CN115035028A - Left ventricular ejection fraction automatic calculation method based on ultrasonic image - Google Patents
Left ventricular ejection fraction automatic calculation method based on ultrasonic image Download PDFInfo
- Publication number
- CN115035028A CN115035028A CN202210479098.8A CN202210479098A CN115035028A CN 115035028 A CN115035028 A CN 115035028A CN 202210479098 A CN202210479098 A CN 202210479098A CN 115035028 A CN115035028 A CN 115035028A
- Authority
- CN
- China
- Prior art keywords
- left ventricle
- model
- diastole
- calculating
- echocardiogram
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/06—Measuring blood flow
- A61B8/065—Measuring blood flow to determine blood output from the heart
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/5215—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10132—Ultrasound image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30048—Heart; Cardiac
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Pathology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Heart & Thoracic Surgery (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Geometry (AREA)
- Cardiology (AREA)
- Hematology (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
- Image Processing (AREA)
Abstract
The invention provides an automatic calculation method of left ventricular ejection fraction based on an ultrasonic image, and relates to the technical field of computer vision. Firstly, acquiring a plurality of echocardiograms with view types, left ventricle masks, mitral valve rings and apex key point labels as a sample data set; then constructing a deep neural network model for view classification, left ventricle segmentation and mark point detection, and training and verifying the deep neural network model by using a sample data set to obtain optimal model parameters; loading the optimal model parameters into the constructed deep neural network model, and inputting an ultrasonic image to be evaluated to obtain a view category, a left ventricle mask and a key point coordinate; and finally, calculating the left ventricular ejection fraction based on a biplane Simpson's method.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to an automatic calculation method for left ventricular ejection fraction based on an ultrasonic image.
Background
Left ventricular contractile function is an important parameter for the evaluation of cardiovascular disease. The clinically preferred indicator for assessing Left Ventricular systolic function is Left Ventricular Ejection Fraction (LVEF). Clinically calculating LVEF requires a doctor to manually trace the left ventricular intima, and is affected by the difference in the quality of the ultrasound image, the difference between observers, and the calculation method based on manual tracing has poor repeatability, thereby reducing the accuracy of clinical diagnosis. To reduce these errors in clinical studies, it is often time consuming and labor intensive to require multiple physicians to repeatedly calculate LVEF for each patient.
There are studies on the automatic acquisition of LVEF using a deep neural network model, mainly involving two different technical routes: (1) LVEF was predicted directly using a video classification model. According to the method, an ultrasonic video sequence is used as input, the LVEF is used as a label, and the LVEF of a segment of image sequence can be directly predicted by training a video classification network. Although this method can really achieve end-to-end prediction without any post-processing, it requires enormous amount of data to obtain accurate results, which requires long, extensive data accumulation and is difficult for most research institutes to do. (2) The LVEF was calculated by extracting the left ventricle in Apical Four Chamber heart view (A4C) and Apical Two Chamber heart view (A2C) using an image segmentation model. The method takes an ultrasonic image as an input, a left ventricle mask as a label, and extracts a left ventricle from an echocardiogram through a training image segmentation network. Selecting a left ventricle mask corresponding to two images of End Diastole (ED) and End Systole (ES), finding the positions of a mitral valve annulus and a cardiac apex, calculating an End Diastole Volume (EDV) and a systolic membrane Volume (ESV) by using a biplane Simpson's method, and finally obtaining the LVEF. Although these studies are aimed at calculating LVEF, they are often stopped by left ventricular segmentation and do not provide a mitral annulus, a method for determining the apex location, and a process for calculating EF. In summary, these methods can automatically obtain the results related to LVEF, but are difficult to be popularized in practical clinical applications.
The method for clinically calculating the left ventricular ejection fraction needs manual tracing of the left ventricular intima, and has time-consuming and labor-consuming operation and poor repeatability. Some methods proposed by related researches are based on huge data scale, data set establishment is difficult, and some methods only provide left ventricle segmentation results and do not realize real end-to-end calculation.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides an automatic calculation method of left ventricular ejection fraction based on an ultrasonic image.
