequences. Based on the area of ​​the myocardial region in the segmented image sequence, the action cycle can be automatically calculated, and a segmented myocardial image at the end of cardiac systole can be obtained. No longer relying on manual selection of the region of interest, the remaining myocardial segmented images can be automatically registered based on the segmented myocardial image at the end of cardiac systole. Based on the registered myocardial segmented images, a function model can be automatically fitted, and based on the fitted function model, myocardial blood volume, mean blood flow velocity, and myocardial blood flow can be automatically calculated and parametric images generated. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating a fully automated quantitative analysis method for myocardial acoustic contrast imaging in one embodiment of this application.

[0047] Figure 2 This is a flowchart of an AI myocardial segmentation algorithm in one embodiment of this application;

[0048] Figure 3 This is a schematic diagram of the convex hull algorithm in one embodiment of this application;

[0049] Figure 4 This is an AI segmentation result from one embodiment of this application;

[0050] Figure 5 This refers to the convex hull region of the AI ​​segmentation result in one embodiment of this application;

[0051] Figure 6This is the result of subtracting the myocardial region from the convex hull region in one embodiment of this application;

[0052] Figure 7 This is the result of morphological filtering on the myocardial region in one embodiment of this application;

[0053] Figure 8 This is an original MCE image of the region corresponding to the smallest myocardium in one embodiment of this application;

[0054] Figure 9 This is a time-intensity curve in one embodiment of this application;

[0055] Figure 10 This is a parameter image of myocardial blood flow in one embodiment of this application. Detailed Implementation

[0056] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0057] In this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number or order of the indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] See Figure 1 One embodiment of this application provides a fully automated quantitative analysis method for myocardial acoustic contrast imaging, including steps S100 to S400. Wherein:

[0060] Step S100: Acquire myocardial acoustic contrast images.

[0061] Myocardial acoustic contrast (MCE) images of the patient were acquired.

[0062] Step S200: Segment the acquired myocardial acoustic contrast image to obtain a myocardial segmentation image.

[0063] Myocardial acoustic contrast (MCE) image sequences are input into a pre-trained deep neural network model, which is based on an 8-layer U-Net model. The 8-layer U-Net model includes a contraction path and an expansion path. The contraction path consists of 9 encoding blocks, each containing two 3×3 convolutional layers, followed by a batch normalization layer and a ReLU activation layer. The first 8 encoding blocks are downsampled using a 2×2 max-pooling layer. The expansion path consists of 8 decoding blocks, each containing two 3×3 convolutional layers, followed by a batch normalization layer and a ReLU activation layer. The first decoding block is upsampled using a 2×2 deconvolutional layer. Feature maps of corresponding blocks in the contraction and expansion paths are skipped along the channel dimension. Residual structures are incorporated into each block of the 8-layer U-Net model. In the expansion path, an auxiliary path is added during the first 7 upsampling operations.

[0064] The training process of this deep neural network model requires image transformation and augmentation operations.

[0065] Specifically, several myocardial ultrasound (MCE) images were acquired to form an initial image set. During each iteration of the training process, each myocardial ultrasound image in the initial image set was transformed N times to obtain N augmented images, where N ≥ 8. Each transformation included random rotation, random translation, random cropping, and random scaling. The random rotation angle ranged from -2.5° to 2.5°, the random translation pixel ranged from -0.02 × image width (image height) to 0.02 × image width (image height), the random cropping angle ranged from -2.5° to 2.5°, and the random scaling ratio ranged from 0.90 to 1.10. The initial image set and the augmented images together constituted the training image set.

[0066] Image segmentation processing flow is as follows Figure 2 As shown.

[0067] Step 300: Calculate the cardiac chamber area based on the myocardial segmentation image, automatically extract the myocardial segmentation results at the end of cardiac systole using the periodic change characteristics of the cardiac chamber area, and use the extracted end-systole segmentation results as a mask to extract the corresponding position of each image as the target region for registration, so as to generate a one-to-one correspondence of registered image sequence.

[0068] The area of ​​the heart chambers in each image is calculated based on the myocardial segmentation results output by the deep neural network.

[0069] Specifically, the convex hull algorithm is applied to the myocardial segmentation results. Figure 3 , Figure 4 ), which yields the convex hull region ( Figure 5 After subtracting the myocardial region from the convex hull region ( Figure 6Morphological filtering was used to remove excess parts to obtain the cardiac chamber area. Figure 7 ).

[0070] The cardiac cycle is calculated based on the cardiac chamber area in each image.

[0071] Specifically, the heart chamber region is extracted, the heart chamber area is calculated, and a heart chamber area change curve is generated. The minimum value in the heart chamber area change curve is found, the index of each minimum value is recorded, and the number of interval frames is calculated. The mode of all interval frames is taken as the cardiac cycle.

