CN117547306B

CN117547306B - Left ventricular ejection fraction measurement method, system and device based on M-type ultrasound

Info

Publication number: CN117547306B
Application number: CN202410031339.1A
Authority: CN
Inventors: 陈群
Original assignee: Shenzhen Wisonic Medical Technology Co ltd
Current assignee: Shenzhen Wisonic Medical Technology Co ltd
Priority date: 2024-01-09
Filing date: 2024-01-09
Publication date: 2024-03-15
Anticipated expiration: 2044-01-09
Also published as: CN117547306A

Abstract

The invention is suitable for the technical field of ultrasonic imaging, and provides a left ventricular ejection fraction measuring method, a system and a device based on M-type ultrasonic, wherein the method comprises the steps of acquiring an M-type ultrasonic image acquired from a target area; analyzing the information of each included cardiac cycle according to the M-type ultrasonic image; locating all key points according to the M-type ultrasonic image; screening out each effective key point from all the located key points according to the information of each cardiac cycle contained in the M-type ultrasonic image; according to the cardiac cycle information contained in the M-type ultrasonic image and the screened effective key points, matching the target key points of each cardiac cycle; calculating left ventricular ejection fraction of each cardiac cycle according to the matched target key points of each cardiac cycle; and corresponding display is carried out according to the defined display mode. The invention solves the problems of insufficient reliability and poor interpretability of the result output by the existing measuring method.

Description

Left ventricular ejection fraction measurement method, system and device based on M-type ultrasound

Technical Field

The invention belongs to the technical field of ultrasonic imaging, and particularly relates to a left ventricular ejection fraction measuring method, system and device based on M-type ultrasonic.

Background

The left ventricular ejection fraction (Left ventricular ejection fraction, LVEF) refers to the proportion of the volume of left ventricular end-diastole that is pumped out during the left ventricular systole in each cardiac cycle, and is an important indicator that reflects whether the systolic function is normal. Among them, the method for measuring LVEF is mainly echocardiography, cardiac catheter interventional therapy, magnetic Resonance Imaging (MRI) or Computed Tomography (CT), etc., and is most commonly used at present due to the noninvasive nature and low cost of the echocardiography measuring method.

In clinical measurement of LVEF using echocardiography, a doctor needs to determine the End Systole (ES) and End Diastole (ED) of the left ventricle on an ultrasound image time series, then manually measure a partial measurement term, transform the partial measurement term into the left ventricular volumes at these two moments, namely End Diastole Volume (EDV) and End Systole Volume (ESV), respectively, and calculate the LVEF based on the existing formula. According to the different ultrasonic modes, the methods for measuring the LVEF by utilizing the echocardiogram can be divided into two types, wherein the first type is a measurement method based on B-type ultrasonic, the two-dimensional B-type ultrasonic image time sequence is detected, the second type is a measurement method based on M-type ultrasonic, the one-dimensional M-type ultrasonic image time sequence is detected, and the two methods are very commonly used clinically.

The measurement method based on M-mode ultrasound requires a doctor to respectively determine the End systole (ED) time and the End diastole (ES) time of the left ventricle on an M-mode ultrasound image, manually measure the inner diameter (Left ventricular diameter, LVID) of the left ventricle, the rear wall thickness (Left ventricular posterior wall, LVPW) of the left ventricle, the inter-chamber thickness (Interventricular septum, IVS) and the inner diameter (Right ventricle diameter, RVD) of the right ventricle at the two times, and then calculate LVEF based on the existing formula. Specifically, to determine the LVEF process for a complete cardiac cycle, the physician is required to first determine the ES time and the ED time in a cardiac cycle, then manually measure the measurements, draw 4 line segments at ES time, RVDs, IVSs, LVIDs, LVPWs each, and draw 4 line segments at ED time, RVDd, IVSd, LVIDd, LVPWd each.

Therefore, the method for measuring LVEF based on M-mode ultrasound clinically requires more manual operation steps, wherein the technology capable of automatically measuring LVEF can greatly reduce the operation time of doctors and relieve the burden in clinic, however, related research in the industry still has some problems as follows:

(1) The related research is mainly focused on B-type ultrasonic, the technical means based on M-type ultrasonic are very deficient, and few schemes based on M-type ultrasonic exist, but the scheme is imperfect and has defects;

(2) The image content of the M-type ultrasonic image is complex, one image can have a plurality of cardiac cycles, LVEF information contained in each cardiac cycle can be complete or missing, most of the existing schemes directly utilize deep learning end-to-end output results, then follow-up calculation is carried out, and reliability judgment is not carried out on the output of a deep learning model, so that reliability is insufficient;

(3) 2) the M-type ultrasonic image often contains a plurality of cardiac cycles, the same measurement item has a plurality of results, the value of the maximum left ventricular end diastole inner diameter is directly taken as the subsequent calculation input when the final LVEF is calculated, and the output LVEF has larger errors, no problem of results and insufficient reliability due to the calculation mode;

(4) The existing scheme completely depends on the result of deep learning output, so that the interpretation is not strong and is not easily accepted clinically.

Disclosure of Invention

The embodiment of the invention aims to provide a left ventricular ejection fraction measuring method, system and device based on M-type ultrasound, and aims to solve the problems of insufficient reliability and poor interpretability of results output by the existing measuring method.

The embodiment of the invention is realized in such a way that the left ventricular ejection fraction measuring method based on M-type ultrasound comprises the following steps:

acquiring an M-type ultrasonic image acquired from a target area, wherein the vertical axis of the M-type ultrasonic image is the depth of different tissues, and the horizontal axis is time;

analyzing each cardiac cycle information contained in the M-type ultrasonic image according to the acquired M-type ultrasonic image, wherein the cardiac cycle information comprises end systole time and end diastole time;

locating all key points on the M-type ultrasonic image according to the acquired M-type ultrasonic image, wherein the key points comprise position information and category information;

screening out each effective key point from all the located key points according to the information of each cardiac cycle contained in the M-type ultrasonic image;

according to the cardiac cycle information contained in the M-type ultrasonic image and the screened effective key points, matching each target key point of each cardiac cycle, wherein the target key points are effective key points corresponding to paired end diastole time and end systole time in the cardiac cycle information;

calculating left ventricular ejection fraction of each cardiac cycle according to the matched target key points of each cardiac cycle;

And correspondingly displaying the M-type ultrasonic image, each cardiac cycle information, all key points, each effective key point, each target key point and each left ventricular ejection fraction according to the defined display mode.

Still further, the step of analyzing the cardiac cycle information contained in the M-mode ultrasound image from the acquired M-mode ultrasound image includes:

determining a motion curve of the rear wall of the left ventricle according to the acquired M-type ultrasonic image, and carrying out extremum solving processing on the motion curve of the rear wall of the left ventricle to obtain extremum information, wherein the extremum information comprises extremum points and corresponding position information;

traversing extreme points in the extreme information from left to right in sequence, and forming two adjacent extreme points into a point pair according to the sequence of the maximum value/the minimum value;

and determining each effective point pair according to the calculated transverse distance and longitudinal distance of all the point pairs and the corresponding preset parameter range, and determining each cardiac cycle information according to each effective point pair, wherein the position information corresponding to the maximum value of the effective point pair is end diastole time, and the position information corresponding to the minimum value of the effective point pair is end systole time.

Still further, the step of determining a motion profile of the rear wall of the left ventricle from the acquired M-mode ultrasound image includes:

traversing from left to right to obtain each column of pixels of the M-type ultrasonic image in sequence, and smoothing each pixel column;

respectively calculating the segmentation threshold value of each pixel column after the smoothing processing, and carrying out threshold processing on each pixel column according to the segmentation threshold value to obtain each new pixel column;

performing convolution processing on each new pixel column according to the gradient detection check to obtain gradient information of each new pixel column;

determining the left ventricle back wall position in each new pixel column according to the gradient values of the upper edge and the lower edge of each bright band in each new pixel column, wherein the bright bands are continuous non-zero value pixels which are clustered together after threshold processing;

and smoothing the point set formed by the positions of the left ventricular rear wall in each new pixel row and connecting the point set to obtain a motion curve of the left ventricular rear wall.

Further, the step of calculating the division threshold value of each pixel column after the smoothing process, and performing the thresholding on each pixel column according to the division threshold value to obtain each new pixel column includes:

traversing all gray values as threshold values, and calculating the inter-class variance of each pixel column under each threshold value;

Selecting a gray value with the maximum inter-class variance in each pixel column as a segmentation threshold value of each pixel column;

and setting the gray value of the pixel in each pixel column smaller than the corresponding segmentation threshold to be zero according to the segmentation threshold of each pixel column, so as to obtain each new pixel column.

Further, the step of determining the left ventricular posterior wall position in each new pixel column according to the gradient values of the upper and lower edges of each bright band in each new pixel column includes:

sequentially counting the bright band information in each new pixel column from top to bottom, wherein the bright band information comprises the number of bright bands and the number of pixels of each bright band;

traversing each bright band in the bright band information in each new pixel column to obtain gradient values of the upper edge and the lower edge of each bright band;

determining a bright band probability value of each bright band according to the sum of absolute values of gradient values of the upper edge and the lower edge of each bright band, wherein the bright band probability value represents the probability that each bright band is a left ventricular back wall bright band;

determining a bright band with the highest probability value in each new pixel column as a target bright band according to all the probability values of the bright bands in each new pixel column;

and determining the position of the left ventricle back wall in each new pixel column according to the position information of the target brightness band in each new pixel column.

Still further, the step of locating all key points on the M-mode ultrasonic image according to the acquired M-mode ultrasonic image includes:

extracting local features of all key points according to the existing marked data set, respectively establishing respective local feature templates for each type of key points based on the local features of each key point, and classifying and storing the key points;

traversing the obtained M-type ultrasonic image to obtain each image block, and extracting the local characteristics of each image block;

according to the comparison of the local features of each image block and the stored various local feature templates, each target image block with the distance from the local feature template smaller than a preset threshold value is matched;

and determining the position information and the category information of each key point according to the position information of each target image block and the type information of the matched local feature template.