An automatic calculation method for left ventricular ejection fraction based on an ultrasonic image specifically comprises the following steps:
step 1: acquiring echocardiograms of A4C and A2C views, and a left ventricle mask corresponding to each echocardiogram, wherein the coordinate labels of the marking points of the apex, the ventricular septum mitral annulus and the side wall mitral annulus are used as sample data sets;
the obtained echocardiogram is a single-frame image with time continuity analyzed from a dcm file following a medical digital image transmission protocol, and an image sequence formed by the echocardiograms according to a time sequence is an echocardiogram sequence; the left ventricle mask is a binary image which is marked with a left ventricle area and corresponds to two frames of the echocardiograms ED and ES, wherein the gray value of a background pixel is 0, and the gray value of the left ventricle area pixel is 1; the mark point coordinate labels are the mark point coordinates of the apex, the ventricular septum mitral valve ring and the side wall mitral valve ring corresponding to the two frames ED and ES of the ultrasonic cardiogram;
step 2: initializing a deep neural network model for ultrasonic view identification, left ventricle segmentation and marking point detection, and pre-training the deep neural network model by using a sample data set to obtain a pre-training model;
the deep neural network model for ultrasonic view identification, left ventricle segmentation and marking point detection is a multi-task learning model, and classification, segmentation and marking point detection results can be obtained simultaneously;
the model consists of a backbone network and three decoders; the backbone network is a ResNet50 model which uses hole convolution and removes a full connection layer; the three decoders include a classification decoder, a segmentation decoder and a mark point detection decoder; wherein, the classification decoder is a simple full connection layer and is used for identifying the view type of the echocardiogram; the segmentation decoder consists of a cavity convolution pyramid and is used for segmenting the left ventricle; the mark point detection decoder consists of three upsampling convolutional layers and is used for positioning the position of a mark point;
and 3, step 3: loading model parameters and configuration files of a pre-training model, and inputting an echocardiogram sequence which needs to calculate ejection fraction into the model one by one; the model automatically identifies the view category of the ultrasonic cardiogram, segments the left ventricle and detects the positions of the marking points of the apex, the interacvular septum mitral valve ring and the side wall mitral valve ring;
the model parameters of the pre-training model are the model parameters obtained by training in the step 2, and the configuration file is a file for storing the learning rate, the Batch size and the input image size set by the model operation;
the echocardiogram sequence needing to calculate the ejection fraction is the echocardiogram sequence which is obtained in the step 1 and does not participate in training the deep neural network model;
and 4, step 4: calculating the area of the left ventricle by using the segmentation result, finding a plurality of maximum values and minimum values of the area of the left ventricle in the echocardiogram sequence, and respectively taking the echocardiograms corresponding to the adjacent maximum values and minimum values as images of an end diastole ED and an end systole ES;
determining an end diastole and an end systole; firstly, calculating the area of the left ventricle segmentation result of each image in the whole echocardiogram sequence, wherein the method takes the number of pixels with the gray value of 1 in the segmentation result image as the area of the left ventricle; traversing the area of the left ventricle in the echocardiogram sequence, finding a plurality of maximum values and minimum value points, and determining the end diastole and the end systole of a single cardiac cycle by using the adjacent maximum values and minimum values;
and 5: on the two images of the determined end diastole and end systole, using a connecting line of the midpoint of two markers of the ventricular septum mitral valve ring and the lateral wall mitral valve ring and the apex marker as a long axis of the left ventricle, equally dividing the long axis of the left ventricle into a line segments, taking an extension line perpendicular to the long axis of the left ventricle on the equally divided points to the inner membrane of the left ventricle, and respectively storing the coordinates of the intersection points of each extension line and the inner membrane of the left ventricle according to four conditions of the end diastole of A4C, the end systole of A4C, the end diastole of A2C and the end systole of A2C;
determining the left ventricular base and long axis; using a connecting line segment of two marking points of the mitral valve ring as a left ventricle base; the connecting line of the central point of the two marking points of the mitral valve ring and the apex marking point is used as the long axis of the left ventricle, and the method for acquiring the coordinate of the central point comprises the following steps:
