Three-dimensional cardiac image segmentation methods, devices, equipment and storage media

Abstract

This invention relates to the field of image segmentation technology, and in particular to a method, apparatus, device, and storage medium for segmenting three-dimensional heart images. The method includes: extracting the intermediate layer region of interest (ROI) of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model; spreading the intermediate layer ROI to the basal and apical layers of the three-dimensional heart image to be segmented to obtain an overall ROI image; and segmenting the overall ROI image using a preset overall segmentation model to obtain a segmentation result for the three-dimensional heart image to be segmented. Because this invention first segments the intermediate layer image of the three-dimensional heart image to be segmented, then spreads the segmentation result to the entire three-dimensional heart image to be segmented, and finally segments the entire three-dimensional heart image to obtain a segmentation result, compared with existing methods, this invention can reduce the image data involved in segmentation and improve the accuracy of the segmentation result.

Description

Technical Field

[0001] This invention relates to the field of image segmentation technology, and in particular to a three-dimensional cardiac image segmentation method, apparatus, device, and storage medium. Background Technology

[0002] The heart is the most vital organ for maintaining human life. It powers blood circulation, pumping blood throughout the body via heartbeats. Heart disease directly impacts a person's health and life. Statistics show that approximately 17.5 million people die from heart disease globally each year, posing a significant threat to human life and safety. Therefore, researching rapid and effective methods for diagnosing heart disease is of paramount importance.

[0003] Currently, because the heart is a constantly functioning organ, its shape changes continuously during movement. Furthermore, cardiac motion parameters play a significant role in describing local abnormalities and early minor lesions. Clinically, the left and right ventricles of the entire three-dimensional heart image at different time phases (different moments in diastole and systole) are generally segmented directly. However, with the significant improvement in the temporal and spatial resolution of imaging equipment, the entire three-dimensional heart image involves a large amount of image data, which makes the existing segmentation methods more difficult and results in lower accuracy.

[0004] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this invention is to provide a three-dimensional heart image segmentation method, apparatus, device, and storage medium, aiming to solve the technical problem of low segmentation accuracy in the prior art when segmenting three-dimensional heart images.

[0006] To achieve the above objectives, the present invention provides a three-dimensional cardiac image segmentation method, the method comprising the following steps:

[0007] The intermediate layer region of interest is extracted from the three-dimensional heart image to be segmented by a preset intermediate layer segmentation model; the intermediate layer region of interest is diffused to the basal and apical layers of the three-dimensional heart image to be segmented to obtain the overall region of interest image;

[0008] The overall region of interest image is segmented using a preset overall segmentation model to obtain the segmentation result of the three-dimensional heart image to be segmented.

[0009] Optionally, the step of extracting the intermediate layer region of interest of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model includes:

[0010] A segmentation mask for the three-dimensional heart image to be segmented is obtained by using a pre-set intermediate layer segmentation model;

[0011] The region of interest in the middle layer of the three-dimensional heart image to be segmented is extracted based on the segmentation mask.

[0012] Optionally, before the step of obtaining the segmentation mask of the three-dimensional heart image to be segmented through a preset intermediate layer segmentation model, the method further includes:

[0013] Organize time-series cardiac images into short-axis images, wherein the time-series images are the original cardiac sequence images;

[0014] The intermediate layer image of the three-dimensional heart image to be segmented is obtained from the short axis image;

[0015] Accordingly, the step of obtaining the segmentation mask of the three-dimensional heart image to be segmented through a preset intermediate layer segmentation model includes:

[0016] The segmentation mask of the intermediate layer image is obtained by using a preset intermediate layer segmentation model.

[0017] Optionally, before the step of extracting the intermediate layer region of interest of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model, the method further includes:

[0018] Acquire a three-dimensional heart image of the sample and label the three-dimensional heart image of the sample;

[0019] The intermediate layer image is obtained from the labeled three-dimensional heart image of the sample, and the initial intermediate layer segmentation model is trained using the intermediate layer image;

[0020] When the training results meet the preset conditions, the initial intermediate layer segmentation model after training will be used as the preset intermediate layer segmentation model.

[0021] Optionally, after the step of using the trained initial intermediate layer segmentation model as the preset intermediate layer segmentation model when the training results meet preset conditions, the method further includes:

[0022] The region of interest is diffused to the basal and apical layers of the three-dimensional cardiac image of the sample to obtain an overall image of the region of interest of the sample.