Further, the step of screening each effective key point from all the located key points according to each cardiac cycle information contained in the M-mode ultrasonic image includes:

determining a first depth priori weight of each key point according to the ordinate of all the key points;

determining a second depth priori weight of each key point according to the longitudinal distance between all the key points and each key point of other categories in the neighborhood;

Determining the time prior weight of each key point according to the transverse distance between all the key points and each key point of other categories in the neighborhood;

determining the prior weight of the cardiac cycle of each key point according to the transverse distance between all the key points and the abscissa in the information of each cardiac cycle;

determining effective probability information of each key point in each positioned key point according to the first depth priori weight, the second depth priori weight, the time priori weight and the cardiac cycle priori weight determined by each key point, wherein the effective probability information is a probability value of each key point as an effective key point;

and screening out each effective key point according to the position information, the category information and the effective probability information of each key point.

Further, the calculation formula for determining the first depth priori weight of each key point according to the ordinate of all key points is as follows:

，

wherein,first depth a priori weight for current keypoint,/->Is the abscissa of the current key point, +.>Is the ordinate of the current key point, +.>Is standard deviation super parameter->A reasonable depth of the tissue structure of the category corresponding to the current key point, <' > is given>Is an exponential function based on a natural base e.

Further, the calculation formula for determining the second depth prior weight of each key point according to the longitudinal distance between all the key points and each key point of other categories in the neighborhood is as follows:

，

wherein,a second depth prior weight for the current keypoint,/->Is standard deviation super parameter->Is the abscissa of the current key point, +.>Is the ordinate of the current key point, +.>For the number of other categories of keypoints in the neighborhood of the current keypoint, +.>Representing the +.>Ordinate of each key point, +.>Is the +.>Reasonable depth of the individual keypoints to the current keypoint, +.>Is an exponential function based on a natural base e.

Further, the calculation formula for determining the time prior weight of each key point according to the lateral distance between all the key points and each key point of other categories in the neighborhood is as follows:

，

wherein,time prior weight for current key point, +.>Is standard deviation super parameter->Is the abscissa of the current key point, +.>Is the ordinate of the current key point, +.>For the number of other categories of keypoints in the neighborhood of the current keypoint, +.>Is the +.>The abscissa of the individual key points, +. >Based on natural baseeAn exponential function of the base.

Furthermore, the calculation formula for determining the prior weight of the cardiac cycle of each key point according to the transverse distance between all the key points and the abscissa in the information of each cardiac cycle is as follows:

，

wherein,prior weight for cardiac cycle for the current key point,/->Is the standard deviation super-parameter of the standard deviation,is the abscissa of the current key point, +.>Is the ordinate of the current key point, +.>Is the abscissa of the end systole or end diastole time in the cardiac cycle information closest to the current key point.

Further, the calculation formula for determining the effective probability information of each key point in the located key points according to the first depth priori weight, the second depth priori weight, the time priori weight and the cardiac cycle priori weight determined by each key point is as follows:

，

wherein,effective probability information for the current key point, < +.>Is the abscissa of the current key point, +.>Threshold +.>Is an empirical parameter, ++>Representing a threshold binarization function, in whichLess than threshold->Time 0 is greater than threshold +.>Unchanged time and->For the probability value of the current key point, +. >First depth a priori weight for current keypoint,/->A second depth prior weight for the current keypoint,/->Time prior weight for current key point, +.>The cardiac cycle for the current keypoint is weighted a priori.

Still further, the step of calculating left ventricular ejection fraction for each cardiac cycle based on the matched target keypoints for each cardiac cycle includes:

calculating the left ventricular internal diameter at the end systole moment and the left ventricular internal diameter at the end diastole moment in each cardiac cycle according to the matched target key points of each cardiac cycle;

calculating the end systole volume and the end diastole volume in each cardiac cycle according to the inner diameter of the left ventricle at the end systole moment and the inner diameter of the left ventricle at the end diastole moment in each cardiac cycle;

the left ventricular ejection fraction of each cardiac cycle is calculated from the end systole volume and the end diastole volume in each cardiac cycle.

It is also an object of another embodiment of the present invention to provide a left ventricular ejection fraction measurement system based on M-mode ultrasound, the system comprising:

the input module is used for acquiring M-type ultrasonic images acquired in a target area, wherein the vertical axis of the M-type ultrasonic images is the depth of different tissues, and the horizontal axis is time;

The analysis module is used for analyzing each cardiac cycle information contained in the M-type ultrasonic image according to the acquired M-type ultrasonic image, wherein the cardiac cycle information comprises end systole time and end diastole time;

the positioning module is used for positioning all key points on the M-type ultrasonic image according to the acquired M-type ultrasonic image, wherein the key points comprise position information and category information;

the screening module is used for screening each effective key point from all the located key points according to the information of each cardiac cycle contained in the M-type ultrasonic image;

the matching module is used for matching each target key point of each cardiac cycle according to cardiac cycle information contained in the M-type ultrasonic image and each screened effective key point, wherein the target key points are effective key points corresponding to paired end diastole time and end systole time in the cardiac cycle information;

the calculation module is used for calculating the left ventricular ejection fraction of each cardiac cycle according to the matched target key points of each cardiac cycle;

and the display module is used for correspondingly displaying the M-type ultrasonic image, each cardiac cycle information, all key points, each effective key point, each target key point and each left ventricular ejection fraction according to the defined display mode.

It is also an object of another embodiment of the present invention to provide an M-mode ultrasound based left ventricular ejection fraction measuring device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed performs the M-mode ultrasound based left ventricular ejection fraction measuring method as described above.

According to the left ventricular ejection fraction measuring method based on the M-type ultrasound, provided by the embodiment of the invention, the M-type ultrasound image is processed in a plurality of steps, including cardiac cycle analysis, key point positioning, key point screening, key point matching, left ventricular ejection fraction calculation and the like, so that the accuracy and the reliability of the left ventricular ejection fraction measurement are improved, and specifically, each cardiac cycle contained in the M-type ultrasound image can be analyzed by cardiac cycle analysis; the key point positioning can initially position each key point on the M-type ultrasonic image; the key point screening screens invalid false key points by combining cardiac cycle information, and reliable effective key points are reserved; the key point matching can be combined with the cardiac cycle information to match paired target key points in each complete cardiac cycle; the left ventricular ejection fraction is calculated to obtain each target key point through calculation and matching, so that the accurate left ventricular ejection fraction can be obtained, and the reliability of the result is ensured; and through carrying out corresponding display on the M-type ultrasonic image, each cardiac cycle information, all key points, each effective key point, each target key point and each left ventricular ejection fraction according to a display mode defined by a user, the measurement result of the final left ventricular ejection fraction can be checked, and meanwhile, intermediate results such as cardiac cycle information, key points, effective key points and target key points can be checked, so that the interpretability and the visual effect in clinical use are improved, and the problems of insufficient reliability and poor interpretability of the results output by the conventional measurement method are solved.

Drawings

FIG. 1 is a flow chart of a left ventricular ejection fraction measurement method based on M-mode ultrasound according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a left ventricular ejection fraction measurement system based on M-mode ultrasound according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of a left ventricular ejection fraction measuring device based on M-mode ultrasound according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of an M-mode ultrasound image in a left ventricular ejection fraction measurement method based on M-mode ultrasound according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of an M-mode ultrasound image in an M-mode ultrasound-based left ventricular ejection fraction measurement method according to an embodiment of the present invention, wherein the M-mode ultrasound image includes a plurality of complete cardiac cycles;

FIG. 6 is a schematic diagram showing an indication of the inclusion of multiple end systole and end diastole in an M-mode ultrasound image in an M-mode ultrasound-based left ventricular ejection fraction measurement method according to an embodiment of the present invention;

FIG. 7 is a schematic diagram showing an M-mode ultrasound image in an M-mode ultrasound-based left ventricular ejection fraction measurement method according to an embodiment of the present invention, wherein the M-mode ultrasound image includes a plurality of measurement items and corresponding key points;

the following detailed description will further illustrate the invention with reference to the above-described drawings.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

In the present invention, unless explicitly specified and limited otherwise, the terms "mounted," "connected," "secured," and the like are to be construed broadly and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

Example 1

Referring to fig. 1, a flowchart of a method for measuring left ventricular ejection fraction based on M-mode ultrasound according to a first embodiment of the present invention is shown, for convenience of explanation, only a portion related to the embodiment of the present invention is shown, and the method includes:

Step S10, acquiring an M-type ultrasonic image acquired from a target area, wherein the vertical axis of the M-type ultrasonic image is the depth of different tissues, and the horizontal axis is time;

in one embodiment of the present invention, M-mode ultrasound is an ultrasound imaging method that displays a curve of the distance between interfaces and probes on a certain sound beam in the body over time. It is mainly used for analyzing the movement of heart and large blood vessels, such as the movement of heart wall, blood vessel wall or valve. Wherein the M-mode ultrasound image is a two-dimensional image, the horizontal axis of which represents time and the vertical axis represents depth (i.e. amplitude of motion) of different tissues, i.e. intensity of the reflected echo signal. Wherein each point on the horizontal axis corresponds to a cardiac cycle time and each point on the vertical axis corresponds to a location of the tissue interface. Therefore, the M-type ultrasonic image can reflect the motion curves of tissue interfaces with different depths in different cardiac cycles, so that the functions of the tissues are evaluated.

In one embodiment of the invention, the target area is the heart area of the chest of the human body, and the M-type ultrasonic scanning is performed on the target area through the long axis section of the left ventricle at the left edge of the sternum by adopting a conventional ultrasonic probe and instrument, so that an M-type ultrasonic image is acquired. Wherein the acquired M-mode ultrasound images reflect the systolic and diastolic procedures of the heart, as well as the depth and motion of the different tissues. Wherein fig. 4 is a schematic view of an M-mode ultrasound image acquired in the present embodiment. As shown in fig. 4, the characteristics of ventricular wave group, atrial wave group, mitral valve E-point and a-point, and the motion curves of the left ventricular back wall and the ventricular septum can be observed on the M-type ultrasonic image.