wherein the content of the first and second substances,is the coordinate of the midpoint of the end diastole,is the coordinate of the midpoint at the end of systole,the mitral annulus marker coordinates for the end diastole ventricular septum,marking point coordinates for the end-systolic ventricular septum mitral valve annulus,the coordinates of the marker points for the lateral wall mitral annulus at end diastole,marking point coordinates for the mitral valve annulus at the last systole phase;
step 6: based on the intersection points stored in the step 5, respectively calculating the end diastole volume and the end systole volume of the left ventricle by using a biplane Simpson's method, and then calculating the percentage of the absolute value of the difference value of the two volumes to the end diastole volume so as to obtain the LVEF;
calculating the volumes of the left ventricles at the end systole and the end diastole according to the intersection point coordinates stored in the step 5, and then calculating the percentage of the absolute value of the difference value of the volumes of the left ventricles at the end systole and the end diastole and the volume of the left ventricle at the end diastole so as to obtain the LVEF; based on the stored intersection point coordinates, estimating the left ventricular end diastole volume and the end systole volume by using a biplane Simpson's method, and calculating the percentage of the absolute value of the difference value of the two volumes to the end diastole volume so as to obtain the LVEF:
wherein V is the left ventricular volume, L max Is A4C and A2Maximum of major axis length in C view, a i And b i V is the length of the line segment connected with the intersection point corresponding to the same time of A4C and A2C ES To contract end volume, V ED End diastolic volume.
The invention has the beneficial technical effects that:
the invention provides an automatic LVEF calculation method based on an ultrasonic image, aiming at the problems that manual tracing is needed for calculating the score of blood ejected from a left ventricle, the calculation result is influenced by the clinical experience of a doctor and the repeatability is poor, a deep neural network is used for automatically carrying out view classification, left ventricle segmentation and marking point detection, the learning and time cost of tracing data is reduced, the result difference caused by the subjective experience of the doctor is avoided, and the automatic LVEF calculation is realized.
The invention aims to provide an ultrasonic image-based LVEF automatic calculation method, which uses a deep neural network to automatically identify the A4C and A2C views of an echocardiogram, segment the left ventricle and detect the position of the mitral valve ring and the apex so as to calculate the LVEF of a single cardiac cycle. The result error caused by subjective difference is overcome, and the time and the learning cost of a doctor are reduced.
By adopting the automatic LVEF calculation method, the automatic calculation of the ejection fraction is realized by using a deep learning method, and the test is carried out on a plurality of echocardiograms, and the result shows that the LVEF can be calculated more accurately on the ultrasonic images with different qualities, and the calculation speed meets the real-time requirement. The method has the advantages of simple implementation, high operation speed, high result accuracy, no need of manual interaction in the processing process and capability of meeting the application requirements.
Drawings
FIG. 1 is a flowchart illustrating a method for automatically calculating left ventricular ejection fraction based on ultrasound images according to an embodiment of the present invention;
FIG. 2 is a diagram of a deep neural network model architecture according to an embodiment of the present invention;
FIG. 3 is a sample data set diagram according to an embodiment of the present invention. Wherein a is a view of A4C, b is a view of A2C, c is the left ventricle mask, d is the coordinate label of the mark point, and e is the schematic diagram of the position of the mark point.
FIG. 4 is a schematic diagram of an embodiment of the present invention for calculating the area of the left ventricular segmentation result in the ultrasound sequence and determining the end diastole ED and the end systole ES;
FIG. 5 is a schematic diagram of the left ventricle in an ED and ES frame image partitioned in accordance with an embodiment of the present invention;
FIG. 6 is a schematic diagram of calculating left ventricular end-diastolic volume and end-systolic volume and calculating left ventricular ejection fraction according to an embodiment of the present invention.
Detailed Description
The invention is further explained below with reference to the figures and examples;
the invention provides an LVEF automatic calculation method based on an ultrasonic image, which automatically identifies an input echocardiogram view, segments a left ventricle in a whole ultrasonic image sequence, detects and tracks the positions of two marking points of a mitral valve ring and a cardiac apex marking point, and automatically determines images of a diastole end and a systole end by utilizing a segmentation result so as to calculate the LVEF of a single cardiac cycle.