[0072] Based on the cardiac cycle, images of all end-systolic phases of the heart and corresponding end-systolic myocardial segmentation results are extracted from the original MCE images.

[0073] Based on the myocardial segmentation results at the end of cardiac systole, the image with the smallest area in the myocardial segmentation results at the end of cardiac systole is selected as a mask, and the corresponding region in the corresponding frame of the original MCE image is extracted.

[0074] After using the myocardial region within the masked image as the target region, the Large Displacement Diffeomorphic Metric Mapping (LDDMM) method is used to register the corresponding regions in the corresponding frames of the extracted original MCE images, ensuring that the points within the myocardial region of each image correspond one-to-one. The registered image sequence is then saved and transferred to the quantitative calculation module for parameter calculation.

[0075] Step S400: Fit a function model based on the pixel values ​​of each point in the registered image sequence, and generate a parameter image based on the registered image sequence and the fitted function model.

[0076] Based on the registered image sequence, a fitting function model is performed.

[0077] Specifically, the grayscale value at each point in the myocardial region of each image in the registered image sequence is extracted, and a nonlinear Levenberg-Marquardt algorithm is used to fit a function model for each point:

[0078] I(t)=A(1-e -βt )

[0079] Where I is the image grayscale value, A is the peak contrast intensity, representing myocardial blood volume, and β is the contrast replenishment rate, representing the average blood flow velocity. After determining the parameters A and β at each point, MBF = A * β(MBF - Myocardial blood flow) can be calculated, representing myocardial blood flow.

[0080] Based on the fitted function model, plot the time-intensity curve for each point. Figure 8 ).

[0081] Based on the parameter values ​​at each point, a parametric image of the myocardial region is drawn. Figure 9 ).

[0082] In one embodiment, a fully automated quantitative analysis system for myocardial acoustic contrast imaging is provided, comprising:

[0083] Image acquisition module, used to acquire myocardial acoustic contrast images;

[0084] The image segmentation module is used to segment the acquired myocardial acoustic contrast images to obtain myocardial segmentation images;

[0085] The image processing module is used to calculate the cardiac chamber area based on the myocardial segmentation image, automatically extract the myocardial segmentation results at the end of cardiac systole by utilizing the periodic change characteristics of the cardiac chamber area, and use the extracted end-systole segmentation results as a mask to extract the corresponding position of each image as the target region for registration, so as to generate a one-to-one correspondence of registered image sequence.

[0086] The quantitative calculation module is used to fit a function model based on the pixel values ​​of each point in the registered image sequence, and to generate a parametric image based on the registered image sequence and the fitted function model. Figure 10 ).

[0087] The fully automated myocardial acoustic contrast quantitative analysis system of this embodiment works on the same principle and process as the fully automated myocardial acoustic contrast quantitative analysis method described above, and will not be repeated here.

[0088] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered to be within the scope of this specification. When technical features of different embodiments are embodied in the same drawing, it can be regarded as the drawing also disclosing examples of combinations of the various embodiments involved.

[0089] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A fully automated quantitative analysis method for myocardial sonographic contrast, characterized in that, The method comprises the following steps: acquiring myocardial contrast echocardiography images; segmenting the acquired myocardial contrast echocardiography images to obtain myocardial segmentation images; calculating the heart cavity area according to the myocardial segmentation images, automatically extracting the myocardial segmentation result at the end of cardiac systole by using the characteristic of periodic change of the heart cavity area, and using the extracted myocardial segmentation result at the end of cardiac systole as a mask to extract the corresponding position of each image as a target region for registration to generate a point-to-point corresponding registered image sequence, specifically comprising: for each myocardial segmentation image, calculating a convex hull image of the myocardium; extracting the myocardial segmentation result at the end of cardiac systole by using the convex hull region of the myocardium of each image; using the myocardial segmentation result at the end of cardiac systole as a target image to register the remaining myocardial segmentation results to generate a registered image sequence; subtracting the myocardial segmentation image from the convex hull image of the myocardium to obtain an initial heart cavity region image; processing to obtain an ideal heart cavity region image by using morphological filtering; calculating the heart cavity area by using each heart cavity region image, recording and generating a heart cavity area change curve; finding the minimum value in the heart cavity area change curve, recording the index of each minimum value and calculating the interval frame number, and taking the mode of all interval frame numbers as the cardiac cycle; extracting all end-systolic images in the myocardial segmentation result image sequence and all end-systolic images in the original MCE image sequence through the cardiac cycle; using the image with the smallest heart cavity area in the end-systolic segmentation image sequence as a mask to extract the images in the mask region of all images in the end-systolic original image sequence to generate a mask image sequence; using the mask image sequence as a target region and using the large deformation differential homeomorphism method to register all images of the end-systolic segmentation image sequence to ensure that the pixel points in the myocardial region of each image correspond one by one, and saving and inputting the registered image sequence into a quantitative calculation module for parameter calculation; fitting a function model according to the pixel values of each point in the registered image sequence, and generating a parameter image according to the registered image sequence and the fitted function model.

2. The fully automated myocardial sonographic contrast quantification method according to claim 1, characterized in that, The method further comprises the following steps: based on the acquired myocardial contrast echocardiography images, a plurality of myocardial contrast echocardiography images are collected to form an initial image sequence; the initial image sequence is input into a deep neural network model to obtain myocardial segmentation images; the deep neural network model is trained based on a U-Net model; the training path of the deep neural network model comprises a contraction path and an expansion path.

3. The fully automated myocardial sonographic contrast quantification method according to claim 2, characterized in that, The contraction path comprises 9 encoding blocks, each encoding block comprising two 3x3 convolution layers, each convolution layer being followed by a batch normalization layer and a ReLU activation layer, and a 2x2 maximum pooling layer being used after the first 8 encoding blocks for down-sampling; the feature maps of the corresponding blocks in the contraction path and the expansion path are connected in the channel dimension by jumping.

4. The fully automated myocardial sonographic contrast quantification method according to claim 3, characterized in that, The expansion path comprises 8 decoding blocks, each decoding block comprising two 3x3 convolution layers, each convolution layer being followed by a batch normalization layer and a ReLU activation layer, and a 2x2 deconvolution layer being used before each decoding block for up-sampling.

5. The fully automated myocardial sonographic contrast quantification method according to claim 3, characterized in that, In each iteration of the training process, each myocardial ultrasound contrast image in the initial image sequence is transformed N times to obtain an N-image expanded image sequence, each transformation including a random rotation, a random translation, a random shear, and a random scaling performed in sequence, and the initial image sequence and the expanded image sequence together constitute a training image set.

6. The fully automated myocardial sonographic contrast quantification method according to claim 1, characterized in that, The function model is fitted according to the pixel values of each point in the registered image sequence, and a parameter image is generated according to the registered image sequence and the fitted function model, specifically including: extracting the gray value of each point in the registered image sequence; unknown parameters in the function model are calculated by fitting the extracted gray value; a time-intensity curve and a parameter image are drawn according to the mathematical model.

7. The fully automated myocardial sonographic contrast quantification method according to claim 6, characterized in that, The function model is: ; Where I is the image gray value, A is the peak contrast intensity, and β is the contrast replenishment rate.

8. A fully automated myocardial sonographic contrast quantification system, characterized in that It includes: an image acquisition module for acquiring myocardial acoustic contrast images; an image segmentation module for segmenting the acquired myocardial acoustic contrast images to obtain myocardial segmentation images; an image processing module for calculating the heart chamber area according to the myocardial segmentation images, automatically extracting the myocardial segmentation result at the end of cardiac systole using the periodic change feature of the heart chamber area, and extracting the corresponding position of each image as a target region using the extracted myocardial segmentation result at the end of cardiac systole as a mask to perform registration, thereby generating a registered image sequence with one-to-one correspondence between points, specifically for: calculating the convex hull image of the myocardium for each myocardial segmentation image; extracting the myocardial segmentation result at the end of cardiac systole using the convex hull region of the myocardium of each image; using the myocardial segmentation result at the end of cardiac systole as a target image to perform registration on the remaining myocardial segmentation results to generate a registered image sequence; subtracting the myocardial segmentation image from the convex hull image of the myocardium to obtain an initial heart chamber region image; processing using morphological filtering to obtain an ideal heart chamber region image; calculating the heart chamber area using each heart chamber region image, recording and generating a heart chamber area change curve, finding the minimum value in the heart chamber area change curve, recording the index of each minimum value and calculating the interval frame number, and taking the mode of all interval frame numbers as the cardiac cycle; extracting all end-systolic images in the myocardial segmentation result image sequence and all end-systolic images in the original MCE image sequence through the cardiac cycle; using the image with the smallest heart chamber area in the end-systolic segmentation image sequence as a mask to extract the images in the corresponding mask region in all images in the end-systolic original image sequence to generate a mask image sequence; using the mask image sequence as a target region to perform registration on all images in the end-systolic segmentation image sequence using the large deformation differential homeomorphism method to ensure one-to-one correspondence of pixel points in the myocardial region of each image, and saving and transmitting the registered image sequence to a quantitative calculation module for parameter calculation; a quantitative calculation module for fitting a function model according to the pixel values of each point in the registered image sequence, and generating a parameter image according to the registered image sequence and the fitted function model.

Images