Further, in an embodiment of the present invention, the data input source for acquiring the M-type ultrasound image may be an ultrasound machine, a cloud server for storing the ultrasound image, or other devices for storing the ultrasound image data.

Step S20, analyzing each cardiac cycle information contained in the M-type ultrasonic image according to the acquired M-type ultrasonic image, wherein the cardiac cycle information comprises end systole time and end diastole time;

in one embodiment of the present invention, since the vertical axis of the M-mode ultrasound image represents the depth of the human tissue and the horizontal axis represents the time, motion curves of tissues with different depths within a certain period of time can be observed according to the M-mode ultrasound image, and the motion curves include cardiac cycle information, so that the number of complete cardiac cycles of the M-mode ultrasound image in an effective time range and specific positions of end systole/end diastole moments in each cardiac cycle on the horizontal axis can be obtained through analysis of the M-mode ultrasound image, thereby realizing analysis of the respective cardiac cycle information included in the M-mode ultrasound image. It is shown with reference to fig. 5 indicating that the M-mode ultrasound image includes 4 complete cardiac cycles and with reference to fig. 6 indicating the respective end systole and end diastole times in the M-mode ultrasound image.

The analysis of the cardiac cycle information contained in the M-type ultrasound image may be implemented by various embodiments, and may be performed by a deep learning method, or may be performed by a conventional image processing algorithm, or may be performed by other methods, which are not limited to any one of them, and are set according to actual use needs, and are not specifically limited herein. In the embodiment of the invention, two specific embodiments are specifically adopted to realize analysis of each cardiac cycle information contained in the M-type ultrasonic image.

In a first specific embodiment of the invention, a deep learning method is adopted, a large number of left ventricular M-mode ultrasonic images are collected in clinic, then a professional doctor marks the M-mode ultrasonic images, wherein the marked content comprises specific positions (i.e. x coordinates) of systole end moment/diastole end moment of each cardiac cycle on a transverse axis, on the images, the marked M-mode ultrasonic images are used for training a neural network model, and the trained neural network model can acquire cardiac cycle information of the input M-mode ultrasonic images.

More specifically, first, a large number of left ventricular M-mode ultrasound images are collected in the clinic and the images are labeled by a specialized doctor. The content of the label includes the specific positions on the horizontal axis of the end systole and end diastole moments of each cardiac cycle, i.e. the x-coordinates, and the image contains several complete cardiac cycles. These annotations can be used as true values for training data to guide learning of the neural network. The labeling process should ensure the accuracy and consistency of labeling to avoid the influence of labeling errors or noise on the neural network.

Next, a suitable neural network model is selected for extracting features from the M-mode ultrasound image and predicting the time points of the cardiac cycle. The neural network model may use the structure of a common Convolutional Neural Network (CNN), such as VGGNet, resNet, efficientNet, vision Transformer, etc. These network structures are all based on a stack of convolutional and pooling layers, which can extract features of different levels from the image, from low-level edges and textures to high-level shapes and semantics. Generally, the deeper the network structure, the more features can be extracted, but the more easily the problem of overfitting or gradient disappearance is encountered. Therefore, it is necessary to select an appropriate network structure according to the size and complexity of data and the limitation of computing resources.

Then, a suitable loss function is defined for measuring the difference between the predicted value and the actual value of the neural network, and a suitable optimization algorithm is used for updating the parameters of the neural network to minimize the loss function. Wherein the loss function may use a Mean Square Error (MSE) or a Mean Absolute Error (MAE), which is a loss function commonly used in regression tasks, capable of reflecting the distance between the predicted value and the true value. The optimization algorithm may use random gradient descent (SGD) or its variants, such as Adam, RMSProp, etc., which are all gradient-based optimization algorithms that can adjust parameters of the neural network according to the gradient of the loss function to minimize the loss function.

Furthermore, the neural network model is trained by using the marked M-type ultrasonic image and the corresponding cardiac cycle time point. The training process can be divided into the following steps:

the marked data set is divided into a training set, a verification set and a test set. The training set is used for training the neural network model, the verification set is used for evaluating the performance of the model and adjusting the super parameters of the model, and the testing set is used for finally testing the generalization capability of the model.

The M-type ultrasonic image is preprocessed, such as clipping, scaling, normalization, enhancement and the like, so as to adapt to the input requirement of the neural network model and improve the quality and diversity of data.

The training set is divided into a plurality of small batches (batches), each batch containing a number of M-mode ultrasound images and corresponding cardiac cycle time points. And then inputting each batch into the neural network model in turn to obtain a predicted value, and calculating a loss function. And then updating parameters of the neural network by using an optimization algorithm according to the gradient of the loss function. Wherein the process is called an iteration (iteration).

The iterative process is repeated until all data in the training set is input once through the neural network model. Wherein this process is called an exercise cycle (epoch).

And repeating the training period process until the performance of the neural network model reaches the optimum or no longer improves on the verification set or reaches the preset maximum training period number. Wherein this process is called training phase (training phase).

After the training phase is completed, the neural network model with the best performance on the verification set is selected as the final neural network model. The test set is then used to evaluate the generalization ability of the model, i.e., the performance on unseen data. Wherein the process is called the testing phase.

Finally, the trained neural network model is used for acquiring the cardiac cycle information of the input M-type ultrasonic image. At this time, a new M-type ultrasonic image is input into the trained neural network model to obtain a predicted value of the neural network model, namely, the x coordinate of the moment points of the end systole and the end diastole of each cardiac cycle.

In a second embodiment of the present invention, an image processing method is adopted, specifically, the step of analyzing the cardiac cycle information included in the M-type ultrasound image according to the acquired M-type ultrasound image includes:

Firstly, determining a motion curve of the rear wall of the left ventricle in an M-type ultrasonic image by an image processing method, wherein the motion rule of the motion curve of the rear wall of the left ventricle is consistent with the motion rule of the left ventricle, and then knowing that the position of the rear wall of the left ventricle is more under the M-type ultrasonic image than other moments in a cardiac cycle at the end diastole of the left ventricle; at the end of left ventricular systole, the left ventricular posterior wall position is above the M-ultrasound image than at other times in the cardiac cycle; therefore, the end diastole of the left ventricle is represented when the position of the back wall of the left ventricle is lowest, the end systole of the left ventricle is represented when the position of the back wall of the left ventricle is highest, and the positions correspond to extreme points of the curve, so that extreme value processing is carried out on the motion curve of the back wall of the left ventricle by a derivative or difference method, the maximum point and the minimum point of the curve are found, the maximum value is defined as a candidate point of the end diastole, the minimum value is defined as a candidate point of the end systole, and the extreme points are combined and position information is recorded, namely extreme value information.

Secondly, because there may be some errors or interferences in the extreme points of the curve, not necessarily all the extreme points are the real states of the left ventricle, so that it is necessary to screen and match the extreme points according to some priori knowledge to obtain the real end diastole and end systole moments, specifically, traversing the extreme points in the extreme information from left to right, and forming a point pair by two adjacent extreme points according to the order of maximum/minimum, namely end diastole/end systole. For example, a first maximum point and a first minimum point form a pair of points, a second maximum point and a second minimum point form a pair of points, and so on, which represent the state of the left ventricle for a complete cardiac cycle. Then for each point pair, the lateral distance and the longitudinal distance thereof, i.e. the difference between the abscissa and the ordinate of the two extreme points, are calculated, and then it is determined whether they are valid point pairs or not according to some empirical parameters. The empirical parameters include the time range of the human cardiac cycle [ thr ] _time0 , thr _time1 ]And normal left ventricular systolic diastolic undulating distance range [ thr _depth0 , thr _depth1 ]They can be obtained by statistics or measurements, whose role is to limit the transverse and longitudinal distances of the point pairs within a reasonable range, in particular, Time range of transverse distance [ thr ] of human cardiac cycle _time0 , thr _time1 ]Is limited by the normal left ventricular systolic diastolic undulating distance range [ thr ] _depth0 , thr _depth1 ]The limitation should be excluded if the lateral distance and the longitudinal distance of the point pair are out of range, indicating that the point pair is not a true left ventricular condition. In this case, the pair of valid points can be retained after these restrictions, and is denoted as end diastole time/end systole time.

And finally, determining the information of each cardiac cycle according to each effective point pair, wherein the number of the effective point pairs is the number of the complete cardiac cycles, the position information corresponding to the maximum value of the effective point pair is the end diastole moment, and the position information corresponding to the minimum value of the effective point pair is the end systole moment. The number of complete cardiac cycles in the M-mode ultrasound image and the end systole and end diastole times of each cardiac cycle can thus be determined. The heart rate of the left ventricle, i.e. the number of cardiac cycles per minute, can also be calculated by determining the number of valid point pairs, which is also one of the important indicators reflecting the function of the left ventricle. Since each effective point pair represents the state of the left ventricle of a complete cardiac cycle, the number of the complete cardiac cycles can be obtained by counting the number of the effective point pairs, and then the heart rate of the left ventricle can be calculated according to the sampling frequency and the time length of the M-type ultrasonic image.

Further, in an embodiment of the present invention, the determining the motion curve of the rear wall of the left ventricle according to the acquired M-mode ultrasound image may be implemented by:

The motion curve of the left ventricular back wall (LVPW) of the M-mode ultrasound image is usually clearly visible on the whole time axis, and has high brightness and easy identification, so that the motion curve of the left ventricular back wall is identified through threshold segmentation and gradient detection processing. Specifically, since each column of pixels of the M-mode ultrasound image represents a motion condition of the left ventricle at a moment, the processing object of the method is that each column of the M-mode ultrasound image is traversed in turn from left to right, at first, the whole M-mode ultrasound image is obtained each time, one column of pixels in the M-mode ultrasound image is taken as the motion condition of the left ventricle at the moment, each column in fig. 4-7 represents a pixel column which is a one-dimensional vector, wherein each element represents a gray value of a pixel of a depth, the larger the gray value is, the higher the brightness of a tissue structure representing the depth is, and otherwise, the lower the brightness is.