An automatic calculation method for left ventricular ejection fraction based on ultrasound image is shown in fig. 1, and specifically includes the following steps:
calculating ejection fraction using the biplane Simpson's method requires segmenting the left ventricle in echocardiography A4C and A2C views and detecting the location of the apex, the interventricular mitral valve annulus and the lateral mitral valve annulus. Therefore, the present invention uses the echocardiograms of the two views A4C and A2C, and the corresponding left ventricle mask, apex, septal mitral annulus, and lateral mitral annulus marker point coordinate labels of each echocardiogram as the sample data set.
Step 1: acquiring echocardiograms of A4C and A2C views, and a left ventricle mask corresponding to each echocardiogram, wherein the coordinate labels of the marking points of the apex, the ventricular septum mitral annulus and the side wall mitral annulus are used as sample data sets; a sample data set schematic is shown in fig. 3.
The obtained echocardiograms are single-frame images with time continuity analyzed from a dcm file following a medical digital image transmission protocol, and an image sequence formed by the echocardiograms according to a time sequence is an echocardiogram sequence; the left ventricle mask is a binary image which is marked with a left ventricle area and corresponds to two frames of the echocardiograms ED and ES, wherein the gray value of a background pixel is 0, and the gray value of the left ventricle area pixel is 1; the mark point coordinate labels are the mark point coordinates of the apex, the ventricular septum mitral valve ring and the side wall mitral valve ring corresponding to the two frames of the echocardiograms ED and ES;
step 2: initializing a deep neural network model for ultrasonic view identification, left ventricle segmentation and marking point detection, and pre-training the deep neural network model by using a sample data set to obtain a pre-training model; the structure diagram of the deep neural network model is shown in figure 2.
The deep neural network model for ultrasonic view identification, left ventricle segmentation and marking point detection is a multi-task learning model, and classification, segmentation and marking point detection results can be obtained simultaneously;
the model consists of a backbone network and three decoders. The backbone network is a ResNet50 model that uses hole convolution and removes the full connectivity layer. The three decoders include a classification decoder, a segmentation decoder and a marker detection decoder. Wherein the classification decoder is a simple full connection layer for identifying the view type of the echocardiogram. The segmentation decoder consists of a hole convolution pyramid, used to segment the left ventricle. The mark point detection decoder consists of three upsampling convolutional layers and is used for positioning the position of a mark point;
and 3, step 3: loading model parameters and configuration files of a pre-training model, and inputting an echocardiogram sequence which needs to calculate ejection fraction into the model one by one; the model automatically identifies the view type of the echocardiogram, segments the left ventricle and detects the positions of the marking points of the apex, the ventricular septum mitral valve ring and the side wall mitral valve ring;
the echocardiogram sequence needing to calculate the ejection fraction is the echocardiogram sequence which is obtained in the step 1 and does not participate in training the deep neural network model;
the model parameters of the pre-training model are the model parameters obtained by training in the step 2, and the configuration file is a file for storing the learning rate, the Batch size and the input image size set by the model operation;
and 4, step 4: calculating the area of the left ventricle by using the segmentation result, finding a plurality of maximum values and minimum values of the area of the left ventricle in the echocardiogram sequence, and respectively taking the echocardiograms corresponding to the adjacent maximum values and minimum values as images of an end diastole ED and an end systole ES;
determining an end diastole and an end systole; firstly, calculating the area of the left ventricle segmentation result of each image in the whole echocardiogram sequence, wherein the method takes the number of pixels with the gray value of 1 in the segmentation result image as the area of the left ventricle; traversing the area of the left ventricle in an echocardiogram sequence, finding a plurality of maximum values and minimum value points, and determining the end diastole and the end systole of a single cardiac cycle by using the adjacent maximum values and minimum values; as shown in fig. 4.
