A processing method and device for measuring characteristic size of m-mode echocardiogram

Abstract

Embodiments of the present application relate to a processing method and device for measuring characteristic sizes of M-mode echocardiogram, the method comprising: receiving an M-mode echocardiogram as a first image; performing binary mask extraction on image regions corresponding to motion state change curves of a right ventricular anterior wall, a right ventricle, an interventricular septum, a left ventricle and a left ventricular posterior wall on the first image based on a preset image instance segmentation model; performing characteristic size change curve fitting on each mask; identifying diastolic and systolic key points according to a left ventricular characteristic size curve; confirming a diastolic characteristic size set according to a diastolic key point sequence and all characteristic size curves; confirming a systolic characteristic size set according to a systolic key point sequence and all characteristic size curves; and outputting the diastolic and systolic characteristic size sets as measurement results. Through the present application, the measurement efficiency of M-mode echocardiogram can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a processing method and apparatus for measuring characteristic dimensions of M-mode echocardiography. Background Technology

[0002] M-mode echocardiography displays the changes in the motion of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall over time. Analysis of M-mode echocardiography yields two sets of characteristic cardiac dimensions at two typical time points (diastole and systole): a diastolic characteristic dimension set and a systolic characteristic dimension set. Each set consists of five characteristic dimensions at the corresponding time point: right ventricular anterior wall thickness, right ventricular diameter, interventricular septum thickness, left ventricular diameter, and left ventricular posterior wall thickness. Typically, during this analysis, analysts mark typical time points on the M-mode echocardiography based on their experience and manually or semi-manually measure the corresponding diastolic and systolic characteristic dimension sets on the various motion change curves. Clearly, this conventional method of measuring characteristic dimensions in M-mode echocardiography relies mainly on manual labor. Due to human factors, this conventional method cannot achieve high measurement efficiency or guarantee stable measurement quality. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by providing a method, apparatus, electronic device, and computer-readable storage medium for measuring the feature dimensions of M-mode echocardiography. First, a binary mask is extracted from the image regions of the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall in M-mode echocardiography based on an image instance segmentation model (using the Mask R-CNN model by default). Then, based on each binary mask, the feature dimension change curves of the right ventricular anterior wall thickness, right ventricular diameter, interventricular septum thickness, left ventricular diameter, and left ventricular posterior wall thickness are fitted. Next, a peak-finding algorithm is used to identify key points during diastole and systole on the left ventricular feature dimension curves. Then, based on the identified key points during diastole and systole, corresponding dimension data are obtained from each feature dimension change curve to form corresponding diastolic and systolic feature dimension sets. Finally, the obtained diastolic and systolic feature dimension sets are output as the feature dimension measurement results of the M-mode echocardiography. This invention improves the measurement efficiency of M-mode echocardiography while ensuring stable measurement quality.

[0004] To achieve the above objectives, a first aspect of the present invention provides a method for measuring characteristic dimensions of M-mode echocardiography, the method comprising:

[0005] Receive an M-mode echocardiogram as the corresponding first image;

[0006] Based on a preset image instance segmentation model, the image regions of the first image corresponding to the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle and left ventricular posterior wall are processed by binary mask extraction to generate the corresponding first right ventricular anterior wall mask, first right ventricle mask, first interventricular septum mask, first left ventricle mask and first left ventricular posterior wall mask.

[0007] The first right ventricular anterior wall mask, the first right ventricle mask, the first interventricular septum mask, the first left ventricle mask, and the first left ventricular posterior wall mask are respectively subjected to feature size change curve fitting processing to generate corresponding first right ventricular anterior wall feature size curve, first right ventricle feature size curve, first interventricular septum feature size curve, first left ventricle feature size curve, and first left ventricular posterior wall feature size curve;

[0008] Based on the first left ventricular characteristic size curve, key points during diastole and systole are identified to generate corresponding first diastolic key point sequences and first systolic key point sequences; the first diastolic key point sequence includes the horizontal coordinates of multiple first diastolic key point pixels; the first systolic key point sequence includes the horizontal coordinates of multiple first systolic key point pixels.

[0009] Based on the first diastolic key point sequence and all feature size curves, the diastolic feature size set is confirmed and a corresponding first diastolic feature size set is generated.

[0010] Based on the first contraction phase key point sequence and all feature size curves, the contraction phase feature size set confirmation process is performed to generate the corresponding first contraction phase feature size set;

[0011] The first set of diastolic feature dimensions and the first set of systolic feature dimensions are output as the feature dimension measurement results of the first image.

[0012] Preferably, the first image includes an image region of the motion state change curve corresponding to the anterior wall of the right ventricle, an image region of the motion state change curve corresponding to the right ventricle, an image region of the motion state change curve corresponding to the interventricular septum, an image region of the motion state change curve corresponding to the left ventricle, and an image region of the motion state change curve corresponding to the posterior wall of the left ventricle; and the horizontal time axes of the image regions of the motion state change curve corresponding to the anterior wall of the right ventricle, the image regions of the motion state change curve corresponding to the right ventricle, the image regions of the motion state change curve corresponding to the interventricular septum, the image regions of the motion state change curve corresponding to the left ventricle, and the image regions of the motion state change curve corresponding to the posterior wall of the left ventricle are aligned.

[0013] Preferably, the network structure of the image instance segmentation model is implemented based on the network structure of the Mask R-CNN model;

[0014] The image instance segmentation model includes a first feature extraction backbone network, a first feature pyramid network, a first region proposal network, a first region of interest alignment network, a first convolutional network, a first fully connected network, a second fully connected network, a first mask extraction network, and a first fusion module. The output of the first feature extraction backbone network is connected to the input of the first feature pyramid network. The output of the first feature pyramid network is connected to the input of the first region proposal network. The output of the first region proposal network is connected to the input of the first region of interest alignment network. The output of the first region of interest alignment network is connected to the input of the first convolutional network. The output of the first convolutional network is connected to the inputs of the first fully connected network, the second fully connected network, and the first mask extraction network, respectively. The input of the first fusion module is connected to the outputs of the first fully connected network, the second fully connected network, and the first mask extraction network, respectively. The first feature extraction backbone network is defaulted to a ResNet-50 network.

[0015] The first feature extraction backbone network is used to perform feature extraction processing on the input image of the model to generate five feature maps of different sizes, namely the first feature map, the second feature map, the third feature map, the fourth feature map, and the fifth feature map; the (height, width) of the input image is (H, W), then the (height, width) of the corresponding first, second, third, fourth, and fifth feature maps are (H / 2, W / 2), (H / 4, W / 4), (H / 8, W / 8), (H / 16, W / 16), and (H / 32, W / 32), respectively;

[0016] The first feature pyramid network is used to upsample the input fifth feature map to obtain a first upsampled feature map with the same size as the fourth feature map, and to fuse the first upsampled feature map and the fourth feature map to obtain a corresponding first fused feature map; then, it upsamples the first fused feature map to obtain a second upsampled feature map with the same size as the third feature map, and to fuse the second upsampled feature map and the third feature map to obtain a corresponding second fused feature map; then, it upsamples the second fused feature map to obtain a third upsampled feature map with the same size as the second feature map, and to fuse the third upsampled feature map and the second feature map to obtain a corresponding third fused feature map; finally, it upsamples the third fused feature map to obtain a fourth upsampled feature map with the same size as the first feature map, and to fuse the fourth upsampled feature map and the first feature map to obtain a corresponding fourth fused feature map;

[0017] The first region proposal network is used to perform image region pre-identification processing on the motion state change curve of the fourth fused feature map to obtain multiple first region recognition boxes;

[0018] The first region of interest alignment network is used to perform recognition box fusion alignment processing on all first region recognition boxes to obtain multiple first fused recognition boxes; and extracts the sub-feature map covered by each first fused recognition box in the fourth fused feature map as the corresponding first recognition box feature map; and takes the offset position of each first fused recognition box in the fourth fused feature map as the corresponding first recognition box feature map offset position.

[0019] The first convolutional network is used to perform feature extraction processing on each of the first recognition box feature maps to generate corresponding second recognition box feature maps; the height and width of the second recognition box feature map are the same as those of the first recognition box feature maps. Figure 1 To;

[0020] The first fully connected network is used to classify and predict the feature map type of each second recognition box feature map and use the prediction result as the corresponding first mask map type; the first mask map type includes the right ventricular anterior wall motion state change curve type, the right ventricular motion state change curve type, the interventricular septum motion state change curve type, the left ventricular motion state change curve type, and the left ventricular posterior wall motion state change curve type.

[0021] The second fully connected network is used to perform regression prediction on the offset position and size of each second recognition box feature map in the model input image based on the offset position of the first recognition box feature map corresponding to each second recognition box feature map and the size correspondence between the fourth fused feature map and the model input image, and use the prediction result as the corresponding first mask map offset position and first mask map size;

[0022] The first mask extraction network is used to perform binary mask transformation on the feature maps of each second recognition box to generate the corresponding first mask map; the pixel value of each pixel on the first mask map has only two possible values;

[0023] The first fusion module is used to adjust the mask image size of each first mask image according to the corresponding first mask image size to generate a corresponding second mask image; and to form a corresponding first mask image data set by each second mask image, the corresponding first mask image type, the first mask image offset position, and the first mask image size; and to output the obtained first mask image data set as the model output result.

[0024] Furthermore, the offset position and size of the first mask of the second mask, which is of the type of right ventricular anterior wall motion state change curve, are matched with the region offset position and region size of the image region of the motion state change curve corresponding to the right ventricular anterior wall on the first image.

[0025] The offset position and size of the first mask of the second mask, which is of the type of right ventricular motion state change curve, match the region offset position and region size of the image region of the motion state change curve corresponding to the right ventricle on the first image;

[0026] The offset position and size of the first mask of the second mask, which is of the type of ventricular septal motion state change curve, match the region offset position and region size of the image region of the motion state change curve corresponding to the ventricular septum on the first image;

[0027] The offset position and size of the first mask of the second mask, which is of the left ventricular motion state change curve type, match the region offset position and region size of the image region of the motion state change curve corresponding to the left ventricle on the first image;

[0028] The offset position and size of the first mask of the second mask, which is of the left ventricular posterior wall motion state change curve type, match the region offset position and region size of the image region of the motion state change curve corresponding to the left ventricular posterior wall on the first image.

[0029] Furthermore, the image instance segmentation model is trained using a supervised training method; the model loss function used during the training process is a preset first model loss function L. all L all =λ1*L CLS +λ2*L OFF +λ3*L MASK Among them, the classification loss function L CLS Used to calculate the loss based on the output of the first fully connected network and the corresponding preset ground truth training labels; offset loss function L OFF Used to calculate the loss based on the output of the second fully connected network and the corresponding preset ground truth training labels; Mask loss function L MASK Used to extract the network's output based on the first mask image and calculate the loss using the corresponding preset ground truth training labels; λ1, λ2, and λ3 are preset weighting coefficients.

[0030] Preferably, the step of extracting binary mask images of the image regions in the first image corresponding to the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall based on a preset image instance segmentation model to generate corresponding first right ventricular anterior wall mask, first right ventricle mask, first interventricular septum mask, first left ventricle mask, and first left ventricular posterior wall mask specifically includes:

[0031] The first image is input into the image instance segmentation model for processing to obtain multiple sets of first mask image data; the first mask image data set includes the second mask image, the first mask image type, the first mask image offset position, and the first mask image size;

[0032] The second mask, whose first mask image type is the right ventricular anterior wall motion state change curve, is used as the corresponding first right ventricular anterior wall mask image; the second mask, whose first mask image type is the right ventricular motion state change curve, is used as the corresponding first right ventricular mask image; the second mask, whose first mask image type is the interventricular septum motion state change curve, is used as the corresponding first interventricular septum mask image; the second mask, whose first mask image type is the left ventricular motion state change curve, is used as the corresponding first left ventricular mask image; and the second mask, whose first mask image type is the left ventricular posterior wall motion state change curve, is used as the corresponding first left ventricular posterior wall mask image.

[0033] Preferably, the step of performing characteristic size change curve fitting processing on the first right ventricular anterior wall mask image, the first right ventricular mask image, the first interventricular septum mask image, the first left ventricular mask image, and the first left ventricular posterior wall mask image to generate corresponding first right ventricular anterior wall characteristic size curve, first right ventricular characteristic size curve, first interventricular septum characteristic size curve, first left ventricular characteristic size curve, and first left ventricular posterior wall characteristic size curve specifically includes:

[0034] The first right ventricular anterior wall mask, the first right ventricular mask, the first interventricular septum mask, the first left ventricular mask, and the first left ventricular posterior wall mask are respectively processed by format binarization to generate corresponding first binary image, second binary image, third binary image, fourth binary image, and fifth binary image; the pixel values ​​of each pixel in the first, second, third, fourth, and fifth binary images have only two possible values, namely the preset foreground pixel value and the background pixel value;

[0035] Image denoising processing is performed on the first, second, third, fourth, and fifth binary images respectively;

[0036] The first, second, third, fourth, or fifth binary images after noise reduction are processed by extracting the depth of the upper and lower edges of the foreground region to obtain the corresponding first depth data sequence, second depth data sequence, third depth data sequence, fourth depth data sequence, or fifth depth data sequence; the first, second, third, fourth, or fifth depth data sequence is composed of multiple depth data groups sorted together; the depth data group includes the first pixel horizontal coordinate and the first depth;

[0037] Based on the first, second, third, fourth, or fifth depth data sequence, perform corresponding depth-pixel abscissa curve fitting processing to generate the corresponding first fitting curve, second fitting curve, third fitting curve, fourth fitting curve, or fifth fitting curve;

[0038] The first fitted curve obtained is used as the corresponding first right ventricular anterior wall characteristic size curve; the second fitted curve obtained is used as the corresponding first right ventricular characteristic size curve; the third fitted curve obtained is used as the corresponding first interventricular septum characteristic size curve; the fourth fitted curve obtained is used as the corresponding first left ventricular characteristic size curve; and the fifth fitted curve obtained is used as the corresponding first left ventricular posterior wall characteristic size curve.

[0039] Furthermore, the step of performing foreground region upper and lower edge depth extraction processing on the first, second, third, fourth, or fifth binary images after noise reduction to obtain corresponding first, second, third, fourth, or fifth depth data sequences specifically includes:

[0040] Use the first, second, third, fourth, or fifth binary image as the corresponding current binary image;

[0041] The current binary image is scanned column by column; each column scan yields a set of pixels, denoted as the current column pixel set, and the x-coordinate of the pixels in the current scanned column on the current binary image is marked as the corresponding current column pixel x-coordinate. Based on the first mask offset position of the mask image corresponding to the current binary image, the pixel x-coordinate corresponding to the current column pixel on the first image is confirmed to obtain the corresponding first pixel x-coordinate. If the number of pixels in the current column pixel set whose pixel value is a foreground pixel value exceeds a preset minimum foreground pixel count threshold, then the pixels in the current column pixel set whose pixel value is a foreground pixel value are extracted to form the corresponding foreground pixel set. The system combines and extracts pixels with minimum and maximum vertical coordinates from the foreground pixel set as the corresponding upper and lower edge pixels. It then calculates the absolute difference between the vertical coordinates of these upper and lower edge pixels on the current binary image to generate the corresponding first upper and lower edge pixel distance. The product of this first upper and lower edge pixel distance and a preset pixel distance-real distance scale is used as the corresponding first depth. The first pixel horizontal coordinate and the first depth form the corresponding depth data group. Finally, after all columns have been scanned, all the obtained depth data are sorted from left to right according to the corresponding first pixel horizontal coordinate to generate the corresponding current depth data sequence.

[0042] Based on the correspondence between the current binary image and the first, second, third, fourth, or fifth binary images, the obtained current depth data sequence is output as the corresponding first, second, third, fourth, or fifth depth data sequence.

[0043] Furthermore, the step of performing corresponding depth-pixel abscissa curve fitting processing based on the first, second, third, fourth, or fifth depth data sequences to generate corresponding first, second, third, fourth, or fifth fitting curves specifically includes:

[0044] The first, second, third, fourth, or fifth depth data sequence is taken as the corresponding current depth data sequence; and a two-dimensional coordinate system space is constructed with the pixel horizontal coordinate as the horizontal axis and the depth as the vertical axis, denoted as the corresponding depth-pixel horizontal coordinate space;

[0045] In the depth-pixel horizontal coordinate space, the first pixel horizontal coordinate and the first depth of each depth data group in the current depth data sequence are used as the corresponding horizontal axis coordinate and vertical axis coordinate of the plotting point to mark the plotting point and generate the corresponding first plotting point; and all the obtained first plotting points are sequentially connected to obtain the corresponding first plotting curve; and the first plotting curve is filtered; and the filtered first plotting curve is used as the corresponding current fitting curve;

[0046] Based on the correspondence between the current depth data sequence and the first, second, third, fourth, or fifth depth data sequences, the obtained current fitting curve is output as the corresponding first, second, third, fourth, or fifth fitting curve.

[0047] Preferably, the step of identifying key points during diastole and systole based on the first left ventricular characteristic size curve to generate corresponding first diastolic key point sequences and first systolic key point sequences specifically includes:

[0048] Based on a preset peak-finding algorithm, the peak points of the first left ventricular characteristic size curve are identified to obtain multiple first peak points; the first pixel abscissa corresponding to each first peak point is used as the corresponding first diastolic key point pixel abscissa; and the obtained first diastolic key point pixel abscissas are sorted from left to right to generate the corresponding first diastolic key point sequence.

[0049] The first left ventricular feature size curve is flipped vertically to generate a corresponding second left ventricular feature size curve; and the peak point is identified on the second left ventricular feature size curve based on the peak finding algorithm to obtain multiple second peak points; the first pixel abscissa corresponding to each second peak point is used as the corresponding first systolic key point pixel abscissa; and all the obtained first systolic key point pixel abscissas are sorted from left to right to generate the corresponding first systolic key point sequence.

[0050] Preferably, the step of generating the corresponding first diastolic feature size set by confirming the diastolic feature size set based on the first diastolic key point sequence and all feature size curves specifically includes:

[0051] The abscissa of each pixel of the first diastolic key point in the first diastolic key point sequence is traversed. During traversal, the abscissa of the currently traversed first diastolic key point pixel is taken as the abscissa of the corresponding current key point pixel. The first depth corresponding to the abscissa of the current key point pixel on the first right ventricular anterior wall feature size curve, the first right ventricular feature size curve, the first interventricular septum feature size curve, the first left ventricular feature size curve, and the first left ventricular posterior wall feature size curve is taken as the corresponding first right ventricular anterior wall thickness, first right ventricular diameter, first interventricular septum thickness, first left ventricular diameter, and first... The thickness of the left ventricular posterior wall is determined; and based on the correspondence between the pixel horizontal coordinates and the time axis of the first image, the time information corresponding to the horizontal coordinate of the current key point pixel is confirmed to generate the corresponding first diastolic time; and the horizontal coordinate of the current key point pixel, the first diastolic time, the first right ventricular anterior wall thickness, the first right ventricular diameter, the first interventricular septum thickness, the first left ventricular diameter, and the first left ventricular posterior wall thickness are combined to form the corresponding first diastolic feature size data set; and at the end of the traversal, all the obtained first diastolic feature size data sets constitute the corresponding first diastolic feature size set.

[0052] Preferably, the step of generating the corresponding first contraction period feature size set by confirming the contraction period feature size set based on the first contraction period key point sequence and all feature size curves specifically includes:

[0053] The abscissa of each pixel of the first systolic key point in the first systolic key point sequence is traversed. During traversal, the abscissa of the currently traversed first systolic key point pixel is taken as the corresponding abscissa of the current key point pixel. The first depth on the first right ventricular anterior wall feature size curve, the first right ventricular feature size curve, the first interventricular septum feature size curve, the first left ventricular feature size curve, and the first left ventricular posterior wall feature size curve, corresponding to the abscissa of the current key point pixel, is taken as the corresponding second right ventricular anterior wall thickness, second right ventricular ventricular diameter, second interventricular septum thickness, second left ventricular diameter, and... The second left ventricular posterior wall thickness; and based on the correspondence between the pixel horizontal coordinates of the first image and the time axis, confirm the time information corresponding to the current key point pixel horizontal coordinate to generate the corresponding first systolic time; and the current key point pixel horizontal coordinates, the first systolic time, the second right ventricular anterior wall thickness, the second right ventricular diameter, the second interventricular septum thickness, the second left ventricular diameter, and the second left ventricular posterior wall thickness form the corresponding first systolic feature size data set; and at the end of the traversal, all the obtained first systolic feature size data sets constitute the corresponding first systolic feature size set.

[0054] A second aspect of the present invention provides an apparatus for implementing the method described in the first aspect above, comprising: a receiving module, an image instance segmentation module, a feature size curve fitting module, a diastolic and systolic key point identification module, a diastolic feature size measurement module, a systolic feature size measurement module, and an output module;

[0055] The receiving module is used to receive an M-mode echocardiogram as the corresponding first image;

[0056] The image instance segmentation module is used to perform binary mask image extraction processing on the image regions of the motion state change curves corresponding to the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle and left ventricular posterior wall on the first image based on a preset image instance segmentation model to generate the corresponding first right ventricular anterior wall mask image, first right ventricle mask image, first interventricular septum mask image, first left ventricle mask image and first left ventricular posterior wall mask image.

[0057] The feature size curve fitting module is used to perform feature size change curve fitting processing on the first right ventricular anterior wall mask, the first right ventricle mask, the first interventricular septum mask, the first left ventricle mask, and the first left ventricular posterior wall mask to generate corresponding first right ventricular anterior wall feature size curve, first right ventricle feature size curve, first interventricular septum feature size curve, first left ventricle feature size curve, and first left ventricular posterior wall feature size curve, respectively.

[0058] The diastolic and systolic key point identification module is used to identify diastolic and systolic key points based on the first left ventricular characteristic size curve to generate corresponding first diastolic key point sequences and first systolic key point sequences; the first diastolic key point sequence includes multiple first diastolic key point pixel abscissas; the first systolic key point sequence includes multiple first systolic key point pixel abscissas.

[0059] The diastolic characteristic size measurement module is used to generate the corresponding first diastolic characteristic size set by performing diastolic characteristic size set confirmation processing based on the first diastolic key point sequence and all characteristic size curves;

[0060] The contraction period feature size measurement module is used to generate a corresponding first contraction period feature size set by performing a contraction period feature size set confirmation process based on the first contraction period key point sequence and all feature size curves.

[0061] The output module is used to output the first diastolic feature size set and the first systolic feature size set as the feature size measurement results of the first image.

[0062] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0063] The processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the method described in the first aspect above;

[0064] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0065] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the method described in the first aspect above.

[0066] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for measuring the feature dimensions of M-mode echocardiography. First, a binary mask is extracted from the image regions of the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall in M-mode echocardiography based on an image instance segmentation model (using the Mask R-CNN model by default). Then, based on each binary mask, the feature dimension change curves of the right ventricular anterior wall thickness, right ventricular diameter, interventricular septum thickness, left ventricular diameter, and left ventricular posterior wall thickness are fitted. Next, a peak-finding algorithm is used to identify key points during diastole and systole on the left ventricular feature dimension curves. Then, based on the identified key points during diastole and systole, corresponding dimension data are obtained from each feature dimension change curve to form corresponding diastolic and systolic feature dimension sets. Finally, the obtained diastolic and systolic feature dimension sets are output as the feature dimension measurement results of the M-mode echocardiography. This invention improves the measurement efficiency of M-mode echocardiography while ensuring the stability of measurement quality. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of a method for measuring characteristic dimensions of M-mode echocardiography provided in Embodiment 1 of the present invention;

[0068] Figure 2a This is a schematic diagram of an M-mode echocardiogram provided in Embodiment 1 of the present invention;

[0069] Figure 2b This is a block diagram of the image instance segmentation model provided in Embodiment 1 of the present invention;

[0070] Figure 3 This is a module structure diagram of a processing device for measuring characteristic dimensions of M-mode echocardiography provided in Embodiment 2 of the present invention;

[0071] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0073] Embodiment 1 of the present invention provides a processing method for measuring characteristic dimensions of M-mode echocardiography, such as... Figure 1The schematic diagram shows a method for measuring characteristic dimensions of M-mode echocardiography provided in Embodiment 1 of the present invention. This method mainly includes the following steps:

[0074] Step 1: Receive the M-mode echocardiogram as the corresponding first image;

[0075] The first image includes image regions corresponding to the motion state change curves of the anterior wall of the right ventricle, the right ventricle, the interventricular septum, the left ventricle, and the posterior wall of the left ventricle; and the horizontal time axes of the image regions corresponding to the motion state change curves of the anterior wall of the right ventricle, the right ventricle, the interventricular septum, the left ventricle, and the posterior wall of the left ventricle are aligned.

[0076] Here, as Figure 2a As shown in the schematic diagram of the M-mode echocardiography provided in Embodiment 1 of the present invention, the received M-mode echocardiography displays the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall over time. The image regions corresponding to the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall are respectively: the image region corresponding to the motion state change curve of the right ventricular anterior wall, the image region corresponding to the motion state change curve of the right ventricle, the image region corresponding to the motion state change curve of the interventricular septum, the image region corresponding to the motion state change curve of the left ventricle, and the image region corresponding to the motion state change curve of the left ventricular posterior wall. These five image regions have the same width and the horizontal time axis is aligned vertically. Under normal circumstances, the order of these five image regions from top to bottom on the M-mode echocardiography is the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall.

[0077] Here, in Embodiment 1 of the present invention, after obtaining the first image, the first image will be processed by the image instance segmentation model in the subsequent step 2. In order to better understand the subsequent steps, the image instance segmentation model will be explained as follows.

[0078] The network structure of the image instance segmentation model in Embodiment 1 of this invention is based on the network structure of the Mask R-CNN model; for a detailed description of the Mask R-CNN model, please refer to the paper "Mask R-CNN", which will not be elaborated further here;

[0079] As can be seen from the network structure of the Mask R-CNN model, the image instance segmentation model in Embodiment 1 of this invention should include a first feature extraction backbone network, a first feature pyramid network (FPN), a first region proposal network (RPN), a first region of interest alignment (ROIA lign) network, a first convolutional neural network (CNN) network, a first fully connected neural network, a second fully connected network, a first mask extraction network, and a first fusion module, as shown below. Figure 2b The block structure diagram of the image instance segmentation model provided in Embodiment 1 of the present invention is shown below;

[0080] As can be seen from the network connections of the Mask R-CNN model, the output of the first feature extraction backbone network is connected to the input of the first feature pyramid network; the output of the first feature pyramid network is connected to the input of the first region proposal network; the output of the first region proposal network is connected to the input of the first region of interest alignment network; the output of the first region of interest alignment network is connected to the input of the first convolutional network; the output of the first convolutional network is connected to the inputs of the first fully connected network, the second fully connected network, and the first mask extraction network, respectively; and the input of the first fusion module is connected to the outputs of the first fully connected network, the second fully connected network, and the first mask extraction network, respectively.

[0081] The image instance segmentation model of Embodiment 1 of this invention is used to extract binary mask images of five specified image regions from an input M-mode echocardiogram. These five specified image regions are: image regions corresponding to the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall. To achieve this objective, the technical implementation of each network / module of the image instance segmentation model of Embodiment 1 of this invention is as follows:

[0082] The first feature extraction backbone network is used to extract features from the input image of the model and generate five feature maps of different sizes, namely the first feature map, the second feature map, the third feature map, the fourth feature map, and the fifth feature map. Here, if the (height, width) of the input image is (H, W), then the (height, width) of the corresponding first, second, third, fourth, and fifth feature maps are (H / 2, W / 2), (H / 4, W / 4), (H / 8, W / 8), (H / 16, W / 16), and (H / 32, W / 32), respectively. The first feature extraction backbone network of Embodiment 1 of this invention is defaulted to the ResNet-50 network.

[0083] The first feature pyramid network is used to upsample the input fifth feature map to obtain a first upsampled feature map with the same size as the fourth feature map, and then fuse the first upsampled feature map with the fourth feature map to obtain a corresponding first fused feature map; then upsample the first fused feature map to obtain a second upsampled feature map with the same size as the third feature map, and then fuse the second upsampled feature map with the third feature map to obtain a corresponding second fused feature map; then upsample the second fused feature map to obtain a third upsampled feature map with the same size as the second feature map, and then fuse the third upsampled feature map with the second feature map to obtain a corresponding third fused feature map; finally upsample the third fused feature map to obtain a fourth upsampled feature map with the same size as the first feature map, and then fuse the fourth upsampled feature map with the first feature map to obtain a corresponding fourth fused feature map; here, the first feature pyramid network of the first embodiment of the present invention performs upsampling and fusion of low-level features and high-level features step by step, thereby obtaining an enhanced feature map with low-level spatial information and high-level semantic information, namely the fourth fused feature map;

[0084] The first region proposal network is used to pre-identify the image regions of motion state change curves on the fourth fused feature map to obtain multiple first region identification boxes. Here, the first region proposal network in Embodiment 1 of the present invention is actually based on the fourth fused feature map to identify the regions of interest (ROIs) that satisfy the five specified image region features (image regions of motion state change curves corresponding to the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall), and marks each identified ROI image region with a bounding box (bbox), so multiple first region identification boxes will be output. It should be noted that for the same specified image region feature (image regions of motion state change curves corresponding to the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, or left ventricular posterior wall), the number of first region identification boxes output by the first region proposal network may be more than one, so the subsequent first region of interest alignment network is needed to fuse and align the multiple first region identification boxes.

[0085] The first region of interest alignment network is used to perform fusion alignment processing on all first region recognition boxes to obtain multiple first fused recognition boxes; and extracts the sub-feature maps covered by each first fused recognition box in the fourth fused feature map as the corresponding first recognition box feature map; and takes the offset position of each first fused recognition box in the fourth fused feature map as the corresponding first recognition box feature map offset position; here, the number of first fused recognition boxes output by the first region of interest alignment network in Embodiment 1 of the present invention is 5 by default, which correspond to five specified image region features (image regions corresponding to the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle and left ventricular posterior wall) respectively;

[0086] The first convolutional network is used to perform feature extraction processing on the feature maps of each first bounding box to generate corresponding second bounding box feature maps; the height and width of the second bounding box feature map are the same as those of the first bounding box feature maps. Figure 1 To;

[0087] The first fully connected network is used to classify and predict the feature map type of each second recognition box feature map and use the prediction result as the corresponding first mask map type. The first mask map type includes the right ventricular anterior wall motion state change curve type, the right ventricular motion state change curve type, the interventricular septum motion state change curve type, the left ventricular motion state change curve type, and the left ventricular posterior wall motion state change curve type. Here, the first fully connected network in Embodiment 1 of the present invention is actually a classification and prediction network. It should be noted that the first fully connected network is specifically used to classify and predict the feature map type of each second recognition box feature map and use the prediction result as the corresponding first mask map type. When determining the corresponding first mask map type, the feature map type of the current second recognition box feature map is first classified and predicted to obtain a classification vector with five classification probabilities. The feature map type corresponding to the highest classification probability in the classification vector is then used as the first mask map type corresponding to the current second recognition box feature map. Here, the feature map types corresponding to the first fully connected network are five types: right ventricular anterior wall motion state change curve type, right ventricular motion state change curve type, interventricular septum motion state change curve type, left ventricular motion state change curve type, and left ventricular posterior wall motion state change curve type. Therefore, each first mask map type also has five types.

[0088] The second fully connected network is used to perform regression prediction on the offset position and size of each second recognition box feature map in the model input image based on the offset position of the first recognition box feature map corresponding to each second recognition box feature map and the size correspondence between the fourth fused feature map and the model input image, and uses the prediction result as the corresponding first mask map offset position and first mask map size; here, the second fully connected network in Embodiment 1 of the present invention is actually a regression prediction network for the global offset position of the mask map, and the obtained first mask map offset position is the offset position on the M-mode echocardiogram input by the model;

[0089] The first mask extraction network is used to perform binary mask transformation on the feature maps of each second recognition box to generate the corresponding first mask map; the pixel value of each pixel on the first mask map has only two possible values; here, the first mask extraction network of Embodiment 1 of the present invention will provide five first mask maps, each first mask map corresponding to a first mask map type and a set of first mask map offset positions and first mask map sizes; and when the first mask extraction network of Embodiment 1 of the present invention outputs five first mask maps, it sets the pixel values ​​of the background pixels of the five masks to the same background pixel value, but sets different colors for the pixel values ​​of the foreground pixels of the five masks;

[0090] The first fusion module is used to adjust the mask image size of each first mask image according to the corresponding first mask image size to generate a corresponding second mask image; and to form a corresponding first mask image data set by each second mask image and the corresponding first mask image type, first mask image offset position and first mask image size; and to output the obtained first mask image data set as the model output result.

[0091] It should be noted that, for a second mask of type 1 (right ventricular anterior wall motion state change curve), the offset position and size of the first mask should match the region offset position and size of the image region corresponding to the motion state change curve of the right ventricular anterior wall on the first image; for a second mask of type 1 (right ventricular motion state change curve), the offset position and size of the first mask should match the region offset position and size of the image region corresponding to the motion state change curve of the right ventricle on the first image; for a second mask of type 1 (interventricular septum .... The mask size should match the region offset position and region size of the image region corresponding to the motion state change curve of the interventricular septum on the first image; the first mask offset position and first mask size of the second mask of the left ventricular motion state change curve type should match the region offset position and region size of the image region corresponding to the motion state change curve of the left ventricle on the first image; the first mask offset position and first mask size of the second mask of the left ventricular posterior wall motion state change curve type should match the region offset position and region size of the image region corresponding to the motion state change curve of the left ventricular posterior wall on the first image.

[0092] Additionally, it should be noted that, in the first embodiment of this invention, when training the image instance segmentation model, the recommended approach from the paper "Mask R-CNN" is to use supervised training; and during the model training process, the first model loss function L is used. all As the model loss function, L all =λ1*L CLS +λ2*L OFF +λ3*L MASK Among them, the classification loss function L CLS Used to calculate the loss based on the output of the first fully connected network and the corresponding preset ground truth training labels; offset loss function L OFF Used to calculate the loss based on the output of the second fully connected network and the corresponding preset ground truth training labels; Mask loss function L MASK Used to extract the network's output from the first mask image and calculate the loss using the corresponding preset ground truth training labels; λ1, λ2, and λ3 are preset weighting coefficients.

[0093] Step 2: Based on the preset image instance segmentation model, perform binary mask image extraction on the image regions of the motion state change curves corresponding to the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle and left ventricular posterior wall on the first image to generate the corresponding first right ventricular anterior wall mask, first right ventricle mask, first interventricular septum mask, first left ventricle mask and first left ventricular posterior wall mask.

[0094] Specifically, this includes: Step 21, processing the first image into the image instance segmentation model to obtain multiple sets of first mask image data;

[0095] The first mask image data set includes a second mask image, a first mask image type, a first mask image offset position, and a first mask image size;

[0096] Here, the processing flow of the image instance segmentation model for the first image is as shown in the technical implementation of the model above, and will not be repeated here.

[0097] Step 22: Use the second mask image of the first mask image type being the right ventricular anterior wall motion state change curve type as the corresponding first right ventricular anterior wall mask image; use the second mask image of the first mask image type being the right ventricular motion state change curve type as the corresponding first right ventricular mask image; use the second mask image of the first mask image type being the interventricular septum motion state change curve type as the corresponding first interventricular septum mask image; use the second mask image of the first mask image type being the left ventricular motion state change curve type as the corresponding first left ventricular mask image; and use the second mask image of the first mask image type being the left ventricular posterior wall motion state change curve type as the corresponding first left ventricular posterior wall mask image.

[0098] Step 3: Perform feature size change curve fitting processing on the mask images of the first right ventricular anterior wall, the first right ventricle, the first interventricular septum, the first left ventricle, and the first left ventricular posterior wall to generate corresponding feature size curves for the first right ventricular anterior wall, the first right ventricle, the first interventricular septum, the first left ventricle, and the first left ventricular posterior wall.

[0099] Specifically, step 31 involves performing format binarization processing on the mask images of the first right ventricular anterior wall, the first right ventricle, the first interventricular septum, the first left ventricle, and the first left ventricular posterior wall to generate corresponding first, second, third, fourth, and fifth binary images.

[0100] Among them, the pixel values ​​of each pixel in the first, second, third, fourth and fifth binary images have only two possible values: the preset foreground pixel value and the background pixel value.

[0101] Here, as mentioned earlier, although the first right ventricular anterior wall mask, the first right ventricular mask, the first interventricular septum mask, the first left ventricular mask, and the first left ventricular posterior wall mask are all binary masks, the colors of the foreground pixels are different. Therefore, the current step is to format the foreground pixels of these five masks to the same color, that is, to uniformly set the pixel values ​​of the foreground pixels to the preset foreground pixel values.

[0102] Step 32: Perform image denoising processing on the first, second, third, fourth, and fifth binary images respectively;

[0103] Here, in this embodiment of the invention, the same processing method is used when performing image denoising on the first, second, third, fourth and fifth binary images. Specifically, the discrete foreground pixels on the background are eliminated first; then, the discrete background pixel noise is eliminated in the connected regions of the foreground pixels; then, the multiple connected regions composed of multiple foreground pixels on the image are scored as a whole according to the region connectivity and region area, and the small connected regions with scores lower than the preset minimum score threshold are deleted.

[0104] Step 33: Extract the depth of the upper and lower edges of the foreground region from the first, second, third, fourth or fifth binary images after noise reduction to obtain the corresponding first depth data sequence, second depth data sequence, third depth data sequence, fourth depth data sequence or fifth depth data sequence;

[0105] The first, second, third, fourth, or fifth depth data sequences are each composed of multiple depth data groups sorted together; each depth data group includes the first pixel's horizontal coordinate and the first depth.

[0106] Specifically, this includes: step 331, using the first, second, third, fourth, or fifth binary image as the corresponding current binary image;

[0107] Step 332: Perform column scanning on the current binary image; for each column scanned, a set of pixels is obtained and denoted as the current column pixel set. The x-coordinate of the pixel in the current scanned column on the current binary image is marked as the x-coordinate of the current column pixel. Based on the first mask offset position of the mask image corresponding to the current binary image, the x-coordinate of the current column pixel is confirmed to obtain the corresponding first pixel x-coordinate. If the number of pixels in the current column pixel set whose pixel value is a foreground pixel value exceeds a preset minimum foreground pixel count threshold, then the pixels in the current column pixel set whose pixel value is a foreground pixel value are extracted to form the corresponding foreground. The system collects a set of pixels and extracts the pixels with the minimum and maximum vertical coordinates from the foreground pixel set as the corresponding upper and lower edge pixels. It then calculates the absolute difference between the vertical coordinates of the upper and lower edge pixels on the current binary image to generate the corresponding first upper and lower edge pixel distances. The product of these first upper and lower edge pixel distances and a preset pixel distance-actual distance scale is used as the corresponding first depth. The system then forms a depth data group based on the first pixel horizontal coordinate and the first depth. Finally, after scanning all columns, the system sorts all the obtained depth data from left to right according to the corresponding first pixel horizontal coordinate to generate the corresponding current depth data sequence.

[0108] Step 333: Based on the correspondence between the current binary image and the first, second, third, fourth, or fifth binary images, output the obtained current depth data sequence as the corresponding first, second, third, fourth, or fifth depth data sequence;

[0109] Step 34: Perform corresponding depth-pixel abscissa curve fitting processing on the first, second, third, fourth or fifth depth data sequences to generate the corresponding first fitting curve, second fitting curve, third fitting curve, fourth fitting curve or fifth fitting curve.

[0110] Specifically, it includes: step 341, taking the first, second, third, fourth or fifth depth data sequence as the corresponding current depth data sequence; and constructing a two-dimensional coordinate system space with pixel horizontal coordinate as the horizontal axis coordinate and depth as the vertical axis coordinate, denoted as the corresponding depth-pixel horizontal coordinate space;

[0111] Step 342: In the depth-pixel horizontal coordinate space, the first pixel horizontal coordinate and the first depth of each depth data group in the current depth data sequence are used as the corresponding horizontal axis coordinate and vertical axis coordinate of the plotting point to mark the plotting point and generate the corresponding first plotting point; and all the obtained first plotting points are sequentially connected to obtain the corresponding first plotting curve; and the first plotting curve is filtered; and the filtered first plotting curve is used as the corresponding current fitting curve.

[0112] Step 343: Based on the correspondence between the current depth data sequence and the first, second, third, fourth, or fifth depth data sequences, output the obtained current fitted curve as the corresponding first, second, third, fourth, or fifth fitted curve.

[0113] Step 35: The first fitted curve obtained is used as the corresponding first right ventricular anterior wall characteristic size curve; the second fitted curve obtained is used as the corresponding first right ventricular characteristic size curve; the third fitted curve obtained is used as the corresponding first interventricular septum characteristic size curve; the fourth fitted curve obtained is used as the corresponding first left ventricular characteristic size curve; and the fifth fitted curve obtained is used as the corresponding first left ventricular posterior wall characteristic size curve.

[0114] Step 4: Based on the characteristic size curve of the first left ventricle, identify key points during diastole and systole to generate corresponding key point sequences for the first diastole and systole.

[0115] The first diastolic keypoint sequence includes the x-coordinates of multiple first diastolic keypoint pixels; the first systolic keypoint sequence includes the x-coordinates of multiple first systolic keypoint pixels.

[0116] Specifically, it includes: Step 41, identifying multiple first peak points by performing peak point identification on the first left ventricular characteristic size curve based on a preset peak finding algorithm; taking the first pixel abscissa corresponding to each first peak point as the corresponding first diastolic key point pixel abscissa; and generating the corresponding first diastolic key point sequence by sorting all the obtained first diastolic key point pixel abscissas from left to right.

[0117] Here, Embodiment 1 of the present invention supports a variety of peak finding algorithms, including comparison discrimination method, first-order difference combined with comparison discrimination method, second-order difference discrimination method, zero-crossing detection combined with extreme value detection discrimination method, CFAR algorithm, etc.

[0118] Step 42: The first left ventricular feature size curve is flipped vertically to generate the corresponding second left ventricular feature size curve; and the peak points of the second left ventricular feature size curve are identified based on the peak finding algorithm to obtain multiple second peak points; the first pixel abscissa corresponding to each second peak point is used as the corresponding first systolic key point pixel abscissa; and all the obtained first systolic key point pixel abscissas are sorted from left to right to generate the corresponding first systolic key point sequence.

[0119] Here, after flipping the first left ventricular characteristic size curve vertically, the peak becomes a valley and the valley becomes a peak. The peak point identified from the second left ventricular characteristic size curve based on the peak-finding algorithm is also the valley point on the first left ventricular characteristic size curve. It is known that the left ventricular diameter should gradually increase during diastole and gradually decrease during systole. Therefore, in Embodiment 1 of the present invention, each peak point on the first left ventricular characteristic size curve when the left ventricular diameter reaches its maximum is taken as a diastolic key point, and each valley point on the first left ventricular characteristic size curve when the left ventricular diameter reaches its minimum is taken as a systolic key point. After obtaining the diastolic and systolic key points, Embodiment 1 of the present invention extracts the first pixel abscissa corresponding to each key point as the corresponding first diastolic and systolic key point pixel abscissa. Here, the first pixel abscissa is the pixel abscissa in the first image.

[0120] Step 5: Based on the first diastolic key point sequence and all characteristic size curves, perform diastolic characteristic size set confirmation processing to generate the corresponding first diastolic characteristic size set;

[0121] Specifically, this includes: traversing the x-coordinates of the pixels of each first diastolic key point in the first diastolic key point sequence; during traversal, using the x-coordinate of the currently traversed first diastolic key point pixel as the corresponding x-coordinate of the current key point pixel; and using the first depth corresponding to the current key point pixel x-coordinate on the first right ventricular anterior wall characteristic size curve, the first right ventricular characteristic size curve, the first interventricular septum characteristic size curve, the first left ventricular characteristic size curve, and the first left ventricular posterior wall characteristic size curve as the corresponding first right ventricular anterior wall thickness, first right ventricular intraventricular diameter, first interventricular septum thickness, and first left ventricular posterior wall thickness. The first diastolic time is generated by confirming the time information corresponding to the pixel abscissa of the current key point according to the correspondence between the pixel abscissa of the first image and the time axis; and the first diastolic time is generated by the pixel abscissa of the current key point, the first diastolic time, the first right ventricular anterior wall thickness, the first right ventricular diameter, the first interventricular septum thickness, the first left ventricular diameter, and the first left ventricular posterior wall thickness; and at the end of the traversal, the first diastolic feature size data set is formed by all the obtained first diastolic feature size data sets.

[0122] Step 6: Based on the first contraction phase key point sequence and all feature size curves, perform contraction phase feature size set confirmation processing to generate the corresponding first contraction phase feature size set;

[0123] Specifically, this includes: traversing the x-coordinates of the pixels of each key point in the first systolic phase of the key point sequence; during traversal, using the x-coordinate of the currently traversed key point pixel as the corresponding x-coordinate of the current key point pixel; and using the first depth corresponding to the x-coordinate of the current key point pixel on the first right ventricular anterior wall feature size curve, the first right ventricular feature size curve, the first interventricular septum feature size curve, the first left ventricular feature size curve, and the first left ventricular posterior wall feature size curve as the corresponding second right ventricular anterior wall thickness, second right ventricular diameter, second interventricular septum thickness, second left ventricular diameter, and second left ventricular posterior wall thickness; confirming the time information corresponding to the x-coordinate of the current key point pixel based on the correspondence between the pixel x-coordinates of the first image and the time axis to generate the corresponding first systolic time; and forming the corresponding first systolic feature size data group from the current key point pixel x-coordinate, the first systolic time, the second right ventricular anterior wall thickness, the second right ventricular diameter, the second interventricular septum thickness, the second left ventricular diameter, and the second left ventricular posterior wall thickness; and at the end of the traversal, forming the corresponding first systolic feature size set from all the obtained first systolic feature size data groups.

[0124] Step 7: Output the first diastolic feature size set and the first systolic feature size set as the feature size measurement results of the first image.

[0125] Figure 3 This is a module structure diagram of a processing device for measuring characteristic dimensions of M-mode echocardiography provided in Embodiment 2 of the present invention. This device can be a terminal device or server implementing the method of the embodiments of the present invention, or it can be a device connected to the aforementioned terminal device or server to implement the method of the embodiments of the present invention. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a receiving module 201, an image instance segmentation module 202, a feature size curve fitting module 203, a diastolic and systolic key point recognition module 204, a diastolic feature size measurement module 205, a systolic feature size measurement module 206, and an output module 207.

[0126] The receiving module 201 is used to receive the M-mode echocardiogram as the corresponding first image.

[0127] The image instance segmentation module 202 is used to perform binary mask image extraction processing on the image regions of the motion state change curves corresponding to the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle and left ventricular posterior wall on the first image based on a preset image instance segmentation model to generate the corresponding first right ventricular anterior wall mask, first right ventricle mask, first interventricular septum mask, first left ventricle mask and first left ventricular posterior wall mask.

[0128] The feature size curve fitting module 203 is used to perform feature size change curve fitting processing on the mask image of the first right ventricular anterior wall, the mask image of the first right heart ventricle, the mask image of the first interventricular septum, the mask image of the first left heart ventricle, and the mask image of the first left ventricular posterior wall to generate the corresponding feature size curves of the first right ventricular anterior wall, the first right heart ventricle, the first interventricular septum, the first left heart ventricle, and the first left ventricular posterior wall.

[0129] The diastolic and systolic key point recognition module 204 is used to identify diastolic and systolic key points based on the first left ventricular characteristic size curve to generate corresponding first diastolic key point sequences and first systolic key point sequences; the first diastolic key point sequence includes the horizontal coordinates of multiple first diastolic key point pixels; the first systolic key point sequence includes the horizontal coordinates of multiple first systolic key point pixels.

[0130] The diastolic characteristic size measurement module 205 is used to generate the corresponding first diastolic characteristic size set by performing diastolic characteristic size set confirmation processing based on the first diastolic key point sequence and all characteristic size curves.

[0131] The contraction period feature size measurement module 206 is used to generate the corresponding first contraction period feature size set by performing a contraction period feature size set confirmation process based on the first contraction period key point sequence and all feature size curves.

[0132] The output module 207 is used to output the first diastolic feature size set and the first systolic feature size set as the feature size measurement results of the first image.

[0133] The processing device for measuring the characteristic dimensions of M-mode echocardiography provided in this embodiment of the invention can execute the method steps in the above-described method embodiments. Its implementation principle and technical effects are similar, and will not be repeated here.

[0134] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, the receiving module can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and called and executed by a processing element of the device. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0135] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more application-specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs). Furthermore, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor capable of calling program code. Additionally, these modules can be integrated together as a system-on-a-chip (SoC).

[0136] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer described above can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions described above can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, Bluetooth, microwave, etc.) means. The computer-readable storage medium described above can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The aforementioned available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0137] Figure 4 This is a schematic diagram of an electronic device provided in Embodiment 3 of the present invention. This electronic device can be the aforementioned terminal device or server, or it can be a terminal device or server connected to the aforementioned terminal device or server that implements the method of the embodiments of the present invention. Figure 4 As shown, the electronic device may include: a processor 301 (e.g., CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transmission and reception operations of the transceiver 303. The memory 302 may store various instructions for performing various processing functions and implementing the methods and processes provided in the above embodiments of the present invention. Preferably, the electronic device involved in the embodiments of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to realize communication connections between components. The communication port 306 is used for communication between the electronic device and other peripherals.

[0138] exist Figure 4The system bus mentioned can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include Non-Volatile Memory, such as at least one disk storage device.

[0139] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0140] It should be noted that the embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to execute the methods and processes provided in the above embodiments.

[0141] This invention also provides a chip for executing instructions, which is used to execute the methods and processes provided in the above embodiments.

[0142] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for measuring the feature dimensions of M-mode echocardiography. First, a binary mask is extracted from the image regions of the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall in M-mode echocardiography based on an image instance segmentation model (using the Mask R-CNN model by default). Then, based on each binary mask, the feature dimension change curves of the right ventricular anterior wall thickness, right ventricular diameter, interventricular septum thickness, left ventricular diameter, and left ventricular posterior wall thickness are fitted. Next, a peak-finding algorithm is used to identify key points during diastole and systole on the left ventricular feature dimension curves. Then, based on the identified key points during diastole and systole, corresponding dimension data are obtained from each feature dimension change curve to form corresponding diastolic and systolic feature dimension sets. Finally, the obtained diastolic and systolic feature dimension sets are output as the feature dimension measurement results of the M-mode echocardiography. This invention improves the measurement efficiency of M-mode echocardiography while ensuring the stability of measurement quality.

[0143] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0144] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0145] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for measuring characteristic dimensions of M-mode echocardiography, characterized in that, The method includes: Receive an M-mode echocardiogram as the corresponding first image; Based on a preset image instance segmentation model, the image regions of the first image corresponding to the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle and left ventricular posterior wall are processed by binary mask extraction to generate the corresponding first right ventricular anterior wall mask, first right ventricle mask, first interventricular septum mask, first left ventricle mask and first left ventricular posterior wall mask. The first right ventricular anterior wall mask, the first right ventricle mask, the first interventricular septum mask, the first left ventricle mask, and the first left ventricular posterior wall mask are respectively subjected to feature size change curve fitting processing to generate corresponding first right ventricular anterior wall feature size curve, first right ventricle feature size curve, first interventricular septum feature size curve, first left ventricle feature size curve, and first left ventricular posterior wall feature size curve; Based on the first left ventricular characteristic size curve, key points during diastole and systole are identified to generate corresponding first diastolic key point sequences and first systolic key point sequences; the first diastolic key point sequence includes the horizontal coordinates of multiple first diastolic key point pixels; the first systolic key point sequence includes the horizontal coordinates of multiple first systolic key point pixels. Based on the first diastolic key point sequence and all feature size curves, the diastolic feature size set is confirmed and a corresponding first diastolic feature size set is generated. Based on the first contraction phase key point sequence and all feature size curves, the contraction phase feature size set confirmation process is performed to generate the corresponding first contraction phase feature size set; The first set of diastolic feature dimensions and the first set of systolic feature dimensions are output as the feature dimension measurement results of the first image.

2. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 1, characterized in that, The first image includes an image region corresponding to the motion state change curve of the anterior wall of the right ventricle, an image region corresponding to the motion state change curve of the right ventricle, an image region corresponding to the motion state change curve of the interventricular septum, an image region corresponding to the motion state change curve of the left ventricle, and an image region corresponding to the motion state change curve of the posterior wall of the left ventricle; and the horizontal time axes of the image regions corresponding to the motion state change curve of the anterior wall of the right ventricle, the image regions corresponding to the right ventricle, the image regions corresponding to the interventricular septum, the image regions corresponding to the left ventricle, and the image regions corresponding to the motion state change curve of the posterior wall of the left ventricle are aligned.

3. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 2, characterized in that, The network structure of the image instance segmentation model is implemented based on the network structure of the Mask R-CNN model; The image instance segmentation model includes a first feature extraction backbone network, a first feature pyramid network, a first region proposal network, a first region of interest alignment network, a first convolutional network, a first fully connected network, a second fully connected network, a first mask extraction network, and a first fusion module. The output of the first feature extraction backbone network is connected to the input of the first feature pyramid network. The output of the first feature pyramid network is connected to the input of the first region proposal network. The output of the first region proposal network is connected to the input of the first region of interest alignment network. The output of the first region of interest alignment network is connected to the input of the first convolutional network. The output of the first convolutional network is connected to the inputs of the first fully connected network, the second fully connected network, and the first mask extraction network, respectively. The input of the first fusion module is connected to the outputs of the first fully connected network, the second fully connected network, and the first mask extraction network, respectively. The first feature extraction backbone network is defaulted to a ResNet-50 network. The first feature extraction backbone network is used to perform feature extraction processing on the input image of the model to generate five feature maps of different sizes, namely the first feature map, the second feature map, the third feature map, the fourth feature map, and the fifth feature map; the (height, width) of the input image is (H, W), then the (height, width) of the corresponding first, second, third, fourth, and fifth feature maps are (H / 2, W / 2), (H / 4, W / 4), (H / 8, W / 8), (H / 16, W / 16), and (H / 32, W / 32), respectively; The first feature pyramid network is used to upsample the input fifth feature map to obtain a first upsampled feature map with the same size as the fourth feature map, and to fuse the first upsampled feature map and the fourth feature map to obtain a corresponding first fused feature map; then, it upsamples the first fused feature map to obtain a second upsampled feature map with the same size as the third feature map, and to fuse the second upsampled feature map and the third feature map to obtain a corresponding second fused feature map; then, it upsamples the second fused feature map to obtain a third upsampled feature map with the same size as the second feature map, and to fuse the third upsampled feature map and the second feature map to obtain a corresponding third fused feature map; finally, it upsamples the third fused feature map to obtain a fourth upsampled feature map with the same size as the first feature map, and to fuse the fourth upsampled feature map and the first feature map to obtain a corresponding fourth fused feature map; The first region proposal network is used to perform image region pre-identification processing on the motion state change curve of the fourth fused feature map to obtain multiple first region recognition boxes; The first region of interest alignment network is used to perform frame fusion alignment processing on all the first region recognition boxes to obtain multiple first fused recognition boxes; And extract the sub-feature maps covered by each of the first fusion recognition boxes in the fourth fusion feature map as the corresponding first recognition box feature maps; And the offset position of each of the first fused recognition boxes in the fourth fused feature map is taken as the corresponding first recognition box feature map offset position; The first convolutional network is used to perform feature extraction processing on each of the first recognition box feature maps to generate the corresponding second recognition box feature maps; The height and width of the second recognition box feature map are the same as those of the first recognition box feature map; The first fully connected network is used to classify and predict the feature map type of each second recognition box feature map and use the prediction result as the corresponding first mask map type; the first mask map type includes the right ventricular anterior wall motion state change curve type, the right ventricular motion state change curve type, the interventricular septum motion state change curve type, the left ventricular motion state change curve type, and the left ventricular posterior wall motion state change curve type. The second fully connected network is used to perform regression prediction on the offset position and size of each second recognition box feature map in the model input image based on the offset position of the first recognition box feature map corresponding to each second recognition box feature map and the size correspondence between the fourth fused feature map and the model input image, and use the prediction result as the corresponding first mask map offset position and first mask map size; The first mask extraction network is used to perform binary mask transformation on the feature maps of each second recognition box to generate the corresponding first mask map; The pixel value of each pixel in the first mask image has only two possible values; The first fusion module is used to adjust the mask image size of each first mask image according to the corresponding first mask image size to generate a corresponding second mask image; The first mask data set is composed of each second mask image and the corresponding first mask image type, the first mask image offset position, and the first mask image size; The obtained dataset of all the first mask images is then combined into the model output.

4. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 3, characterized in that, The offset position and size of the first mask of the second mask, which is of the type of right ventricular anterior wall motion state change curve, match the region offset position and region size of the image region of the motion state change curve corresponding to the right ventricular anterior wall on the first image; The offset position and size of the first mask of the second mask, which is of the type of right ventricular motion state change curve, match the region offset position and region size of the image region of the motion state change curve corresponding to the right ventricle on the first image; The offset position and size of the first mask of the second mask, which is of the type of ventricular septum motion state change curve, match the region offset position and region size of the image region of the motion state change curve corresponding to the ventricular septum on the first image; The offset position and size of the first mask of the second mask, which is of the left ventricular motion state change curve type, match the region offset position and region size of the image region of the motion state change curve corresponding to the left ventricle on the first image; The offset position and size of the first mask of the second mask, which is of the left ventricular posterior wall motion state change curve type, match the region offset position and region size of the image region of the motion state change curve corresponding to the left ventricular posterior wall on the first image.

5. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 3, characterized in that, The image instance segmentation model is trained using a supervised training method; the model loss function used during the training process is a preset first model loss function L. all L all =λ1*L CLS +λ2*L OFF +λ3*L MASK ; Among them, the classification loss function L CLS Used to calculate the loss based on the output of the first fully connected network and the corresponding preset ground truth training labels; offset loss function L OFF Used to calculate the loss based on the output of the second fully connected network and the corresponding preset ground truth training labels; Mask loss function L MASK Used to extract the network's output based on the first mask image and calculate the loss using the corresponding preset ground truth training labels; λ1, λ2, and λ3 are preset weighting coefficients.

6. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 3, characterized in that, The method, based on a preset image instance segmentation model, performs binary mask image extraction on the image regions of the first image corresponding to the motion state change curves of the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle, and left ventricular posterior wall to generate corresponding first right ventricular anterior wall mask, first right ventricle mask, first interventricular septum mask, first left ventricle mask, and first left ventricular posterior wall mask, specifically including: The first image is input into the image instance segmentation model for processing to obtain multiple sets of first mask image data; the first mask image data set includes the second mask image, the first mask image type, the first mask image offset position, and the first mask image size; The second mask, whose first mask image type is the right ventricular anterior wall motion state change curve, is used as the corresponding first right ventricular anterior wall mask image; the second mask, whose first mask image type is the right ventricular motion state change curve, is used as the corresponding first right ventricular mask image; the second mask, whose first mask image type is the interventricular septum motion state change curve, is used as the corresponding first interventricular septum mask image; the second mask, whose first mask image type is the left ventricular motion state change curve, is used as the corresponding first left ventricular mask image; and the second mask, whose first mask image type is the left ventricular posterior wall motion state change curve, is used as the corresponding first left ventricular posterior wall mask image.

7. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 1, characterized in that, The process of performing feature size change curve fitting on the first right ventricular anterior wall mask image, the first right ventricular mask image, the first interventricular septum mask image, the first left ventricular mask image, and the first left ventricular posterior wall mask image to generate corresponding first right ventricular anterior wall feature size curve, first right ventricular feature size curve, first interventricular septum feature size curve, first left ventricular feature size curve, and first left ventricular posterior wall feature size curve specifically includes: The first right ventricular anterior wall mask, the first right ventricular mask, the first interventricular septum mask, the first left ventricular mask, and the first left ventricular posterior wall mask are respectively processed by format binarization to generate corresponding first binary image, second binary image, third binary image, fourth binary image, and fifth binary image; the pixel values ​​of each pixel in the first, second, third, fourth, and fifth binary images have only two possible values, namely the preset foreground pixel value and the background pixel value; Image denoising processing is performed on the first, second, third, fourth, and fifth binary images respectively; The first, second, third, fourth, or fifth binary images after noise reduction are processed by extracting the depth of the upper and lower edges of the foreground region to obtain the corresponding first depth data sequence, second depth data sequence, third depth data sequence, fourth depth data sequence, or fifth depth data sequence; the first, second, third, fourth, or fifth depth data sequence is composed of multiple depth data groups sorted together; the depth data group includes the first pixel horizontal coordinate and the first depth; Based on the first, second, third, fourth, or fifth depth data sequence, perform corresponding depth-pixel abscissa curve fitting processing to generate the corresponding first fitting curve, second fitting curve, third fitting curve, fourth fitting curve, or fifth fitting curve; The first fitted curve obtained is used as the corresponding first right ventricular anterior wall characteristic size curve; the second fitted curve obtained is used as the corresponding first right ventricular characteristic size curve; the third fitted curve obtained is used as the corresponding first interventricular septum characteristic size curve; the fourth fitted curve obtained is used as the corresponding first left ventricular characteristic size curve; and the fifth fitted curve obtained is used as the corresponding first left ventricular posterior wall characteristic size curve.

8. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 7, characterized in that, The step of extracting the depth of the upper and lower edges of the foreground region from the first, second, third, fourth, or fifth binary images after noise reduction to obtain the corresponding first, second, third, fourth, or fifth depth data sequences specifically includes: Use the first, second, third, fourth, or fifth binary image as the corresponding current binary image; The current binary image is scanned column by column; each column scan yields a set of pixels, denoted as the current column pixel set, and the x-coordinate of the pixels in the current scanned column on the current binary image is marked as the corresponding current column pixel x-coordinate. Based on the first mask offset position of the mask image corresponding to the current binary image, the pixel x-coordinate corresponding to the current column pixel on the first image is confirmed to obtain the corresponding first pixel x-coordinate. If the number of pixels in the current column pixel set whose pixel value is a foreground pixel value exceeds a preset minimum foreground pixel count threshold, then the pixels in the current column pixel set whose pixel value is a foreground pixel value are extracted to form the corresponding foreground pixel set. The system combines and extracts pixels with minimum and maximum vertical coordinates from the foreground pixel set as the corresponding upper and lower edge pixels. It then calculates the absolute difference between the vertical coordinates of these upper and lower edge pixels on the current binary image to generate the corresponding first upper and lower edge pixel distance. The product of this first upper and lower edge pixel distance and a preset pixel distance-real distance scale is used as the corresponding first depth. The first pixel horizontal coordinate and the first depth form the corresponding depth data group. Finally, after all columns have been scanned, all the obtained depth data are sorted from left to right according to the corresponding first pixel horizontal coordinate to generate the corresponding current depth data sequence. Based on the correspondence between the current binary image and the first, second, third, fourth, or fifth binary images, the obtained current depth data sequence is output as the corresponding first, second, third, fourth, or fifth depth data sequence.

9. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 7, characterized in that, The step of performing depth-pixel abscissa curve fitting processing based on the first, second, third, fourth, or fifth depth data sequences to generate corresponding first, second, third, fourth, or fifth fitting curves specifically includes: The first, second, third, fourth, or fifth depth data sequence is taken as the corresponding current depth data sequence; and a two-dimensional coordinate system space is constructed with the pixel horizontal coordinate as the horizontal axis and the depth as the vertical axis, denoted as the corresponding depth-pixel horizontal coordinate space; In the depth-pixel horizontal coordinate space, the first pixel horizontal coordinate and the first depth of each depth data group in the current depth data sequence are used as the corresponding horizontal axis coordinate and vertical axis coordinate of the plotting point to mark the plotting point and generate the corresponding first plotting point; and all the obtained first plotting points are sequentially connected to obtain the corresponding first plotting curve; and the first plotting curve is filtered; and the filtered first plotting curve is used as the corresponding current fitting curve; Based on the correspondence between the current depth data sequence and the first, second, third, fourth, or fifth depth data sequences, the obtained current fitting curve is output as the corresponding first, second, third, fourth, or fifth fitting curve.

10. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 1, characterized in that, The step of identifying key points during diastole and systole based on the first left ventricular characteristic size curve to generate corresponding first diastolic key point sequences and first systolic key point sequences specifically includes: Based on a preset peak-finding algorithm, the peak points of the first left ventricular characteristic size curve are identified to obtain multiple first peak points; the first pixel abscissa corresponding to each first peak point is used as the corresponding first diastolic key point pixel abscissa; and the obtained first diastolic key point pixel abscissas are sorted from left to right to generate the corresponding first diastolic key point sequence. The first left ventricular feature size curve is flipped vertically to generate a corresponding second left ventricular feature size curve; and the peak point is identified on the second left ventricular feature size curve based on the peak finding algorithm to obtain multiple second peak points; the first pixel abscissa corresponding to each second peak point is used as the corresponding first systolic key point pixel abscissa; and all the obtained first systolic key point pixel abscissas are sorted from left to right to generate the corresponding first systolic key point sequence.

11. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 1, characterized in that, The step of generating the corresponding first diastolic feature size set by confirming the diastolic feature size set based on the first diastolic key point sequence and all feature size curves specifically includes: The abscissa of each pixel of the first diastolic key point in the first diastolic key point sequence is traversed. During traversal, the abscissa of the currently traversed first diastolic key point pixel is taken as the abscissa of the corresponding current key point pixel. The first depth corresponding to the abscissa of the current key point pixel on the first right ventricular anterior wall feature size curve, the first right ventricular feature size curve, the first interventricular septum feature size curve, the first left ventricular feature size curve, and the first left ventricular posterior wall feature size curve is taken as the corresponding first right ventricular anterior wall thickness, first right ventricular diameter, first interventricular septum thickness, first left ventricular diameter, and first... The thickness of the left ventricular posterior wall is determined; and based on the correspondence between the pixel horizontal coordinates and the time axis of the first image, the time information corresponding to the horizontal coordinate of the current key point pixel is confirmed to generate the corresponding first diastolic time; and the horizontal coordinate of the current key point pixel, the first diastolic time, the first right ventricular anterior wall thickness, the first right ventricular diameter, the first interventricular septum thickness, the first left ventricular diameter, and the first left ventricular posterior wall thickness are combined to form the corresponding first diastolic feature size data set; and at the end of the traversal, all the obtained first diastolic feature size data sets constitute the corresponding first diastolic feature size set.

12. The processing method for measuring characteristic dimensions of M-mode echocardiography according to claim 1, characterized in that, The step of generating the corresponding first contraction period feature size set by confirming the contraction period feature size set based on the first contraction period key point sequence and all feature size curves specifically includes: The abscissa of each pixel of the first systolic key point in the first systolic key point sequence is traversed. During traversal, the abscissa of the currently traversed first systolic key point pixel is taken as the corresponding abscissa of the current key point pixel. The first depth on the first right ventricular anterior wall feature size curve, the first right ventricular feature size curve, the first interventricular septum feature size curve, the first left ventricular feature size curve, and the first left ventricular posterior wall feature size curve, corresponding to the abscissa of the current key point pixel, is taken as the corresponding second right ventricular anterior wall thickness, second right ventricular ventricular diameter, second interventricular septum thickness, second left ventricular diameter, and... The second left ventricular posterior wall thickness; and based on the correspondence between the pixel horizontal coordinates of the first image and the time axis, confirm the time information corresponding to the current key point pixel horizontal coordinate to generate the corresponding first systolic time; and the current key point pixel horizontal coordinates, the first systolic time, the second right ventricular anterior wall thickness, the second right ventricular diameter, the second interventricular septum thickness, the second left ventricular diameter, and the second left ventricular posterior wall thickness form the corresponding first systolic feature size data set; and at the end of the traversal, all the obtained first systolic feature size data sets constitute the corresponding first systolic feature size set.

13. An apparatus for implementing the processing method for measuring characteristic dimensions of M-mode echocardiography as described in any one of claims 1-12, characterized in that, The device includes: a receiving module, an image instance segmentation module, a feature size curve fitting module, a diastolic and systolic key point recognition module, a diastolic feature size measurement module, a systolic feature size measurement module, and an output module; The receiving module is used to receive an M-mode echocardiogram as the corresponding first image; The image instance segmentation module is used to perform binary mask image extraction processing on the image regions of the motion state change curves corresponding to the right ventricular anterior wall, right ventricle, interventricular septum, left ventricle and left ventricular posterior wall on the first image based on a preset image instance segmentation model to generate the corresponding first right ventricular anterior wall mask image, first right ventricle mask image, first interventricular septum mask image, first left ventricle mask image and first left ventricular posterior wall mask image. The feature size curve fitting module is used to perform feature size change curve fitting processing on the first right ventricular anterior wall mask, the first right ventricle mask, the first interventricular septum mask, the first left ventricle mask, and the first left ventricular posterior wall mask to generate corresponding first right ventricular anterior wall feature size curve, first right ventricle feature size curve, first interventricular septum feature size curve, first left ventricle feature size curve, and first left ventricular posterior wall feature size curve, respectively. The diastolic and systolic key point identification module is used to identify diastolic and systolic key points based on the first left ventricular characteristic size curve to generate corresponding first diastolic key point sequences and first systolic key point sequences; the first diastolic key point sequence includes multiple first diastolic key point pixel abscissas; the first systolic key point sequence includes multiple first systolic key point pixel abscissas. The diastolic characteristic size measurement module is used to generate the corresponding first diastolic characteristic size set by performing diastolic characteristic size set confirmation processing based on the first diastolic key point sequence and all characteristic size curves; The contraction period feature size measurement module is used to generate a corresponding first contraction period feature size set by performing a contraction period feature size set confirmation process based on the first contraction period key point sequence and all feature size curves. The output module is used to output the first diastolic feature size set and the first systolic feature size set as the feature size measurement results of the first image.

14. An electronic device, characterized in that, include: Memory, processor, and transceiver; The processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the method according to any one of claims 1-12; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the method according to any one of claims 1-12.

Images