And secondly, smoothing each pixel column by using a smoothing function to remove noise interference and obtain a new pixel column. In which, since the quality of the M-mode ultrasound image is affected by various factors, there may be some noise or blurring that may affect the subsequent analysis and calculation, a certain smoothing process is performed on each pixel column, so that it is smoother and clearer. The smoothing function can use common image smoothing methods such as mean smoothing or Gaussian smoothing, and the principle is to replace the value of the original pixel with the average value or weighted average value of the neighborhood pixels of the pixel, so as to achieve the effects of noise reduction and enhancement.

Then, the division threshold value of each pixel column after the smoothing processing is calculated respectively, and each pixel column is subjected to threshold processing according to the division threshold value to obtain each new pixel column, specifically, the steps can be realized by the following steps:

Specifically, since the ultrasound image is a gray-scale image, each pixel has only one gray-scale value, at this time, the pixels can be classified into a foreground and a background according to the gray-scale value, where the foreground is a region of interest, such as a tissue structure of the left ventricle, and the background is a region of no interest, such as a background or other organs. In order to realize the segmentation of the foreground and the background, a proper threshold value needs to be found, so that the gray values of the foreground pixels are all larger than the threshold value, and the gray values of the background pixels are all smaller than the threshold value, thereby achieving the maximum inter-class difference. More specifically, it traverses all possible gray values as thresholds, calculates the inter-class variance under each threshold, and then selects the gray value that maximizes the inter-class variance as the optimal threshold, i.e., the segmentation threshold, so that foreground and background segmentation of the pixel columns can be achieved. At this time, each pixel column needs to calculate the corresponding segmentation threshold. Further, based on the segmentation threshold, pixels smaller than the corresponding segmentation threshold in each pixel column are set to zero, so that each new pixel column is obtained. At this time, the gray value of the background pixel is set to zero, which means that the background pixel is ignored, and only the gray value of the foreground pixel is reserved, which means that the foreground pixel is focused, so that the segmentation of the foreground and the background is realized. Thus in the thresholded new pixel column, the pixel gray values of the left ventricular tissue structure are preserved, while the other pixel gray values are set to zero.

Then, each new pixel column is checked according to the gradient detection to carry out convolution processing to obtain a gradient, wherein a part larger than zero is taken as a positive gradient, and a part smaller than zero is taken as a negative gradient, which is called gradient information. The gradient is a vector representing the change degree and direction of the gray value of the pixel in the image, and can be used for image processing tasks such as edge detection, feature extraction and the like. One of the methods for calculating the image gradient is to use a convolution check image to carry out convolution operation to obtain gradient components in the horizontal direction and the vertical direction, and then calculate the gradient information of the image according to the amplitude and the direction formula of the gradient. In particular, in the embodiment of the present invention, a gradient detection kernel filter= [1, 0, -1] is adopted to perform convolution processing, where the gradient detection kernel filter is a one-dimensional convolution kernel, and can be used to calculate gradient components in a vertical direction of a pixel, that is, gray differences between a current pixel and pixels adjacent to the current pixel. For each new pixel column, it can be regarded as a one-dimensional image, and then the gradient detection kernel filter is used for carrying out convolution processing on the image to obtain the gradient, and the specific steps are as follows: the gradient detection kernel filter is aligned with the first pixel of the current new pixel column, then each element in the gradient detection kernel filter is multiplied with the corresponding pixel in the current new pixel column, and then the products are added to obtain the first element of the gradient. And (3) moving the gradient detection kernel filter upwards by one pixel, aligning the gradient detection kernel filter with a second pixel of the current new pixel column, and repeating the steps to obtain a second element of the gradient. And the like until the gradient detection kernel filter is aligned with the last pixel of the current new pixel column, and the last element of the gradient is obtained. At this time, the gradient is the gradient component in the vertical direction of the current new pixel column, which is a one-dimensional vector, and the length of the gradient is the same as that of the current new pixel column. The positive and negative gradient values are divided into positive and negative gradients, wherein when the gray value of the pixel changes, the gradient value also changes, the positive gradient value is positive, and the negative gradient value is negative, and the larger the absolute value of the gradient value is, the more severe the change is, and the more likely the change is an edge.

Then, determining the left ventricular rear wall position in each new pixel column according to the gradient values of the upper edge and the lower edge of each bright band in each new pixel column, wherein the steps can be realized by the following steps:

Specifically, from top to bottom, the number of bright bands of each new pixel column (after thresholding, continuous non-zero values are gathered together to be called a bright band) is counted in sequence, and the number of bright bands and the number of pixels of each bright band are recorded to be called bright band information. Since only the pixel gray value of the organization structure of the left ventricle is non-zero after the threshold processing, the organization structure of the left ventricle in each new pixel column can be found out by traversing and counting the bright bands of each new pixel column from top to bottom, and the organization structure of the left ventricle consists of four parts of the right ventricle, the ventricular septum, the left ventricle and the left ventricle back wall, which are presented as each bright band in the new pixel column, and the pixel number of each bright band reflects the thickness of the part.

Further, each bright band in the bright band information is traversed in each new pixel column, the absolute value of the reverse gradient at the upper edge of each bright band and the absolute value of the forward gradient at the lower edge are counted, and since the brightness of the left ventricular back wall is always maximum and has a certain thickness, the position of the left ventricular back wall bright band can present the condition that the middle is bright and the two sides are dark, in this case, the gradient values of the upper edge and the lower edge of the left ventricular back wall bright band are larger than those of the edges of other bright bands, namely, the absolute value of the reverse gradient of the upper edge is larger, the absolute value of the forward gradient of the lower edge is larger, therefore, by using the magnitude of the gradient values, the judgment of which bright band is the left ventricular back wall bright band is made, in particular, the absolute value of the reverse gradient at the upper edge of each bright band and the absolute value of the forward gradient at the lower edge are added to obtain a bright band probability value, which represents the probability that the bright band is the left ventricular back wall bright band, and the probability that the bright band is larger.

Further, a bright band with the largest bright band probability value and a corresponding bright band serial number are found out from all bright band probability values of each new pixel column, at this time, the bright band is a left ventricular back wall bright band, and the ordinate of the midpoint of the bright band is the left ventricular back wall position of the new pixel column. The ordinate of the midpoint is the average of the sum of the ordinate of the upper edge and the ordinate of the lower edge of the bright band.

Then, traversing the M-type ultrasonic image from left to right, acquiring a set of motion position points of the rear wall of the left ventricle on the whole M-type ultrasonic image, and realizing acquisition of a motion track of the rear wall of the left ventricle in the whole cardiac cycle, wherein the motion curve of the rear wall of the left ventricle presents periodic fluctuation and reflects the contraction and relaxation changes of the left ventricle by traversing each row of pixels and repeating the analysis and calculation steps to acquire the position of the rear wall of the left ventricle of each row of pixels, and then connecting the position points of the rear wall of the left ventricle to form a set of points which represent the motion curve of the rear wall of the left ventricle on the whole M-type ultrasonic image, as shown in fig. 4-7.

And finally, carrying out smoothing treatment on the motion position point set of the rear wall of the left ventricle on the whole M-shaped ultrasonic image, and then connecting to obtain a motion curve of the rear wall of the left ventricle, wherein the motion curve of the rear wall of the left ventricle is smoother and clearer by the smoothing treatment for further eliminating noise and fluctuation, so that the subsequent analysis and calculation are facilitated. The smoothing process may use some common curve smoothing methods, such as moving average method, exponential smoothing method, spline interpolation method, etc., and the principle is to replace the value of the original curve with the average value or weighted average value of the neighborhood points of the curve, so as to achieve the effects of smoothing and fitting.

Step S30, positioning all key points on the M-type ultrasonic image according to the acquired M-type ultrasonic image, wherein the key points comprise position information and category information;

in one embodiment of the invention, among others, the M-mode ultrasound image is an ultrasound technique that shows the structure and motion of the heart, which can be used to measure the size and function of the heart. In order to measure parameters of the heart, specific points on the heart, such as the inner and outer walls of the left ventricle, the front wall of the right ventricle, the front and rear walls of the ventricular septum, etc., are required to be found first, and these points are called key points. The location and type of its keypoints may reflect the morphology and state of the heart. Therefore, when observing left ventricular function based on an M-mode ultrasound image, important indicators are the left ventricular inner diameter, left ventricular back wall thickness, ventricular septum thickness, and right ventricular inner diameter at the left ventricular end systole and end diastole times. When acquiring the measurement items, the clinician firstly acquires key points on the left ventricle, and converts the distances among the key points into corresponding measurement item values. Specifically, the key points of the positioning are a right ventricular anterior point, a ventricular septum posterior point, a left ventricular posterior wall anterior point and a left ventricular posterior wall posterior point. Wherein each moment (i.e., end systole moment and end diastole moment) has 4 line segments, and the 4 line segments are connected, so that the distance between every two key points is one line segment, specifically, the line segment from the anterior point of the right ventricle to the anterior point of the ventricular septum is the right ventricular inner diameter (RVD), the line segment from the anterior point of the ventricular septum to the posterior point of the ventricular septum is the ventricular septum thickness (IVS), the line segment from the posterior point of the ventricular septum to the anterior point of the posterior wall of the left ventricle is the Left Ventricular Inner Diameter (LVID), and the line segment from the anterior point of the posterior wall of the left ventricle to the posterior point of the posterior wall of the left ventricle is the left ventricular wall thickness (LVPW). At this time, all key points are positioned through the M-type ultrasonic image, and the specific key points comprise position information and category information.

All key points on the positioning M-type ultrasonic image can be realized through various specific embodiments, can be analyzed by adopting a deep learning method, can be analyzed by adopting a traditional image processing algorithm, can be analyzed by adopting other methods, and is not limited to any one of the methods, and the key points are set according to actual use requirements, and are not particularly limited. In the embodiment of the invention, two specific embodiments are specifically adopted to realize the positioning of all key points on the M-type ultrasonic image.