And 5: on the two images of the determined end diastole and end systole, using a connecting line of the midpoint of two markers of the ventricular septum mitral valve ring and the lateral wall mitral valve ring and the apex marker as a long axis of the left ventricle, equally dividing the long axis of the left ventricle into a line segments, taking an extension line perpendicular to the long axis of the left ventricle on the equally divided points to the inner membrane of the left ventricle, and respectively storing the coordinates of the intersection points of each extension line and the inner membrane of the left ventricle according to four conditions of the end diastole of A4C, the end systole of A4C, the end diastole of A2C and the end systole of A2C; FIG. 5 is a drawing illustration of the left ventricle in a partitioned ED and ES frame image; wherein a is typically 20;
determining the left ventricular base and long axis; using the section of the line connecting the two marking points of the mitral valve ring as the left ventricle base; the connecting line of the central point of the two marking points of the mitral valve ring and the apex marking point is used as the long axis of the left ventricle, and the method for acquiring the coordinate of the central point comprises the following steps:
wherein the content of the first and second substances,is the coordinate of the midpoint of the end diastole,is the coordinate of the midpoint at the end of systole,the coordinates of the points are marked for the end diastole ventricular septum mitral annulus,marking point coordinates for the end-systolic ventricular septum mitral valve annulus,the coordinates of the marker points for the lateral wall mitral annulus at end diastole,marking point coordinates for the mitral valve annulus on the lateral wall at the end of systole;
step 6: based on the intersection points stored in the step 5, respectively calculating the end diastole volume and the end systole volume of the left ventricle by using a biplane Simpson's method, and then calculating the percentage of the absolute value of the difference value of the two volumes to the end diastole volume so as to obtain the LVEF; calculating the left ventricular end-diastolic volume and end-systolic volume, and then calculating the left ventricular ejection fraction as shown in fig. 6;
calculating the volumes of the left ventricles at the end systole and the end diastole according to the intersection point coordinates stored in the step 5, and then calculating the percentage of the absolute value of the difference value of the volumes of the left ventricles at the end systole and the end diastole and the volume of the left ventricle at the end diastole so as to obtain the LVEF; based on the stored intersection point coordinates, estimating the left ventricular end diastole volume and the end systole volume by using a biplane Simpson's method, and calculating the percentage of the absolute value of the difference value of the two volumes to the end diastole volume so as to obtain the LVEF:
wherein V is the left ventricular volume, L max Is the maximum of the major axis length in the A4C and A2C views, a i And b i V is the length of the line segment connected by the intersection points corresponding to the same time points of A4C and A2C ES To contract end volume, V ED End diastolic volume.
Claims (7)
1. An automatic calculation method for left ventricular ejection fraction based on an ultrasonic image is characterized by comprising the following steps:
step 1: acquiring echocardiograms of A4C and A2C views, and a left ventricle mask corresponding to each echocardiogram, wherein the coordinate labels of the marking points of the apex, the ventricular septum mitral annulus and the side wall mitral annulus are used as sample data sets;
step 2: initializing a deep neural network model for ultrasonic view identification, left ventricle segmentation and marking point detection, and pre-training the deep neural network model by using a sample data set to obtain a pre-training model;
and step 3: loading model parameters and configuration files of a pre-training model, and inputting an echocardiogram sequence which needs to calculate ejection fraction into the model one by one; the model automatically identifies the view type of the echocardiogram, segments the left ventricle and detects the positions of the marking points of the apex, the ventricular septum mitral valve ring and the side wall mitral valve ring;
and 4, step 4: calculating the area of the left ventricle by using the segmentation result, finding a plurality of maximum values and minimum values of the area of the left ventricle in the echocardiogram sequence, and respectively taking the echocardiograms corresponding to the adjacent maximum values and minimum values as images of an end diastole ED and an end systole ES;
and 5: on the two images of the determined end diastole and end systole, using a connecting line of the midpoint of two marking points of the ventricular septum mitral valve ring and the lateral wall mitral valve ring and the apex marking point as a long axis of the left ventricle, equally dividing the long axis of the left ventricle into a line segments, taking an extension line perpendicular to the long axis of the left ventricle on the equally divided points to the left ventricle inner membrane, and respectively storing the coordinates of the intersection points of each extension line and the left ventricle inner membrane according to four conditions of the end diastole of A4C, the end systole of A4C, the end diastole of A2C and the end systole of A2C;
step 6: and (5) respectively calculating the end diastole volume and the end systole volume of the left ventricle by using a biplane Simpson's method based on the intersection points stored in the step (5), and then calculating the percentage of the absolute value of the difference value of the two volumes to the end diastole volume so as to obtain the LVEF.