[0023] The initial global segmentation model is trained using the sample region of interest image, and when the training result meets the preset condition, the trained initial global segmentation model is used as the preset global segmentation model.

[0024] Optionally, before the step of acquiring the sample three-dimensional heart image and labeling the sample three-dimensional heart image, the method further includes:

[0025] The sample time-series cardiac images are organized into sample short-axis images, wherein the sample time-series images are sample three-dimensional cardiac images;

[0026] The minor axis image of the sample is labeled to obtain the labeled minor axis image of the sample;

[0027] Accordingly, the step of acquiring intermediate layer images from labeled 3D cardiac images and training the initial intermediate layer segmentation model using the intermediate layer images includes:

[0028] Obtain the intermediate layer image from the minor axis image of the labeled sample, and train the initial intermediate layer segmentation model using the intermediate layer image.

[0029] Optionally, the preset intermediate layer segmentation model includes: a two-layer convolution module, a downsampling module, an upsampling module, and a one-dimensional convolution module;

[0030] The two-layer convolution module is connected to the downsampling module, and the upsampling module is connected to both the downsampling module and the one-dimensional convolution module.

[0031] The dual-layer convolution module is used to perform convolution processing on the three-dimensional heart image to be segmented to obtain a sample dataset;

[0032] The downsampling module is used to filter the sample dataset to obtain a downsampled dataset;

[0033] The upsampling module is used to amplify the downsampling dataset to obtain an upsampling dataset;

[0034] The one-dimensional convolution module is used to extract features from the upsampled dataset to obtain a segmentation mask.

[0035] Furthermore, to achieve the above objectives, the present invention also proposes a three-dimensional cardiac image segmentation device, the device comprising:

[0036] The region of interest extraction module is used to extract the intermediate region of interest of the three-dimensional heart image to be segmented by using a preset intermediate layer segmentation model;

[0037] The region of interest diffusion module is used to diffuse the intermediate layer region of interest to the basal and apical layers of the three-dimensional heart image to be segmented, thereby obtaining the overall region of interest image.

[0038] The region of interest segmentation module is used to segment the overall region of interest image using a preset overall segmentation model to obtain the segmentation result of the three-dimensional heart image to be segmented.

[0039] Furthermore, to achieve the above objectives, the present invention also proposes a three-dimensional cardiac image segmentation device, the device comprising: a memory, a processor, and a three-dimensional cardiac image segmentation program stored in the memory and executable on the processor, the three-dimensional cardiac image segmentation program being configured to implement the steps of the three-dimensional cardiac image segmentation method as described above.

[0040] Furthermore, to achieve the above objectives, the present invention also proposes a storage medium storing a three-dimensional heart image segmentation program, which, when executed by a processor, implements the steps of the three-dimensional heart image segmentation method described above.

[0041] This invention extracts the region of interest (ROI) from the intermediate layer of a 3D heart image to be segmented using a preset intermediate layer segmentation model; then, it extends this ROI to the basal and apical layers of the 3D heart image to obtain a global ROI image; finally, it segments this global ROI image using a preset global segmentation model to obtain the segmentation result of the 3D heart image to be segmented. Because this invention first segments the intermediate layer image of the 3D heart image using a preset intermediate layer segmentation model, then extends the segmentation result to the entire 3D heart image, and finally segments the entire 3D heart image using a preset global segmentation model to obtain the segmentation result, compared to existing methods that directly segment the entire 3D heart image, it reduces the amount of image data involved in the segmentation process and improves the accuracy of the segmentation result. Attached Figure Description

[0042] Figure 1 This is a schematic diagram of the structure of a three-dimensional cardiac image segmentation device in the hardware operating environment involved in the embodiments of the present invention;

[0043] Figure 2 This is a flowchart illustrating the first embodiment of the three-dimensional cardiac image segmentation method of the present invention;

[0044] Figure 3 This is a flowchart illustrating the second embodiment of the three-dimensional cardiac image segmentation method of the present invention;

[0045] Figure 4 This is a diagram of the pre-set intermediate layer segmentation model structure in the second embodiment of the three-dimensional heart image segmentation method of the present invention;

[0046] Figure 5 This is a structural block diagram of the first embodiment of the three-dimensional cardiac image segmentation device of the present invention.

[0047] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0048] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0049] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of a three-dimensional cardiac image segmentation device in the hardware operating environment involved in the embodiments of the present invention.

[0050] like Figure 1 As shown, the three-dimensional cardiac image segmentation device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen and an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0051] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the three-dimensional cardiac image segmentation device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0052] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a three-dimensional cardiac image segmentation program.

[0053] exist Figure 1 In the illustrated three-dimensional cardiac image segmentation device, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the three-dimensional cardiac image segmentation device of the present invention can be set in the three-dimensional cardiac image segmentation device, and the three-dimensional cardiac image segmentation device calls the three-dimensional cardiac image segmentation program stored in the memory 1005 through the processor 1001 and executes the three-dimensional cardiac image segmentation method provided in the embodiment of the present invention.

[0054] This invention provides a three-dimensional cardiac image segmentation method, with reference to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the three-dimensional cardiac image segmentation method of the present invention.

[0055] In this embodiment, the three-dimensional heart image segmentation method includes the following steps:

[0056] Step S10: Extract the region of interest in the intermediate layer of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model.

[0057] It should be noted that the method in this embodiment can be applied to scenarios involving the segmentation of three-dimensional cardiac images, or other scenarios requiring image segmentation. The executing entity in this embodiment can be a three-dimensional cardiac image segmentation device with data processing, network communication, and program execution functions, such as a computer, mobile terminal, or other devices capable of performing similar or identical functions. This embodiment and the following embodiments will be specifically described using the aforementioned three-dimensional cardiac image segmentation device (hereinafter referred to as the device).

[0058] It is understandable that the aforementioned three-dimensional heart image to be segmented can be a three-dimensional image of the patient's heart obtained by a magnetic resonance imaging device, or a three-dimensional image obtained by other devices.

[0059] It should be understood that the above-mentioned three-dimensional image to be segmented can be divided into the basal layer, the intermediate layer and the apical layer. Considering that the basal layer image contains the left ventricular outflow tract and the apical layer blood pool is too small, both of which are not conducive to segmentation, the above-mentioned device selects one image from the intermediate layer of the three-dimensional cardiac image to be segmented as the intermediate layer image for segmentation. The specific location of the intermediate layer image can be set according to the actual situation.

[0060] It should be emphasized that the aforementioned preset intermediate layer segmentation model can be obtained by training a convolutional neural network model.

[0061] Furthermore, in order to train a preset intermediate layer segmentation model, before step S10, the method further includes: acquiring a sample three-dimensional heart image and labeling the sample three-dimensional heart image; acquiring an intermediate layer image in the labeled sample three-dimensional heart image and training an initial intermediate layer segmentation model using the intermediate layer image; and when the training result meets a preset condition, using the trained initial intermediate layer segmentation model as the preset intermediate layer segmentation model.

[0062] It should be noted that the above-mentioned sample three-dimensional heart image can also be a three-dimensional image of the patient's heart obtained by magnetic resonance imaging equipment, or a three-dimensional image obtained by other equipment. The above-mentioned equipment can mark the left ventricle and right ventricle in the sample three-dimensional heart image with labels. The above-mentioned labels can also be manually marked by experts. This embodiment does not limit this. The above-mentioned labels can be called the gold standard, that is, the above-mentioned marked sample three-dimensional heart image can be considered as a three-dimensional heart image with the gold standard.

[0063] Understandably, the aforementioned intermediate layer image can refer to an intermediate layer image in the sample three-dimensional heart image that contains the outlines of the left and right ventricles. The device can input several intermediate layer images into the initial intermediate layer segmentation model for training, and the specific number of images can be set according to the actual situation.

[0064] It should be understood that the above-mentioned preset condition can be that the loss value of the initial intermediate layer segmentation model is within the preset loss value range. If the loss value of the initial intermediate layer segmentation model is within the preset loss value range, it can be said that the initial intermediate layer segmentation model has been trained. If it is not within the preset loss value range, it can be said that the initial intermediate layer segmentation model has not been trained and needs to be trained again. The above-mentioned preset loss value range can be set by the user according to the actual situation.

[0065] In a specific implementation, the aforementioned device can label sample 3D heart images to obtain sample 3D heart images with gold standards. The initial intermediate layer segmentation model is trained using the intermediate layer images of the sample 3D heart images with gold standards. When the loss value of the initial intermediate layer segmentation model is within a preset loss value range, the trained initial intermediate layer segmentation model is used as the preset intermediate layer segmentation model. The intermediate layer of the 3D heart image to be segmented is then segmented using the preset intermediate layer segmentation model to obtain a segmentation mask. The region of interest in the intermediate layer is then extracted using the segmentation mask.

[0066] Step S20: Expand the intermediate layer region of interest to the basal and apical layers of the three-dimensional heart image to be segmented to obtain the overall region of interest image.

[0067] It should be noted that the above diffusion can be a repeated iterative process, in which the region of interest in the intermediate layer is diffused layer by layer towards the lower and apical layers of the heart to obtain the region of interest in each layer of the image. Finally, the regions of interest in each layer are combined to obtain the region of interest in the entire three-dimensional heart image to be segmented.

[0068] In its implementation, the aforementioned device can diffuse the region of interest from the intermediate layer to the basal and apical layers layer by layer to obtain an image of the entire region of interest.

[0069] Step S30: Segment the overall region of interest image using a preset overall segmentation model to obtain the segmentation result of the three-dimensional heart image to be segmented.

[0070] Understandably, the aforementioned pre-defined overall segmentation model can be obtained through training a convolutional neural network.

[0071] Furthermore, in order to train a preset overall segmentation model, after the step of using the trained initial intermediate layer segmentation model as the preset intermediate layer segmentation model when the training result of the initial intermediate layer segmentation model meets the preset conditions, the method further includes: expanding the region of interest to the basal and apical layers of the sample three-dimensional heart image to obtain an overall sample region of interest image; training the initial overall segmentation model using the sample region of interest image, and using the trained initial overall segmentation model as the preset overall segmentation model when the training result meets the preset conditions.

[0072] It should be noted that the above-mentioned device diffuses the region of interest from the intermediate layer to the basal and apical layers to obtain an image of the entire region of interest.

[0073] It should be understood that the above preset conditions can be consistent with or inconsistent with the preset loss value range of the initial intermediate layer segmentation model. You can set them according to the actual situation. If the loss value of the initial overall segmentation model is within the preset loss value range, it means that the initial overall segmentation model has been trained. If it is not within the preset loss value range, it means that the initial overall segmentation model has not been trained and needs to be trained again.

[0074] In a specific implementation, the aforementioned device can expand the region of interest of the intermediate layer to the basal and apical layers of the three-dimensional heart image of the sample to obtain the overall region of interest of the sample. The initial overall segmentation model is trained using the overall sample region of interest image. When the loss value of the initial overall segmentation model is within a preset loss range, the initial overall segmentation model trained at this time is used as the preset overall segmentation model.

[0075] In this embodiment, the device described above can label sample 3D heart images to obtain sample 3D heart images with gold standards. An initial intermediate layer segmentation model is trained using the intermediate layer image of the sample 3D heart image with gold standards. When the loss value of the initial intermediate layer segmentation model is within a preset loss value range, the trained initial intermediate layer segmentation model is used as the preset intermediate layer segmentation model. The region of interest (ROI) of the intermediate layer is then diffused to the basal and apical layers of the sample 3D heart image to obtain the overall sample ROI. The initial overall segmentation model is trained using the overall sample ROI image. When the loss value of the initial overall segmentation model is within a preset loss value range, the trained initial overall segmentation model is used as the preset overall segmentation model. The intermediate layer of the 3D heart image to be segmented is segmented using the preset intermediate layer segmentation model to obtain a segmentation mask. The intermediate layer ROI is extracted based on the segmentation mask and diffused layer by layer to the basal and apical layers to obtain the overall ROI image. Finally, the overall ROI image is segmented using the preset overall segmentation model to obtain the segmentation result of the 3D heart image to be segmented. This embodiment first segments the intermediate layer image of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model, then expands the region of interest to the entire three-dimensional heart image to be segmented based on the segmentation results, and finally segments the entire three-dimensional heart image to be segmented using a preset overall segmentation model to obtain the overall segmentation result. Compared with the existing method of directly segmenting the entire three-dimensional heart image to be segmented, this embodiment first segments the intermediate layer image separately, which can reduce the image data involved in the segmentation and improve the accuracy of the segmentation results.

[0076] refer to Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the three-dimensional cardiac image segmentation method of the present invention.

[0077] Based on the first embodiment described above, considering the continuous operation of the heart and the constant changes in its shape during movement, in order to improve segmentation accuracy, the following step is included before step S10:

[0078] Step S01: Obtain the segmentation mask of the three-dimensional heart image to be segmented by using a preset intermediate layer segmentation model;

[0079] Further, before step S01, the method includes: organizing the time-series heart images into short-axis images, wherein the time-series images are the original heart sequence images; obtaining intermediate layer images of the three-dimensional heart images to be segmented based on the short-axis images; correspondingly, step S01 includes: obtaining the segmentation mask of the intermediate layer images through a preset intermediate layer segmentation model.

[0080] It should be noted that the above-mentioned original cardiac sequence images may include diastolic cardiac images and systolic cardiac images. The corresponding cardiac images are different at different times. The above-mentioned device can organize the original cardiac sequence into long axis images and short axis images according to the temporal sequence. Since short axis images can better analyze cardiac diseases, short axis images are used in this embodiment.

[0081] Understandably, the aforementioned device can select a short-axis image at any time phase, select an intermediate layer image from the short-axis image at that time phase, and obtain a segmentation mask through a preset intermediate layer segmentation model. The aforementioned segmentation mask can be an image after the intermediate layer image has been masked by the preset intermediate layer segmentation model, except for the left and right ventricles, which facilitates the extraction of the region of interest.

[0082] Step S02: Extract the intermediate layer region of interest of the three-dimensional heart image to be segmented based on the segmentation mask.

[0083] In its specific implementation, the device can organize the original heart sequence images into short-axis images at different time phases, select intermediate layer images from the short-axis images, segment them using a preset intermediate layer segmentation model to obtain a segmentation mask, and then extract the intermediate layer region of interest based on the segmentation mask.

[0084] Furthermore, when training the initial intermediate layer segmentation model, short-axis images can also be used for training to improve segmentation accuracy. Therefore, before the steps of obtaining sample three-dimensional heart images and labeling the sample three-dimensional heart images, the method further includes: organizing sample time-series heart images into sample short-axis images, wherein the sample time-series images are sample three-dimensional heart images; labeling the sample short-axis images to obtain labeled sample short-axis images; correspondingly, the step of obtaining intermediate layer images in the labeled sample three-dimensional heart images and training the initial intermediate layer segmentation model using the intermediate layer images includes: obtaining intermediate layer images in the labeled sample short-axis images and training the initial intermediate layer segmentation model using the intermediate layer images.

[0085] It should be noted that the above-mentioned three-dimensional cardiac images can also include diastolic cardiac images and systolic cardiac images. The above-mentioned device can organize the sample three-dimensional cardiac images into long axis images and short axis images in the order of time phases. The short axis image is also the above-mentioned sample short axis image.

[0086] In a specific implementation, the device can organize the sample three-dimensional heart image into a sample minor axis image, label the intermediate layer image of the sample minor axis image to obtain the labeled intermediate layer image in the sample minor axis, and then train the initial intermediate layer segmentation model using the intermediate layer image.

[0087] Furthermore, since both methods segment the region of interest, the initial intermediate layer segmentation model and the initial overall segmentation model can be obtained using the same convolutional neural network model through different training methods. For ease of understanding the specific segmentation process, only the preset intermediate layer segmentation model is used as an example. Figure 4 , Figure 4 This is a structural diagram of a preset intermediate layer segmentation model in the second embodiment of the three-dimensional heart image segmentation method of the present invention.

[0088] like Figure 4 As shown, the preset intermediate layer segmentation model includes: a two-layer convolution module, a downsampling module, an upsampling module, and a one-dimensional convolution module; wherein, the two-layer convolution module is connected to the downsampling module, and the upsampling module is connected to both the downsampling module and the one-dimensional convolution module; the two-layer convolution module is used to perform convolution processing on the three-dimensional heart image to be segmented to obtain a sample dataset; the downsampling module is used to filter the sample dataset to obtain a downsampled dataset; the upsampling module is used to enlarge the downsampled dataset to obtain an upsampled dataset; and the one-dimensional convolution module is used to extract features from the upsampled dataset to obtain a segmentation mask.

[0089] It should be noted that the above-mentioned double-layer convolution module includes one double-layer convolution, which in turn includes ResNet Block, BN, ReLU, ResNet Block, BN, and ReLU; the downsampling module includes four downsampling modules, which in turn include MaxPool and DoubleConv; and the upsampling module includes four upsampling modules, which in turn include Upsample and DoubleConv.

[0090] Understandably, the aforementioned initial intermediate layer segmentation model is trained to obtain the aforementioned preset intermediate layer segmentation model. When the preset intermediate layer segmentation model performs segmentation, the intermediate layer image of the 3D heart image to be segmented is taken as input, and sequentially processed by two convolutional layers (3×3,64), followed by downsampling (3×3,128), downsampling (3×3,256), downsampling (3×3,512), and downsampling (3×3,512) for filtering, and upsampling (3×3,256), upsampling (3×3,128), and upsampling (3×3,256), upsampling (3×3,128), and upsampling (3×3,256), upsampling (3×3,128), and upsampling (3×3,128 ... The upsampling (3×3,64) and downsampling (3×3,64) are amplified, and finally, a one-dimensional convolution (1×1,256) is used for feature extraction to obtain the segmentation mask. The double convolution (3×3,64) and upsampling (3×3,64), the downsampling (3×3,128) and upsampling (3×3,64), the downsampling (3×3,256) and upsampling (3×3,128), and the downsampling (3×3,512) and upsampling (3×3,256) are all connected by residual modules for forward propagation. For details, please refer to [reference needed]. Figure 4 .

[0091] It should be understood that during training, intermediate layer images with labeled left and right ventricular contours are used as input, such as... Figure 4 As shown, the first result is obtained by sequentially performing a double convolution (3×3,64), downsampling (3×3,128), downsampling (3×3,256), downsampling (3×3,512), downsampling (3×3,512), upsampling (3×3,256), upsampling (3×3,128), upsampling (3×3,64), and upsampling (3×3,64), followed by the aforementioned one-dimensional convolution (1×1,256). Specifically, the relationships between the double convolution (3×3,64) and upsampling (3×3,64), between downsampling (3×3,128) and upsampling (3×3,64), between downsampling (3×3,256) and upsampling (3×3,128), and between downsampling (3×3,128) and upsampling (3×3,64), and between downsampling (3×3,256) and upsampling (3×3,128), and between downsampling (3×3,64) are as follows: Both (3×3,512) and upsampling (3×3,256) are connected by residual modules for forward propagation; and simultaneously pass through double convolution (3×3,64), downsampling (3×3,128), upsampling (3×3,128), double convolution (3×3,64), and one-dimensional convolution (1×1,256) to obtain the second result; and simultaneously pass through double convolution (3×3,64), downsampling (3×3,128), downsampling (3×3,256), upsampling (3×3,256), double convolution (3×3,128), upsampling (3×3,128), double convolution (3×3,64), and one-dimensional convolution (1×1,256) to obtain the third result.

[0092] Finally, backpropagation is performed using the labels. This involves comparing the aforementioned gold standard with the first result to adjust the first loss function (Loss1), comparing it with the second result to adjust the second loss function (Loss2), and comparing it with the third result to adjust the third loss function (Loss3). The first loss function (Loss1), the second loss function (Loss2), and the third loss function (Loss3) together constitute the total loss function (Loss) to prevent gradient vanishing and to adjust the training weights in a timely manner. The relationships between them are as follows:

[0093]

[0094] k1, k2, and k3 can all be obtained through adaptive learning by the network.

[0095] In this embodiment, the device described above can organize the original heart sequence image into short-axis images at different time phases. Intermediate layer images are selected from the short-axis images and segmented using a preset intermediate layer segmentation model to obtain a segmentation mask. Then, the intermediate layer region of interest is extracted based on the segmentation mask. Simultaneously, sample 3D heart images can be organized into sample short-axis images. The intermediate layer images of the sample short-axis images are labeled to obtain the labeled intermediate layer images in the sample short-axis. The initial intermediate layer segmentation model is then trained using these intermediate layer images. Compared to existing methods that directly segment and train the entire 3D heart image, this embodiment uses short-axis images, which can prevent the continuous movement of the heart from affecting the segmentation and improve segmentation accuracy.

[0096] Furthermore, this embodiment of the invention also proposes a storage medium storing a three-dimensional cardiac image segmentation program, which, when executed by a processor, implements the steps of the three-dimensional cardiac image segmentation method described above.

[0097] In addition, refer to Figure 5 , Figure 5 This is a structural block diagram of a first embodiment of the three-dimensional cardiac image segmentation device of the present invention. The present invention also proposes a three-dimensional cardiac image segmentation device, which includes:

[0098] The region of interest extraction module 501 is used to extract the intermediate region of interest of the three-dimensional heart image to be segmented by using a preset intermediate layer segmentation model;

[0099] The region of interest diffusion module 502 is used to diffuse the intermediate layer region of interest to the basal and apical layers of the three-dimensional heart image to be segmented, so as to obtain the overall region of interest image.

[0100] The region of interest segmentation module 503 is used to segment the overall region of interest image using a preset overall segmentation model to obtain the segmentation result of the three-dimensional heart image to be segmented.

[0101] In this embodiment, the device described above can label sample 3D heart images to obtain sample 3D heart images with gold standards. An initial intermediate layer segmentation model is trained using the intermediate layer image of the sample 3D heart image with gold standards. When the loss value of the initial intermediate layer segmentation model is within a preset loss value range, the trained initial intermediate layer segmentation model is used as the preset intermediate layer segmentation model. The region of interest (ROI) of the intermediate layer is then diffused to the basal and apical layers of the sample 3D heart image to obtain the overall sample ROI. The initial overall segmentation model is trained using the overall sample ROI image. When the loss value of the initial overall segmentation model is within a preset loss value range, the trained initial overall segmentation model is used as the preset overall segmentation model. The intermediate layer of the 3D heart image to be segmented is segmented using the preset intermediate layer segmentation model to obtain a segmentation mask. The intermediate layer ROI is extracted based on the segmentation mask and diffused layer by layer to the basal and apical layers to obtain the overall ROI image. Finally, the overall ROI image is segmented using the preset overall segmentation model to obtain the segmentation result of the 3D heart image to be segmented. This embodiment first segments the intermediate layer image of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model, then expands the region of interest to the entire three-dimensional heart image to be segmented based on the segmentation results, and finally segments the entire three-dimensional heart image to be segmented using a preset overall segmentation model to obtain the overall segmentation result. Compared with the existing method of directly segmenting the entire three-dimensional heart image to be segmented, this embodiment first segments the intermediate layer image separately, which can reduce the image data involved in the segmentation and improve the accuracy of the segmentation results.

[0102] Other embodiments or specific implementations of the three-dimensional cardiac image segmentation device of the present invention can be referred to the above-described method embodiments, and will not be repeated here.

[0103] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0104] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0106] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A three-dimensional cardiac image segmentation method, characterized in that, The three-dimensional cardiac image segmentation method includes the following steps: The intermediate region of interest in the three-dimensional heart image to be segmented is extracted using a preset intermediate layer segmentation model; The region of interest in the intermediate layer is diffused to the basal and apical layers of the three-dimensional heart image to be segmented, thereby obtaining the overall region of interest image; The overall region of interest image is segmented by a preset overall segmentation model to obtain the segmentation result of the three-dimensional heart image to be segmented. Both the preset overall segmentation model and the preset intermediate layer segmentation model are constructed based on the ResNet network. The step of extracting the intermediate layer region of interest of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model includes: A segmentation mask for the three-dimensional heart image to be segmented is obtained by using a preset intermediate layer segmentation model. The segmentation mask is an image after occluding the regions of the intermediate layer image other than the left and right ventricles. The region of interest in the middle layer of the three-dimensional heart image to be segmented is extracted based on the segmentation mask.

2. The three-dimensional cardiac image segmentation method as described in claim 1, characterized in that, Before the step of obtaining the segmentation mask of the three-dimensional heart image to be segmented through a preset intermediate layer segmentation model, the method further includes: Organize time-series cardiac images into short-axis images, wherein the time-series cardiac images are the original cardiac sequence images; The intermediate layer image of the three-dimensional heart image to be segmented is obtained from the short axis image; Accordingly, the step of obtaining the segmentation mask of the three-dimensional heart image to be segmented through a preset intermediate layer segmentation model includes: The segmentation mask of the intermediate layer image is obtained by using a preset intermediate layer segmentation model.

3. The three-dimensional cardiac image segmentation method as described in claim 1 or 2, characterized in that, Before the step of extracting the intermediate layer region of interest of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model, the method further includes: Acquire a three-dimensional heart image of the sample and label the three-dimensional heart image of the sample; The intermediate layer image is obtained from the labeled three-dimensional heart image of the sample, and the initial intermediate layer segmentation model is trained using the intermediate layer image; When the training results meet the preset conditions, the initial intermediate layer segmentation model after training will be used as the preset intermediate layer segmentation model.

4. The three-dimensional cardiac image segmentation method as described in claim 3, characterized in that, After the step of using the trained initial intermediate layer segmentation model as the preset intermediate layer segmentation model when the training results meet the preset conditions, the method further includes: The region of interest is diffused to the basal and apical layers of the three-dimensional cardiac image of the sample to obtain an overall image of the region of interest of the sample. The initial global segmentation model is trained using the sample region of interest image, and when the training result meets the preset condition, the trained initial global segmentation model is used as the preset global segmentation model.

5. The three-dimensional cardiac image segmentation method as described in claim 4, characterized in that, Before the step of acquiring and labeling the sample three-dimensional heart image, the method further includes: The sample time-series cardiac images are organized into sample short-axis images, wherein the sample time-series cardiac images are sample three-dimensional cardiac images; The minor axis image of the sample is labeled to obtain the labeled minor axis image of the sample; Accordingly, the step of acquiring intermediate layer images from labeled 3D cardiac images and training the initial intermediate layer segmentation model using the intermediate layer images includes: Obtain the intermediate layer image from the minor axis image of the labeled sample, and train the initial intermediate layer segmentation model using the intermediate layer image.

6. The three-dimensional cardiac image segmentation method as described in claim 2, characterized in that, The preset intermediate layer segmentation model includes: a two-layer convolution module, a downsampling module, an upsampling module, and a one-dimensional convolution module; The two-layer convolution module is connected to the downsampling module, and the upsampling module is connected to both the downsampling module and the one-dimensional convolution module. The dual-layer convolution module is used to perform convolution processing on the three-dimensional heart image to be segmented to obtain a sample dataset; The downsampling module is used to filter the sample dataset to obtain a downsampled dataset; The upsampling module is used to amplify the downsampling dataset to obtain an upsampling dataset; The one-dimensional convolution module is used to extract features from the upsampled dataset to obtain a segmentation mask.

7. A three-dimensional cardiac image segmentation device, characterized in that, The device includes: The region of interest extraction module is used to extract the intermediate region of interest of the three-dimensional heart image to be segmented using a preset intermediate layer segmentation model; The region of interest diffusion module is used to diffuse the intermediate layer region of interest to the basal and apical layers of the three-dimensional heart image to be segmented, thereby obtaining the overall region of interest image. The region of interest segmentation module is used to segment the overall region of interest image using a preset overall segmentation model to obtain the segmentation result of the three-dimensional heart image to be segmented. Both the preset overall segmentation model and the preset intermediate layer segmentation model are constructed based on the ResNet network. The region of interest extraction module is further configured to obtain a segmentation mask of the three-dimensional heart image to be segmented by a preset intermediate layer segmentation model. The segmentation mask is an image after occluding the regions of the intermediate layer image other than the left and right ventricles. The intermediate layer region of interest of the three-dimensional heart image to be segmented is extracted according to the segmentation mask.

8. A three-dimensional cardiac image segmentation device, characterized in that, The device includes: a memory, a processor, and a three-dimensional cardiac image segmentation program stored in the memory and executable on the processor, the three-dimensional cardiac image segmentation program being configured to implement the steps of the three-dimensional cardiac image segmentation method as claimed in any one of claims 1 to 6.

9. A storage medium, characterized in that, The storage medium stores a three-dimensional cardiac image segmentation program, which, when executed by a processor, implements the steps of the three-dimensional cardiac image segmentation method as described in any one of claims 1 to 6.

Images