In the first specific embodiment of the invention, a deep learning method is adopted, a large number of left ventricle M-shaped ultrasonic images are collected in clinic, and then a professional doctor marks the M-shaped ultrasonic images, wherein the marking content comprises the category of each key point and the specific position on the M-shaped ultrasonic images. And then the marked M-type ultrasonic image is used for training a neural network model, and the trained neural network model can acquire all key points on the input M-type ultrasonic image and position information and category information thereof. The steps used in analyzing the cardiac cycle information included in the M-mode ultrasound image may be generally referred to, and will not be described herein, in which the neural network model may be, for example, RCNN, SSD, YOLO, and these models are all commonly used models for target detection and localization, and may identify different objects from the image, and give specific positions and categories.

In a second embodiment of the present invention, an image processing method is adopted, specifically, the step of locating all key points on the M-type ultrasound image according to the acquired M-type ultrasound image includes:

More specifically, first, a large number of M-mode ultrasound images are collected in the clinic, and the locations and categories of keypoints are labeled on the M-mode ultrasound images by a specialist to form a labeled dataset. Local features of all keypoints are then extracted from the labeled dataset, which may use some commonly used feature extraction algorithms, such as Harr, SIFT, LBP, HOG, etc. The feature extraction algorithm may detect some feature points having significance and invariance, such as corner points, edges, textures, shapes, etc., from the image, and generate a feature vector for each feature point, which is used to describe local information of the feature point, such as gray scale, gradient, direction, shape, etc. Wherein different classes of keypoints should have different feature vectors so that they can be distinguished.

Second, a local feature template is created for each type of keypoint, which may use some commonly used template generation algorithms, such as K-means, PCA, LDA, etc. The template generation algorithm can extract some common features from the feature vectors of each type of key points or reduce the dimension of the features so as to obtain a feature template representing the type of key points. Wherein the dimensions of the feature templates should be the same as, or lower than, the dimensions of the feature vectors to facilitate subsequent comparisons and matches.

Template information is then saved, which may use some commonly used data structures and storage means, such as sets, lists, dictionaries, files, databases, etc. At this time, the feature templates of each type of key points and the corresponding type labels are stored so as to facilitate subsequent retrieval and use. The manner in which it is stored should ensure the integrity and readability of the template information, as well as convenient access and modification.

Then, when testing a new M-mode ultrasound image, a sliding window traversal is used to traverse and take image blocks on the M-mode ultrasound image, which can use some common image processing algorithms, such as OpenCV, PIL, scikit-image, etc. The method comprises the steps of defining a rectangular area with the size of a sliding window which is generally the same as or slightly larger than that of a template, moving the sliding window from the upper left corner of an M-shaped ultrasonic image according to a certain step length along the horizontal direction and the vertical direction, and cutting out an image block in the sliding window after each movement until the whole M-shaped ultrasonic image is traversed. Wherein the size of the image block should be the same as the template size or scaled to be the same as the template size for subsequent comparison and matching.

The local features of the individual image blocks are then extracted with the same feature extraction algorithm, which may use the feature extraction algorithm mentioned previously. By applying the same feature extraction algorithm to each image block, a feature vector is obtained that describes local information of the image block, such as gray scale, gradient, direction, shape, etc. Wherein the dimension of the feature vector should be the same as the dimension of the template or the dimension of the template after conversion to facilitate subsequent comparison and matching.

The local features of the current image block are then compared/matched to the stored classes of local feature template information, which may use some common similarity metric algorithms, such as SSD, NCC, CC, SSIM, etc. By comparing the feature vector of each image block with the local feature templates of each type of key point, it is determined how much distance is between them. If the distance is small, it indicates that the feature of the image block is similar to a local feature template of a certain type, and the center point of the image block can be considered as a key point, and the category of the key point is the category represented by the local feature template. If the distance is large, it is assumed that the features of the image block are not similar to all the local feature templates, and the center point of the image block is not a key point, but some background or noise.

Then, when the distance between the local feature of the image block and a local feature template of a certain class is smaller than a preset threshold value, the image block can be considered to belong to the critical point of the class, and some common threshold selection algorithms such as a fixed threshold value, an adaptive threshold value, an Otsu threshold value and the like can be used. By selecting an appropriate threshold according to the distribution of the similarity scores, the image block is used for judging whether the image block is a key point or not. The threshold value should be selected to ensure the recognition rate and accuracy of the key points, i.e. to identify all key points as much as possible, while avoiding erroneous recognition of image blocks of non-key points.

Finally, the center point of the image block may be considered as the location of the key point, and after determining that the local feature of the image block matches the local feature template of a known key point of a certain class, the image block and the matched local feature template may be considered as belonging to the same class of key point, and then using some common image coordinate transformation algorithms, such as affine transformation, perspective transformation, projective transformation, and the like. And calculating the coordinates of the central point of the image block according to the position and the size of the image block to represent the position information of the key point, and determining the category of the key point according to the matching condition of the image block and the local feature template to represent the type information of the key point. At this time, by performing the same operation on all the image blocks until the whole M-type ultrasonic image is traversed, all the key points in the M-type ultrasonic image, and the position information and the category information thereof can be obtained.

Step S40, screening out each effective key point from all the located key points according to each cardiac cycle information contained in the M-type ultrasonic image;

in one embodiment of the present invention, since the positioning of the keypoints is a relatively difficult task, as many positioning keypoints as possible are considered in practical application, a large number of keypoint results are located in step S30, which is very likely to include false keypoints, so that screening according to a certain priori knowledge is required, so that the false keypoints can be removed, and reliable effective keypoints are reserved. Where a priori knowledge is some common sense of heart structure and motion, in embodiments of the present invention, the available a priori knowledge is as follows:

1. depth is consistent with priori, namely the coordinate of the key point in the longitudinal axis direction of the M-type ultrasonic image:

from knowledge of the heart anatomy, it can be seen that under the long axis tangential plane beside the sternum, the main tissue structures from the beginning of the probe to the deep of the human body are the right ventricle, the ventricular septum, the left ventricle and the left ventricle back wall, so that the front point of the right ventricle is at the position closest to the probe (i.e. the longitudinal axis coordinate is smaller), then the front point of the ventricular septum, the rear point of the left ventricle back wall, the front point of the left ventricle back wall and the rear point of the left ventricle, so that the rear wall thickness of the left ventricle is at the deepest position (i.e. the longitudinal axis coordinate is larger), and meanwhile, the key points have a certain distance difference in depth (the longitudinal axis coordinate) and are not too close to each other or too far within a reasonable range. Furthermore, the depth to which the location of the tissue structure of a normal adult heart belongs is within a range, so that all the located keypoints can be screened based on this range.

2. The time is consistent with the prior, namely the coordinate of the point in the transverse axis direction of the M-type ultrasonic image:

the key points positioned in the embodiment of the invention are the right ventricular front point, the ventricular septum rear point, the left ventricular rear wall front point and the left ventricular rear wall rear point in the end systole moment in one cardiac cycle, and the 5 key points are the same as the end diastole moment, wherein the key points at the same moment have a relatively close time difference, namely the distance in the transverse axis direction of the M-type ultrasonic image is relatively small. For example, in step S30, the position information of a right ventricular anterior point at end systole is located, but there is no other end systole in a period of time before and after the point, and this point is considered to be a false point and should be eliminated.

3. Cardiac cycle prior:

wherein the cardiac cycle information comprises that the M-mode ultrasound image comprises several complete cardiac cycles and the end diastole moment/end systole moment in each cardiac cycle, i.e. the coordinates in the direction of the transversal axis of the M-mode ultrasound image. Since the sampling line of the M-mode ultrasound system is difficult to be kept perpendicular to the right ventricle, the ventricular septum and the left ventricle back wall at the same time when the left ventricle image is actually clinically acquired, the end diastole moment/end systole moment is a time range from the M-mode ultrasound image, so that reasonable key points can be considered to fall in the time ranges, the end systole key points can fall in the end systole time range, the end diastole key points can fall in the end diastole time range, and the distance between each key point and the cardiac cycle moment can be in a reasonable range and cannot be too far or too close. Based on this, unreasonable key points are excluded.

Thus, in combination with the prior, in one embodiment of the present invention, the step of screening each valid keypoint from all the located keypoints according to each cardiac cycle information contained in the M-mode ultrasound image includes:

More specifically, according to the depth consistent priori knowledge, the time consistent priori knowledge and the cardiac cycle priori knowledge, calculating an effective probability for each key point, wherein the effective probability is obtained by combining four factors, and the effective probability is respectively as follows:

1. a first depth priori weight of the keypoint, the weight representing a degree of matching of the depth of the keypoint with a reasonable depth of the tissue structure to which it belongs, the weight being close to 0 when the depth of the keypoint (vertical axis coordinate) deviates from the reasonable depth at which the tissue structure should be located, and close to 1 otherwise, wherein the weight can be calculated using a gaussian function whose parameters are the depth of the keypoint, the reasonable depth of the tissue structure, and a standard deviation.

Specifically, a calculation formula for determining the first depth priori weight of each key point according to the ordinate of all the key points is as follows:

，

2. The second depth of the key point is a priori weighted, the weighted represents the reasonable degree of the depth distance between the key point and the key points of other categories in the neighborhood, when the depth distance (longitudinal distance) from the key points of other categories in the neighborhood to the current key point is too small or too large, the weighted is close to 0, and when the weighted is close to 1 in the reasonable distance, it is noted that the reasonable distances from the key points of different categories to the key points of other categories are different, and the reasonable distances can be obtained through statistics. The weights may be calculated later using a combination of gaussian functions, the parameters of each gaussian function being the depth of the keypoints, the depths of other classes of keypoints within the neighborhood, and a standard deviation, and then the results of all gaussian functions are averaged such that the weights range from 0 to 1.

Specifically, a calculation formula for determining the second depth prior weight of each key point according to the longitudinal distance between all the key points and each key point of other categories in the neighborhood is as follows:

，

wherein the method comprises the steps ofA second depth prior weight for the current keypoint,/->Is standard deviation super parameter->Is the abscissa of the current key point, +.>Is the ordinate of the current key point, +.>For the number of other categories of keypoints in the neighborhood of the current keypoint, +.>Representing the +.>Ordinate of each key point, +.>Is the +.>Reasonable depth of the individual keypoints to the current keypoint, +.>Based on natural base eAn exponential function of the bottom.

3. The time of the key point is a priori weighted, the weighted represents the reasonable degree of the time distance between the key point and the key points of other categories in the neighborhood, when the time distance (transverse distance) between the key point and the key points of other categories in the neighborhood is large, the key point can be considered as false key point which is likely to be detected by error, the weighted is close to 0, otherwise, the weighted is close to 1, wherein the weighted can be calculated by a combination of a plurality of Gaussian functions, the parameter of each Gaussian function is the time of the key point, the time of the key points of other categories in the neighborhood, and a standard deviation, and then the results of all the Gaussian functions are averaged, so that the weighted range is between 0 and 1.

Specifically, a calculation formula for determining the time prior weight of each key point according to the transverse distance between all the key points and each key point of other categories in the neighborhood is as follows:

，

wherein,time prior weight for current key point, +.>Is standard deviation super parameter->Is the abscissa of the current key point, +.>Is the ordinate of the current key point, +.>For the number of other categories of keypoints in the neighborhood of the current keypoint, +.>Is the +.>The abscissa of the individual key points, +.>Based on natural baseeAn exponential function of the base.

4. The cardiac cycle of the keypoint is weighted a priori, which represents a reasonable degree of temporal distance of the keypoint from the end systole moment or end diastole moment of the nearest cardiac cycle, since the specific end systole moment/end diastole moment (transverse axis position) is known, for keypoints belonging to the end systole/end diastole category, when the time difference (transverse distance) between the keypoint and the known end systole moment/end diastole moment is large, the probability that the keypoint belongs to a valid keypoint will be low, where the weight is close to 0, whereas it is close to 1, which weight can be calculated with a gaussian function whose parameters are the time of the keypoint, the time of the nearest cardiac cycle moment, and a standard deviation.

Specifically, a calculation formula for determining the prior weight of the cardiac cycle of each key point according to the transverse distance between all the key points and the abscissa in the information of each cardiac cycle is as follows:

，

Further, in an embodiment of the present invention, effective probability information of each key point is determined in each located key point according to the first depth priori weight, the second depth priori weight, the time priori weight and the cardiac cycle priori weight determined by each key point, and the calculation formula is as follows:

，

wherein,effective probability information for the current key point, < +.>Is the abscissa of the current key point, +.>Threshold +.>Is an empirical parameter, ++>Representing a threshold binarization function, wherein +.>Less than threshold->Time 0 is greater than threshold +.>Unchanged time and->For the probability value of the current key point, +.>First depth a priori weight for current keypoint,/- >A second depth prior weight for the current keypoint,/->Time prior weight for current key point, +.>The cardiac cycle for the current keypoint is weighted a priori. Wherein->When the value of (1) is 0, the current key point is a false key point (i.eLess than threshold->) And->Is a non-zero value (i.e.)>Greater than threshold->). Wherein->And calculating a probability value based on the weight combinations for the current key point.

In one embodiment of the present invention, based on the above formula, the effective probability information of each key point, that is, the probability value that each key point is an effective key point, may be calculated, and then the remaining key points may be subjected to unique screening, that is, only at most one anterior right ventricular point, one anterior ventricular point, one posterior ventricular point, one anterior left ventricular wall point, one posterior left ventricular wall point, and the end diastole may be performed. And reserving a key point with the maximum probability value as a unique key point at the current moment for a plurality of key points with the same category in the same end systole/end diastole, and discarding other key points with lower probability values so as to obtain a screened effective key point result. At this time, in each complete cardiac cycle, there are five active keypoints at end systole and five active keypoints at end diastole.

Step S50, each target key point of each cardiac cycle is matched according to cardiac cycle information contained in the M-type ultrasonic image and each screened effective key point, wherein the target key points are effective key points corresponding to paired end diastole time and end systole time in the cardiac cycle information;

in one embodiment of the present invention, since the left ventricular movement information of a complete cardiac cycle is more valuable to a doctor, for example, in the left ventricular ejection fraction measurement in the embodiment of the present invention, the end systole volume at the end systole time of the left ventricle and the end diastole volume at the end diastole time of the left ventricle need to be acquired, so that the key points are continuously matched through the cardiac cycle information and the screened effective key points, and the matching result only retains each target key point of the complete cardiac cycle, and discards the key points of the incomplete cardiac cycle at the initial or final acquisition stage of the M-type ultrasound image. Therefore, effective key points corresponding to the end diastole time and the end systole time in pairs in each cardiac cycle can be matched in a plurality of cardiac cycles.

When the method is implemented, firstly, according to the analyzed cardiac cycle information, the end diastole time and the end systole time in each cardiac cycle are determined; then, according to the screened effective key points, finding out key points of each category, such as a right ventricle front point, a room interval rear point, a left ventricle rear wall front point and a left ventricle rear wall rear point, wherein each key point has position information and category information; and then matching key points of each category, namely finding out effective key points corresponding to end diastole time and end systole time in pairs in each cardiac cycle, for each cardiac cycle, matching each effective key point according to the information of a complete cardiac cycle of end diastole and end systole, taking the front right ventricular point as an example, firstly determining whether the front right ventricular point exists at the end diastole time in one cardiac cycle, and if the front right ventricular point does not exist in the cardiac cycle, considering that the cardiac cycle has no paired front right ventricular point, and discarding the cardiac cycle. If so, continuously searching whether the right ventricular front point exists at the end systole moment in the current cardiac cycle, if so, judging that the matching is successful, wherein the cardiac cycle can correctly identify the effective key point, specifically the right ventricular front point, at both the end diastole moment and the end systole moment, and if not, judging that the cardiac cycle has no paired right ventricular front points, and discarding the cardiac cycle. By repeating the above steps, and matching other types of keypoints, each target keypoint for each end diastole and end systole of each cardiac cycle can be obtained, wherein the target keypoints for each end diastole and end systole of each cardiac cycle are included as shown in fig. 7.

Step S60, calculating left ventricular ejection fraction of each cardiac cycle according to the matched target key points of each cardiac cycle;

wherein, in one embodiment of the invention, left Ventricular Ejection Fraction (LVEF) is an indicator of the assessment of left ventricular systolic function, which represents the percentage of blood volume expelled during left ventricular systole in each cardiac cycle, as compared to the volume of blood at the end diastole of the left ventricle. Wherein the left ventricular ejection fraction is calculated as follows:

，

wherein the method comprises the steps ofFor left ventricular ejection fraction, EDV (end-diaston volume) is end-diastole volume, i.e. the amount of blood that the left ventricle fills at end-diastole; ESV (end-systole volume) is the end-systole volume, the amount of blood left ventricle remains at the end-systole moment; where the left ventricular ejection fraction is measured using M-mode ultrasound, it is often not possible to directly obtain the values of the end diastole volume and the end systole volume, but rather the value of the Left Ventricular Inner Diameter (LVID), i.e. the diameter of the left ventricle at end diastole and end systole, which can be obtained by the longitudinal distance of the line segment between the post-ventricular point and the anterior point of the left ventricular wall.

Often, values of Left Ventricular Inner Diameter (LVID) are approximately scaled to End Diastole Volume (EDV) and End Systole Volume (ESV), where the scaling formula is as follows:

，

Wherein,left ventricular inner diameter indicative of end diastole, +.>Indicating the left ventricular inner diameter at end systole. It should be noted that the above formula for converting the Left Ventricular Inner Diameter (LVID) into End Diastole Volume (EDV) and End Systole Volume (ESV) is not fixed, and the conversion formula may be slightly different in different hospitals/regions, so that the formula is replaceable, and is set according to actual use requirements, which is not specifically limited herein.

Since the above step S50 matches each target key point including a plurality of complete cardiac cycles, the calculation of the left ventricular ejection fraction is performed for each complete cardiac cycle when calculating the left ventricular ejection fraction.

Wherein, the step of calculating the left ventricular ejection fraction of each cardiac cycle according to the matched target key points of each cardiac cycle comprises the following steps:

Specifically, the target key points of the end diastole time and the end systole time of each cardiac cycle, such as the front right ventricular point, the front ventricular point, the rear ventricular point, the front left ventricular wall point and the rear left ventricular wall point, are matched according to the step S50, wherein each key point has position information and category information.

Then, calculating the left ventricular inner diameter of each cardiac cycle at the end diastole time according to the position information of the key points) And left ventricular inner diameter at end systole time (+)>) Wherein the left ventricular internal diameter is calculated according to the difference of the ordinate of the target key point, and the specific left ventricular internal diameter at the end diastole time (+.>) Is calculated from the difference between the ordinate of the post-ventricular septum point at end diastole and the ordinate of the anterior left ventricular wall point, the left ventricular inner diameter at end systole (>) From the ordinate of the post-ventricular septum point at end-systole and the ordinate of the anterior left ventricular wall pointAnd calculating the difference value.

Then, according to the conversion formula, the left ventricular inner diameter at the end diastole time is calculated ) And left ventricular inner diameter at end systole time (+)>) The values of (2) are converted into the values of End Diastole Volume (EDV) and End Systole Volume (ESV), and then the value of the left ventricular ejection fraction of each cardiac cycle is obtained according to the calculation formula of the left ventricular ejection fraction, or the left ventricular ejection fraction of each cardiac cycle is averaged to obtain the value of the total left ventricular ejection fraction.

Step S70, corresponding display is carried out on the M-type ultrasonic image, each cardiac cycle information, all key points, each effective key point, each target key point and each left ventricular ejection fraction according to the defined display mode;

in one embodiment of the invention, the comprehensive analysis results obtained by combining the M-type ultrasonic image, each cardiac cycle information, all key points, each effective key point, each target key point and each left ventricular ejection fraction are displayed according to a display mode defined by a user for the user to check, so that the interpretability of the embodiment of the invention in clinical use can be greatly improved. The M-shaped ultrasonic image of the left ventricle may contain a plurality of complete cardiac cycles, a user can select to check the left ventricular ejection fraction output by a certain cardiac cycle, and can check intermediate results, including information of each cardiac cycle analyzed by the M-shaped ultrasonic image, all key points of initial positioning, each effective key point after screening and target key points after matching, the display mode is very important for clinic, and a doctor needs to know what the basis for judging by the automatic detection technology is when using the automatic detection technology.

For example, the M-type ultrasound image includes four complete cardiac cycles on a time axis, but in the embodiment of the present invention, only the effective results of two of the complete cardiac cycles are output finally, and the results of the two cardiac cycles have a slight error, firstly, a doctor can choose to check the intermediate result of the cardiac cycle without effective output, and determine what the reason for the output of the final ineffective result is (the reason may be that the image is missing/blurred due to shake in the scanning process of the probe, or that the image acquisition mode is not very standard, and a part of key points are blocked, etc.); for the effective results of the remaining two cardiac cycles, the doctor can check the middle result, and decide which cardiac cycle result to select as the final output according to whether the position of the key point positioning is real, whether the position is accurate and the like. The manner in which this is displayed may be user-defined, with the user selecting to view only the final left ventricular ejection fraction measurements, and also selecting to view the complete or partial intermediate results.

Further, in one embodiment of the present invention, the method further comprises:

a measurement is output according to the left ventricular ejection fraction of each cardiac cycle, wherein the measurement comprises statistical information such as a numerical value, a normal range, an average value, a maximum value, a minimum value, a standard deviation, a variation coefficient and the like of the left ventricular ejection fraction of each cardiac cycle, and evaluation and suggestion of the measurement.

In this embodiment, a measurement is statistically output based on the left ventricular ejection fraction of each cardiac cycle. The method comprises the following specific steps:

carrying out data cleaning on the left ventricular ejection fraction of each cardiac cycle to remove abnormal values and invalid values, and obtaining an effective left ventricular ejection fraction data set;

calculating the statistical information such as the numerical value, the normal range, the average value, the maximum value, the minimum value, the standard deviation, the variation coefficient and the like of the left ventricular ejection fraction of each cardiac cycle according to the effective left ventricular ejection fraction data set, and displaying the statistical information in a table form;

determining a normal range (typically 50% -70%) based on the distribution of the left ventricular ejection fraction of the normal population, and evaluating and suggesting the measurement based on a comparison of the left ventricular ejection fraction of each cardiac cycle to the normal range, e.g., if the left ventricular ejection fraction is below the normal range, suggesting the likelihood of left ventricular contractility, suggesting further examination and treatment; if the left ventricular ejection fraction is within the normal range, it is indicated that the heart is functioning properly and no special treatment is required.

In the embodiment of the invention, based on clinical operation logic, the characteristic that an M-type ultrasonic image contains a plurality of cardiac cycles is combined, the left ventricular ejection fraction calculation problem is converted into a plurality of sub-problems, and each sub-problem is respectively solved by various solutions, wherein each solution can obtain partial effective information, so that the operation time of a doctor can be reduced in clinic, and the burden of the doctor is lightened; the final measurement result is obtained by combining part of effective information obtained by the solutions in a customizable manner, and meanwhile, intermediate results (namely part of effective information) can be checked, so that the output of the embodiment of the invention has interpretability, meanwhile, the output result is stable, part of effective information obtained by the solutions is related to clinical logic, and the method can be used as teaching demonstration, so that doctors unfamiliar with operation can learn to measure the left ventricular ejection fraction under M-type ultrasound.

In summary, in this embodiment, by performing a plurality of steps on the M-type ultrasound image, including cardiac cycle analysis, key point positioning, key point screening, key point matching, left ventricular ejection fraction calculation, etc., the accuracy and reliability of left ventricular ejection fraction measurement are improved, and specifically, the cardiac cycle analysis may analyze each cardiac cycle included in the M-type ultrasound image; the key point positioning can initially position each key point on the M-type ultrasonic image; the key point screening screens invalid false key points by combining cardiac cycle information, and reliable effective key points are reserved; the key point matching can be combined with the cardiac cycle information to match paired target key points in each complete cardiac cycle; the left ventricular ejection fraction is calculated to obtain each target key point through calculation and matching, so that the accurate left ventricular ejection fraction can be obtained, and the reliability of the result is ensured; and through carrying out corresponding display on the M-type ultrasonic image, each cardiac cycle information, all key points, each effective key point, each target key point and each left ventricular ejection fraction according to a display mode defined by a user, the measurement result of the final left ventricular ejection fraction can be checked, and meanwhile, intermediate results such as cardiac cycle information, key points, effective key points and target key points can be checked, so that the interpretability and the visual effect in clinical use are improved, and the problems of insufficient reliability and poor interpretability of the results output by the conventional measurement method are solved.

Example two

Referring to fig. 2, a schematic structural diagram of an M-mode ultrasound-based left ventricular ejection fraction measurement system according to a second embodiment of the present invention is shown, for convenience of illustration, only in relevant parts of the embodiment of the present invention, the system includes:

the input module 11 is used for acquiring an M-type ultrasonic image acquired from a target area, wherein the vertical axis of the M-type ultrasonic image is the depth of different tissues, and the horizontal axis is time;

an analysis module 12, configured to analyze, according to the acquired M-mode ultrasound image, each cardiac cycle information included in the M-mode ultrasound image, where the cardiac cycle information includes a systole end time and a diastole end time;

the positioning module 13 is used for positioning all key points on the M-type ultrasonic image according to the acquired M-type ultrasonic image, wherein the key points comprise position information and category information;

the screening module 14 is used for screening each effective key point from all the located key points according to each cardiac cycle information contained in the M-type ultrasonic image;

the matching module 15 is configured to match each target key point of each cardiac cycle according to cardiac cycle information included in the M-type ultrasound image and each screened effective key point, where the target key points are effective key points corresponding to paired end diastole time and end systole time in the cardiac cycle information;

A calculation module 16 for calculating left ventricular ejection fraction of each cardiac cycle based on the matched target keypoints of each cardiac cycle;

the display module 17 is configured to display the M-type ultrasound image, each cardiac cycle information, all key points, each valid key point, each target key point, and each left ventricular ejection fraction according to a defined display manner.

Further, in one embodiment of the present invention, the analysis module 12 includes:

the curve extremum calculating unit is used for determining a motion curve of the rear wall of the left ventricle according to the acquired M-type ultrasonic image, and carrying out extremum solving processing on the motion curve of the rear wall of the left ventricle to obtain extremum information, wherein the extremum information comprises extremum points and corresponding position information;

the point pair construction unit is used for traversing extreme points in the extreme value information from left to right in sequence and forming two adjacent extreme points into a point pair according to the sequence of the maximum value/the minimum value;

and the cardiac cycle determining unit is used for determining each effective point pair according to the calculated transverse distance and longitudinal distance of all the point pairs and the corresponding preset parameter range, and determining each cardiac cycle information according to each effective point pair, wherein the position information corresponding to the maximum value of the effective point pair is end diastole time, and the position information corresponding to the minimum value of the effective point pair is end systole time.

Further, in one embodiment of the present invention, the curve extremum calculating unit includes:

a pixel column obtaining subunit, configured to sequentially traverse and obtain each column of pixels of the M-type ultrasound image from left to right, and perform smoothing processing on each pixel column;

a pixel column threshold processing subunit, configured to respectively calculate a segmentation threshold of each pixel column after the smoothing processing, and perform threshold processing on each pixel column according to the segmentation threshold to obtain each new pixel column;

the pixel column gradient detection subunit is used for carrying out convolution processing on each new pixel column according to the gradient detection check to obtain gradient information of each new pixel column;

a left ventricle back wall position determining subunit, configured to determine a left ventricle back wall position in each new pixel column according to gradient values of upper and lower edges of each bright band in each new pixel column, where the bright bands are pixels with continuous non-zero values that are clustered together after thresholding;

and the motion curve construction subunit is used for carrying out smoothing processing on the point set formed by the positions of the left ventricular rear wall in each new pixel column and connecting the point set to obtain a motion curve of the left ventricular rear wall.

Further, in one embodiment of the present invention, the pixel column thresholding subunit is configured to:

Further, in one embodiment of the present invention, the left ventricular posterior wall position determination subunit is configured to:

Further, in one embodiment of the present invention, the positioning module 13 includes:

the local feature template establishing unit is used for extracting the local features of all key points according to the existing marked data set, respectively establishing respective local feature templates for each type of key points based on the local features of the key points and carrying out classified storage;

the local feature extraction unit is used for traversing and acquiring each image block on the acquired M-type ultrasonic image and extracting the local feature of each image block;

the local feature comparison unit is used for matching each target image block with the distance from the local feature template being smaller than a preset threshold value according to the comparison of the local features of each image block and the stored various local feature templates;

and the key point determining unit is used for determining the position information and the category information of each key point according to the position information of each target image block and the type information of the matched local feature template.

Further, in one embodiment of the present invention, the screening module 14 includes:

the first depth priori weight determining unit is used for determining the first depth priori weights of all the key points according to the ordinate of all the key points;

The second depth priori weight determining unit is used for determining the second depth priori weight of each key point according to the longitudinal distance between all the key points and each key point of other categories in the neighborhood;

the time priori weight determining unit is used for determining the time priori weight of each key point according to the transverse distance between all the key points and each key point of other categories in the neighborhood;

the cardiac cycle prior weight determining unit is used for determining cardiac cycle prior weights of all key points according to transverse distances between all the key points and the abscissa in the cardiac cycle information;

the effective probability information determining unit is used for determining effective probability information of each key point in the positioned key points according to the first depth priori weight, the second depth priori weight, the time priori weight and the cardiac cycle priori weight determined by each key point, wherein the effective probability information is a probability value of each key point being an effective key point;

the effective key point screening unit is used for screening out each effective key point according to the position information, the category information and the effective probability information of each key point.

Further, in one embodiment of the present invention, the calculation formula of the first depth a priori weight determining unit is:

，

Further, in one embodiment of the present invention, the calculation formula of the second depth a priori weight determining unit is:

，

Further, in one embodiment of the present invention, the calculation formula of the time prior weight determining unit is:

，

wherein,time prior weight for current key point, +.>Is standard deviation super parameter->Is the abscissa of the current key point, +.>Is the ordinate of the current key point, +. >For the number of other categories of keypoints in the neighborhood of the current keypoint, +.>Is the +.>The abscissa of the individual key points, +.>Based on natural baseeAn exponential function of the base.

Further, in one embodiment of the present invention, the calculation formula of the cardiac cycle a priori weight determination unit is:

，

Further, in one embodiment of the present invention, the calculation formula of the effective probability information determining unit is:

，

wherein,effective probability information for the current key point, < +.>Is the abscissa of the current key point, +.>Threshold +.>Is an empirical parameter, ++>Representing a threshold binarization function, in whichLess than threshold->Time 0 is greater than threshold +.>Unchanged time and->For the probability value of the current key point, +.>First depth a priori weight for current keypoint,/->A second depth prior weight for the current keypoint,/->Time prior weight for current key point, +. >The cardiac cycle for the current keypoint is weighted a priori.

Further, in one embodiment of the present invention, the computing module 16 includes:

the left ventricular internal diameter calculation unit is used for calculating the left ventricular internal diameter at the end systole moment and the left ventricular internal diameter at the end diastole moment in each cardiac cycle according to the matched target key points of each cardiac cycle;

a volume calculation unit for calculating a systole volume and a diastole volume in each cardiac cycle from the left ventricular inner diameter at the systole time and the left ventricular inner diameter at the diastole time in each cardiac cycle;

a left ventricular ejection fraction calculation unit for calculating a left ventricular ejection fraction for each cardiac cycle from the end systole volume and the end diastole volume in each cardiac cycle.

The implementation principle and the produced technical effects of the left ventricular ejection fraction measurement system based on M-mode ultrasound provided by the embodiment of the invention are the same as those of the embodiment of the method, and for the sake of brevity, reference is made to the corresponding content in the embodiment of the method.

Example III

In another aspect, referring to fig. 3, an apparatus for measuring left ventricular ejection fraction based on M-mode ultrasound according to a third embodiment of the present invention includes a memory 20, a processor 10, and a computer program 30 stored in the memory 20 and executable on the processor 10, wherein the processor 10 implements the method for measuring left ventricular ejection fraction based on M-mode ultrasound according to the third embodiment of the present invention when executing the computer program 30.

The processor 10 may be, among other things, a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor or other data processing chip for running program code or processing data stored in the memory 20, e.g. executing an access restriction program or the like, in some embodiments.

The memory 20 includes at least one type of readable storage medium including flash memory, a hard disk, a multimedia card, a card memory (e.g., SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. The memory 20 may in some embodiments be an internal memory unit of an M-mode ultrasound based left ventricular ejection fraction measuring device, such as a hard disk of the M-mode ultrasound based left ventricular ejection fraction measuring device. The memory 20 may also be an external memory device of the left ventricular ejection fraction measuring device based on M-mode ultrasound in other embodiments, such as a plug-in hard disk, smart Media Card (SMC), secure Digital (SD) Card, flash Card (Flash Card) or the like provided on the left ventricular ejection fraction measuring device based on M-mode ultrasound. Further, the memory 20 may also include both internal and external memory units of the left ventricular ejection fraction measurement device based on M-mode ultrasound. The memory 20 may be used not only to store application software and various types of data installed in the left ventricular ejection fraction measuring device based on the M-mode ultrasound, but also to temporarily store data that has been output or is to be output.

It should be noted that the configuration shown in fig. 3 does not constitute a limitation of the M-mode ultrasound based left ventricular ejection fraction measuring device, which may include fewer or more components than shown, or may be combined with certain components, or may be arranged with different components, in other embodiments.

The embodiment of the invention also provides a computer readable medium, on which a computer program is stored, which when being executed by a processor, implements the left ventricular ejection fraction measurement method based on M-mode ultrasound as described in the foregoing method embodiment.

Those of skill in the art will appreciate that the logic and/or steps represented in the flow diagrams or otherwise described herein, e.g., a ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

The above examples merely represent a few embodiments of the present invention, which are described in more detail and are not to be construed as limiting the scope of the present invention. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the invention, which are all within the scope of the invention. Accordingly, the scope of protection of the present invention is to be determined by the appended claims.

Claims

1. A left ventricular ejection fraction measurement method based on M-mode ultrasound, the method comprising:

according to the defined display mode, carrying out corresponding display on the M-type ultrasonic image, each cardiac cycle information, all key points, each effective key point, each target key point and each left ventricle ejection fraction;

The step of screening each effective key point from all the located key points according to the information of each cardiac cycle contained in the M-type ultrasonic image comprises the following steps:

screening out each effective key point according to the position information, the category information and the effective probability information of each key point;

The calculation formula for determining the first depth priori weight of each key point according to the ordinate of all key points is as follows:

wherein,first depth a priori weight for current keypoint,/->As the abscissa of the current keypoint,is the ordinate of the current key point, +.>Is standard deviation super parameter->A reasonable depth of the tissue structure of the category corresponding to the current key point, <' > is given>Is an exponential function based on a natural base e.

2. The method of M-mode ultrasound based left ventricular ejection fraction measurement of claim 1, wherein the step of analyzing the respective cardiac cycle information contained in the M-mode ultrasound image from the acquired M-mode ultrasound image comprises:

3. The method of M-mode ultrasound based left ventricular ejection fraction measurement of claim 2, wherein the step of determining a motion profile of the left ventricular posterior wall from the acquired M-mode ultrasound image comprises:

4. The method for measuring left ventricular ejection fraction based on M-mode ultrasound as claimed in claim 3, wherein the steps of calculating the division threshold value of each pixel column after the smoothing process, and thresholding each pixel column according to the division threshold value to obtain each new pixel column include:

5. The method of M-mode ultrasound based left ventricular ejection fraction measurement of claim 3, wherein the step of determining the left ventricular posterior wall position in each new pixel column from the gradient values of the upper and lower edges of each bright band in each new pixel column comprises:

6. The method of M-mode ultrasound based left ventricular ejection fraction measurement of claim 1, wherein the step of locating all keypoints on the M-mode ultrasound image from the acquired M-mode ultrasound image comprises:

7. The left ventricular ejection fraction measurement method based on M-mode ultrasound according to claim 1, wherein the calculation formula for determining the second depth a priori weight of each key point according to the longitudinal distance between all key points and each key point of other categories in the neighborhood is:

wherein,a second depth prior weight for the current keypoint,/->Is the standard deviation super-parameter of the standard deviation,is the abscissa of the current key point, +.>Is the ordinate of the current key point, +.>For the number of other categories of keypoints in the neighborhood of the current keypoint, +.>Representing the +.>Ordinate of each key point, +.>Is the +.>Reasonable depth of the individual keypoints to the current keypoint, +.>Is an exponential function based on a natural base e.

8. The left ventricular ejection fraction measurement method based on M-mode ultrasound according to claim 1, wherein the calculation formula for determining the time prior weight of each key point according to the lateral distance between all key points and each key point of other categories in the neighborhood is:

wherein,time prior weight for current key point, +.>Is standard deviation super parameter->Is the abscissa of the current key point, +.>Is the current key point Ordinate of>For the number of other categories of keypoints in the neighborhood of the current keypoint, +.>Is the +.>The abscissa of the individual key points, +.>Based on natural baseeAn exponential function of the base.

9. The left ventricular ejection fraction measurement method based on M-mode ultrasound according to claim 1, wherein the calculation formula for determining the cardiac cycle a priori weight of each key point according to the lateral distance between all key points and the abscissa in each cardiac cycle information is:

wherein,prior weight for cardiac cycle for the current key point,/->Is standard deviation super parameter->Is the abscissa of the current key point, +.>Is the ordinate of the current key point, +.>Is the distanceThe current key point is the abscissa of the last systole or diastole moment in the latest cardiac cycle information.

10. The left ventricular ejection fraction measurement method based on M-mode ultrasound according to claim 1, wherein the calculation formula for determining the effective probability information of each key point among the located key points according to the first depth priori weight, the second depth priori weight, the time priori weight and the cardiac cycle priori weight determined by each key point is as follows:

wherein, Effective probability information for the current key point, < +.>Is the abscissa of the current key point, +.>Threshold +.>Is an empirical parameter, ++>Representing a threshold binarization function, wherein +.>Less than threshold->Time 0 is greater than threshold +.>Unchanged time and->Is the probability value of the current keypoint,first depth a priori weight for current keypoint,/->A second depth prior weight for the current keypoint,/->Time prior weight for current key point, +.>The cardiac cycle for the current keypoint is weighted a priori.

11. The method for M-mode ultrasound based left ventricular ejection fraction measurement of claim 1, the step of calculating the left ventricular ejection fraction of each cardiac cycle from the matched target keypoints of each cardiac cycle comprising:

12. A left ventricular ejection fraction measurement system based on M-mode ultrasound, the system comprising:

wherein,first depth a priori weight for current keypoint,/->As the abscissa of the current keypoint,is the ordinate of the current key point, +. >Is standard deviation super parameter->A reasonable depth of the tissue structure of the category corresponding to the current key point, <' > is given>Is an exponential function based on a natural base e;

13. An M-mode ultrasound based left ventricular ejection fraction measurement device, characterized in that the device comprises a memory, a processor and a computer program stored on the memory and executable on the processor, which processor executes the M-mode ultrasound based left ventricular ejection fraction measurement method according to any one of claims 1 to 11 when the computer program is executed.