2. The method according to claim 1, wherein the echocardiograms obtained in step 1 are single-frame images with time continuity parsed from a dcm file conforming to the medical digital image transmission protocol, and the image sequence composed of the echocardiograms in time sequence is an echocardiogram sequence; the left ventricle mask is a binary image which is marked with a left ventricle area and corresponds to two frames of the echocardiograms ED and ES, wherein the gray value of a background pixel is 0, and the gray value of the left ventricle area pixel is 1; the mark point coordinate labels are the mark point coordinates of the apex, the ventricular septum mitral valve ring and the side wall mitral valve ring corresponding to two frames of the echocardiograms ED and ES.
3. The method for automatically calculating the left ventricular ejection fraction based on the ultrasonic image according to claim 1, wherein the deep neural network model for ultrasonic view identification, left ventricular segmentation and marker point detection in step 2 is a multitask learning model capable of simultaneously obtaining classification, segmentation and marker point detection results;
the model consists of a backbone network and three decoders; the backbone network is a ResNet50 model which uses hole convolution and removes a full connection layer; the three decoders include a classification decoder, a segmentation decoder and a mark point detection decoder; wherein, the classified decoder is a simple full connection layer for identifying the view type of the echocardiogram; the segmentation decoder consists of a cavity convolution pyramid and is used for segmenting the left ventricle; the mark point detection decoder consists of three upsampling convolutional layers and is used for positioning the position of the mark point.
4. The method for automatically calculating the left ventricular ejection fraction based on the ultrasound image according to claim 1, wherein the model parameters of the pre-trained model in the step 3 are the model parameters obtained by training in the step 2, and the configuration file is a file for storing a learning rate, a Batch size and an input image size set by the running of the model;
the echocardiogram sequence needing to calculate the ejection fraction is the echocardiogram sequence which is obtained in the step 1 and does not participate in training the deep neural network model.
5. The method for automatically calculating the left ventricular ejection fraction based on the ultrasound image as claimed in claim 1, wherein the step 4 is specifically:
determining an end diastole and an end systole; firstly, calculating the area of the left ventricle segmentation result of each image in the whole echocardiogram sequence, wherein the method takes the pixel number with the gray value of 1 in the segmentation result image as the area of the left ventricle; the left ventricular area is then traversed in the echocardiogram sequence to find a plurality of maxima and minima points, and the end diastole and end systole of a single cardiac cycle are determined from adjacent maxima and minima.
6. The method for automatically calculating the left ventricular ejection fraction based on the ultrasound image as claimed in claim 1, wherein the step 5 is specifically:
determining the left ventricular base and long axis; using the section of the line connecting the two marking points of the mitral valve ring as the left ventricle base; the connecting line of the central point of the two marking points of the mitral valve ring and the apex marking point is used as the long axis of the left ventricle, and the method for acquiring the coordinate of the central point comprises the following steps:
wherein, the first and the second end of the pipe are connected with each other,is the coordinate of the midpoint of the end diastole,is the coordinate of the midpoint at the end of systole,the mitral annulus marker coordinates for the end diastole ventricular septum,marking point coordinates for the end-systolic ventricular septum mitral valve annulus,the coordinates of the marker points for the lateral wall mitral annulus at end diastole,the coordinates of the points are marked for the end-systolic sidewall mitral annulus.
7. The method for automatically calculating the left ventricular ejection fraction based on the ultrasound image as claimed in claim 1, wherein step 6 is specifically:
calculating the volumes of the left ventricles at the end stage of systole and the end stage of diastole according to the intersection point coordinates stored in the step 5, and then calculating the percentage of the absolute value of the difference value of the volumes of the left ventricles at the end stage of systole and the end stage of diastole and the volume of the left ventricles at the end stage of diastole, thereby obtaining the LVEF; based on the stored intersection point coordinates, estimating the left ventricular end diastole volume and the end systole volume by using a biplane Simpson's method, and calculating the percentage of the absolute value of the difference value of the two volumes to the end diastole volume so as to obtain the LVEF:
wherein V is the left ventricular volume, L max Is the maximum of the major axis length in the A4C and A2C views, a i And b i V is the length of the line segment connected with the intersection point corresponding to the same time of A4C and A2C ES To contract end volume, V ED End diastolic volume.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210479098.8A CN115035028A (en) | 2022-05-05 | 2022-05-05 | Left ventricular ejection fraction automatic calculation method based on ultrasonic image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210479098.8A CN115035028A (en) | 2022-05-05 | 2022-05-05 | Left ventricular ejection fraction automatic calculation method based on ultrasonic image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115035028A true CN115035028A (en) | 2022-09-09 |
Family
ID=83119106
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210479098.8A Pending CN115035028A (en) | 2022-05-05 | 2022-05-05 | Left ventricular ejection fraction automatic calculation method based on ultrasonic image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115035028A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115587971A (en) * | 2022-09-21 | 2023-01-10 | 四川大学华西医院 | Method and system for monitoring body reaction and hemodynamics based on heart ultrasonic segmental motion |
CN117918889A (en) * | 2024-03-20 | 2024-04-26 | 中国医学科学院北京协和医院 | Automatic calculation method and device for left ventricular cardiac output of transesophageal echocardiography four-chamber cardiac tangential plane |
-
2022
- 2022-05-05 CN CN202210479098.8A patent/CN115035028A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115587971A (en) * | 2022-09-21 | 2023-01-10 | 四川大学华西医院 | Method and system for monitoring body reaction and hemodynamics based on heart ultrasonic segmental motion |
CN115587971B (en) * | 2022-09-21 | 2023-10-24 | 四川大学华西医院 | Organism reaction and hemodynamic monitoring method and system based on heart ultrasonic segment activity |
CN117918889A (en) * | 2024-03-20 | 2024-04-26 | 中国医学科学院北京协和医院 | Automatic calculation method and device for left ventricular cardiac output of transesophageal echocardiography four-chamber cardiac tangential plane |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11813047B2 (en) | Automatic quantification of cardiac MRI for hypertrophic cardiomyopathy | |
CN107730497B (en) | Intravascular plaque attribute analysis method based on deep migration learning | |
CN110197713B (en) | Medical image processing method, device, equipment and medium | |
CN115035028A (en) | Left ventricular ejection fraction automatic calculation method based on ultrasonic image | |
CN109192305B (en) | Heart function automatic analysis method based on deep circulation neural network | |
US11972571B2 (en) | Method for image segmentation, method for training image segmentation model | |
CN112597982B (en) | Image classification method, device, equipment and medium based on artificial intelligence | |
CN110729045A (en) | Tongue image segmentation method based on context-aware residual error network | |
CN111369528A (en) | Coronary artery angiography image stenosis region marking method based on deep convolutional network | |
US20220012875A1 (en) | Systems and Methods for Medical Image Diagnosis Using Machine Learning | |
CN112465772A (en) | Fundus color photograph image blood vessel evaluation method, device, computer equipment and medium | |
de Siqueira et al. | Artificial intelligence applied to support medical decisions for the automatic analysis of echocardiogram images: A systematic review | |
CN114419499A (en) | Cardiac ultrasound video-based ejection fraction data processing method and device | |
CN111508077B (en) | Method, device, equipment and medium for intelligently evaluating heart function by using double-triangle model | |
CN114898882B (en) | Method and system for ultrasound-based assessment of right heart function | |
CN116664592A (en) | Image-based arteriovenous blood vessel separation method and device, electronic equipment and medium | |
CN114010227B (en) | Right ventricle characteristic information identification method and device | |
Zamzmi et al. | Echo doppler flow classification and goodness assessment with convolutional neural networks | |
CN112914610B (en) | Contrast-enhanced echocardiography wall thickness automatic analysis system and method based on deep learning | |
CN115908418A (en) | Method, system, equipment and medium for determining central line of aorta CT image | |
CN113592802B (en) | Mitral valve annular displacement automatic detection system based on ultrasonic image | |
CN113160116B (en) | Method, system and equipment for automatically segmenting inner membrane and outer membrane of left ventricle | |
CN114998582A (en) | Coronary artery blood vessel segmentation method, device and storage medium | |
CN112102352A (en) | Coronary motion tracking method and device for DSA image sequence | |
Bosch et al. | Overview of automated quantitation techniques in 2D echocardiography |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |