CN114202504A - Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium - Google Patents
Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium Download PDFInfo
- Publication number
- CN114202504A CN114202504A CN202111119622.2A CN202111119622A CN114202504A CN 114202504 A CN114202504 A CN 114202504A CN 202111119622 A CN202111119622 A CN 202111119622A CN 114202504 A CN114202504 A CN 114202504A
- Authority
- CN
- China
- Prior art keywords
- blood vessel
- carotid artery
- neural network
- sample image
- convolutional neural
- 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
- A61B8/08—Detecting organic movements or changes, e.g. tumours, cysts, swellings
- A61B8/0891—Detecting organic movements or changes, e.g. tumours, cysts, swellings for diagnosis of blood vessels
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
- A61B8/48—Diagnostic techniques
- A61B8/488—Diagnostic techniques involving Doppler signals
-
- 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
- A61B8/5223—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data for extracting a diagnostic or physiological parameter from medical diagnostic data
-
- 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
-
- 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
- G06T2207/20076—Probabilistic image processing
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/30101—Blood vessel; Artery; Vein; Vascular
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Surgery (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Pathology (AREA)
- Animal Behavior & Ethology (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Geometry (AREA)
- Physiology (AREA)
- Vascular Medicine (AREA)
- Quality & Reliability (AREA)
- Ultra Sonic Daignosis Equipment (AREA)
Abstract
The invention provides a carotid artery ultrasonic automatic Doppler method, which comprises the following steps: acquiring a carotid artery ultrasonic image; randomly extracting a preset number of image blocks on the carotid artery ultrasonic image; inputting the preset number of image blocks into a convolutional neural network classification model to obtain a blood vessel angle in the carotid artery ultrasonic image; inputting the preset number of image blocks into a convolutional neural network regression model to obtain the position and width of the blood vessel in the carotid artery ultrasonic image; and automatically generating a sampling gate on the carotid artery ultrasonic image based on the obtained blood vessel angle, blood vessel position and blood vessel width. The invention can simplify the operation of a doctor during carotid artery ultrasonic Doppler examination and improve the working efficiency of the doctor.
Description
Technical Field
The invention belongs to the technical field of ultrasonic equipment, and particularly relates to an automatic carotid artery ultrasonic Doppler method.
Background
The incidence of cerebrovascular diseases is rising year by year, and the non-invasive detection method is adopted to discover the existence of extracranial arterial vascular diseases at an early stage and treat the diseases in time, so that the method is an effective means for preventing and reducing the incidence of cerebrovascular diseases and has important clinical significance. The above extracranial artery vessels mainly refer to carotid vessels. Carotid vessels detected by carotid ultrasound comprise common carotid artery, internal carotid artery and external carotid artery, and are one of effective means for diagnosing and evaluating carotid lesions. The carotid artery ultrasonic examination has the characteristics of simple and convenient operation, strong repeatability, economy, practicability, easy acceptance by patients and the like; therefore, it is often used for screening normal people and examining the disease condition of high risk group with cerebrovascular disease, including before, during and after operation and follow-up.
Carotid artery ultrasound examination includes comprehensive analysis of two-dimensional structures, doppler spectra, and the like. The two-dimensional structure analysis can observe the inner diameter of the blood vessel and is used for judging whether the lumen is expanded, narrowed, twisted, normally traveled, the thickness of the intima and whether plaque exists. Doppler spectrum analysis acquires a blood flow spectrum to observe blood flow velocity, and judges whether the blood flow velocity is increased or decreased. Blood flow velocity often corresponds to a two-dimensional structure, such as an increase in blood flow velocity often due to a narrowing of the carotid lumen. Doppler spectroscopy is also used to obtain peak systolic blood flow velocity, end diastolic blood flow velocity and internal carotid to common carotid flow velocity ratios.
At present, when a doctor uses ultrasonic equipment to carry out carotid artery ultrasonic Doppler examination, firstly, the position of a probe is adjusted to make a carotid artery ultrasonic long-axis image clearest; then adjusting the Doppler angle (namely the angle of the sampling gate), wherein the range is generally between plus or minus 30 degrees; the width of the parallelogram sampling gate is adjusted, and the width of the sampling gate is generally 1/3 of the width of the blood vessel; finally, a frame of carotid image and a frame of representative Doppler spectrum are recorded, and the peak systolic flow rate is measured. The physician performs the above steps to ensure accurate measurement of peak systolic flow. If the doctor does not perform the above steps, blood flow sampling is incorrect, peak flow rate is increased or decreased artificially, and errors occur when the systolic peak flow rate ratio is calculated to evaluate the luminal stenosis. In this case, if there is clinically significant carotid stenosis, it may lead to erroneous judgment.
However, when doctors work, the number of patients to be detected is large, and the workload is large; if the procedure of carotid ultrasound examination can be simplified by an automated algorithm, the working efficiency of the doctor can be directly improved.
Disclosure of Invention
The method aims to overcome the defects in the prior art, and provides the carotid artery ultrasonic automatic Doppler method so as to simplify the operation of a doctor during carotid artery ultrasonic Doppler examination and improve the working efficiency of the doctor; the application also provides ultrasonic equipment and a storage medium corresponding to the method. In order to achieve the technical purpose, the embodiment of the invention adopts the technical scheme that:
in a first aspect, an embodiment of the present invention provides a carotid artery ultrasound automatic doppler method, where the method includes:
acquiring a carotid artery ultrasonic image;
randomly extracting a preset number of image blocks on the carotid artery ultrasonic image; such that at least one image patch contains a segment of a carotid vessel;
inputting the preset number of image blocks into a convolutional neural network classification model to obtain a blood vessel angle in the carotid artery ultrasonic image;
inputting the preset number of image blocks into a convolutional neural network regression model to obtain the position and width of the blood vessel in the carotid artery ultrasonic image;
and automatically generating a sampling gate on the carotid artery ultrasonic image based on the obtained blood vessel angle, blood vessel position and blood vessel width.
Further, the randomly extracting a preset number of image blocks on the carotid artery ultrasound image includes:
extracting a preset number of image blocks on the carotid artery ultrasonic image according to a preset extraction standard, wherein each image block records position information on the carotid artery ultrasonic image; the preset extraction criteria include: the aspect ratio of the image block is one or more preset aspect ratios, and/or the extraction length is a preset length and the extraction width is a preset width.
Specifically, the inputting the preset number of image blocks into a convolutional neural network classification model to obtain a blood vessel angle in the carotid artery ultrasound image specifically includes:
outputting a first prediction output vector containing a plurality of blood vessel angle prediction probabilities for each image block through the convolutional neural network classification model;
determining image blocks with highest blood vessel angle prediction probability in all first prediction output vectors;
and taking the blood vessel angle corresponding to the category with the highest blood vessel angle prediction probability in the first prediction output vector of the image block with the highest blood vessel angle prediction probability as the blood vessel angle in the carotid artery ultrasonic image.
Specifically, the inputting the preset number of image blocks into a convolutional neural network regression model to obtain the position and width of the blood vessel in the carotid artery ultrasound image includes:
outputting a second prediction output vector containing the predicted blood vessel position and the predicted blood vessel width for each image block through the convolutional neural network regression model;
restoring the image block with the highest blood vessel angle prediction probability to the carotid artery ultrasonic image, and obtaining the blood vessel position and the blood vessel width in the carotid artery ultrasonic image by combining the predicted blood vessel position and the blood vessel width in the image block.
Specifically, the automatically generating a sampling gate on the carotid artery ultrasonic image based on the obtained blood vessel angle, blood vessel position and blood vessel width comprises:
setting the center of the sampling gate at the center point of the position of the blood vessel in the carotid artery ultrasonic image; the angle of the sampling gate is the obtained angle of the blood vessel; the width of the sampling gate is less than or equal to the width of the obtained blood vessel.
Specifically, the convolutional neural network classification model and the convolutional neural network regression model are obtained by training in the following way:
acquiring a sample carotid artery ultrasonic image, wherein the sample carotid artery ultrasonic image is marked with blood vessel angle information, a blood vessel position mark and blood vessel width information;
randomly extracting a preset number of sample image blocks from the sample carotid artery ultrasonic image, and recording the position information of each sample image block on the carotid artery ultrasonic image;
dividing the sample image blocks into positive sample image blocks and negative sample image blocks according to a preset rule; the blood vessel angle of the positive sample image block is the blood vessel angle marked by the sample carotid artery ultrasound image corresponding to the positive sample image block;
training the convolutional neural network classification model according to the positive sample image blocks and the negative sample image blocks;
and training the convolutional neural network regression model according to the positive sample image blocks and the negative sample image blocks.
Further, the dividing the sample image block into a positive sample image block and a negative sample image block according to a predetermined rule includes:
calculating the intersection and comparison of each sample image block and the corresponding blood vessel of the blood vessel position mark in the sample carotid artery ultrasonic image, wherein the intersection and comparison is the ratio of the intersection to the union;
and taking the sample image blocks with the intersection ratio larger than a set threshold value as positive sample image blocks, and otherwise, taking the sample image blocks as negative sample image blocks.
Specifically, the training the convolutional neural network classification model according to the positive sample image blocks and the negative sample image blocks includes:
processing a positive sample image block and a negative sample image block by a first layer group consisting of a plurality of convolution layers, a batch normalization layer and an activation function layer, connecting a global pooling layer, and finally connecting a full-connection layer, wherein the layers in the convolutional neural network classification model are connected through weight parameters, and outputting a first prediction output vector, the numerical value of each dimension in the first prediction output vector represents the prediction probability of the input sample image block corresponding to the dimension, and the prediction probability comprises the blood vessel angle prediction probability of the sample image block;
calculating a first loss function in the convolutional neural network classification model;
and calculating a partial derivative value of the first loss function relative to each weight parameter, and reversely transmitting the partial derivative value back to the convolutional neural network classification model for updating and iteration to obtain the trained convolutional neural network classification model.
Preferably, the first loss function in the convolutional neural network classification model is formulated as:
Loss＝α(1-y’)r(-logy’) (1)
wherein alpha and r are hyperparameters, 0<α<0.8，1<r<3; y' represents the maximum prediction probability in the first prediction output vector; alpha (1-y')rIs the proportion of the whole first loss function, and the (-log') is the error loss generated by one sample image block.
Specifically, the training the convolutional neural network regression model according to the positive sample image blocks and the negative sample image blocks includes:
processing the positive sample image block and the negative sample image block by a second layer group consisting of a plurality of convolution layers, a batch normalization layer and an activation function layer, connecting the positive sample image block and the negative sample image block with a global pooling layer, and finally connecting the positive sample image block and the negative sample image block with a full-connection layer to output a second prediction output vector, wherein the second prediction output vector comprises blood vessel position information and blood vessel width information predicted on the sample image block;
calculating a second loss function in the convolutional neural network regression model;
and reversely transmitting the function value of the second loss function back to the convolution neural network regression model for updating and iteration to obtain the trained convolution neural network regression model.
Preferably, the second loss function in the convolutional neural network regression model is formulated as:
wherein out represents the second prediction output vector of the convolutional neural network regression model, and truth represents the real blood vessel position and the blood vessel width.
In a second aspect, an embodiment of the present invention provides an ultrasound apparatus, including:
a memory storing a computer program;
a processor for executing the computer program, which computer program is executed for performing the steps of the carotid artery ultrasound automatic doppler method as described above.
In a third aspect, an embodiment of the present invention provides a storage medium having a computer program stored therein, where the computer program is configured to execute the steps of the carotid artery ultrasound automatic doppler method as described above.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
1) the operation of a doctor can be simplified during carotid artery ultrasonic Doppler examination through automatic processing, and the working efficiency of the doctor is obviously improved.
2) By means of the artificial intelligence method, the time for a doctor to carry out carotid artery ultrasonic Doppler examination is greatly saved.
Drawings
Fig. 1 is a schematic diagram of a carotid artery ultrasound long-axis image in an embodiment of the invention.
Fig. 2 is a schematic diagram of an extracted image block according to an embodiment of the present invention.
FIG. 3 is a schematic diagram of a generated sample gate in an embodiment of the invention.
FIG. 4 is a flow chart of a method in an embodiment of the invention.
Detailed Description
In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of description and simplicity of description, but do not indicate or imply that the device or element being referred to must have a particular orientation, be constructed and operated in a particular orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
In the description of the present invention, it should be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; the connection can be mechanical connection or electrical connection; the two elements may be directly connected or indirectly connected through an intermediate medium, or may be communicated with each other inside the two elements, or may be wirelessly connected or wired connected. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
In addition, the technical features involved in the different embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
In a first aspect, an embodiment of the present invention provides a carotid artery ultrasound automatic doppler method, including:
acquiring a carotid artery ultrasonic image; here, the carotid artery ultrasound image is preferably a carotid artery ultrasound long-axis image, and carotid artery blood vessels in the carotid artery ultrasound long-axis image are clearer;
randomly extracting a preset number of image blocks on the carotid artery ultrasonic image; such that at least one image patch contains a segment of a carotid vessel; the preset number is usually several tens to several thousands;
inputting the preset number of image blocks into a convolutional neural network classification model to obtain a blood vessel angle in the carotid artery ultrasonic image;
inputting the preset number of image blocks into a convolutional neural network regression model to obtain the position and width of the blood vessel in the carotid artery ultrasonic image;
automatically generating a sampling gate on the carotid artery ultrasonic image based on the obtained blood vessel angle, blood vessel position and blood vessel width; the sampling gate is positioned in the obtained blood vessel position;
according to the embodiment of the invention, the blood vessel angle in the carotid artery ultrasonic image is obtained through the convolutional neural network classification model, the blood vessel position and the blood vessel width in the carotid artery ultrasonic image are obtained through the convolutional neural network regression model, and the prediction accuracy can be improved through separate processing;
according to the embodiment of the invention, through the automatic process of automatic acquisition of the blood vessel information and automatic generation of the sampling gate, the operation of a doctor can be simplified during carotid artery ultrasonic Doppler examination, and the working efficiency of the doctor is obviously improved; by adopting an artificial intelligence method, the time for the doctor to carry out carotid artery ultrasonic Doppler examination is greatly saved;
preferably, the randomly extracting a preset number of image blocks on the carotid artery ultrasound image includes: extracting a preset number of image blocks on the carotid artery ultrasonic image according to a preset extraction standard, wherein each image block records position information on the carotid artery ultrasonic image; the preset extraction criteria include: the length-width ratio of the image block is one or more preset length-width ratios, and/or the extraction length is a preset length and the extraction width is a preset width; according to the preset extraction standard image blocks, more accurate results can be obtained when the blood vessel angle, the blood vessel position and the blood vessel width in the carotid artery ultrasonic image are obtained based on each image block;
specifically, the inputting the preset number of image blocks into a convolutional neural network classification model to obtain a blood vessel angle in the carotid artery ultrasound image specifically includes: outputting a first prediction output vector containing a plurality of blood vessel angle prediction probabilities for each image block through the convolutional neural network classification model; determining image blocks with highest blood vessel angle prediction probability in all first prediction output vectors; taking the blood vessel angle corresponding to the category with the highest blood vessel angle prediction probability in the first prediction output vector of the image block with the highest blood vessel angle prediction probability as the blood vessel angle in the carotid artery ultrasonic image;
specifically, the inputting the preset number of image blocks into a convolutional neural network regression model to obtain the position and width of the blood vessel in the carotid artery ultrasound image includes: outputting a second prediction output vector containing the predicted blood vessel position and the predicted blood vessel width for each image block through the convolutional neural network regression model; restoring the image block with the highest blood vessel angle prediction probability to the carotid artery ultrasonic image, and obtaining the blood vessel position and the blood vessel width in the carotid artery ultrasonic image by combining the predicted blood vessel position and the blood vessel width in the image block;
the processing processes of the convolutional neural network classification model and the convolutional neural network regression model have independent processing processes and mutually matched processing processes, so that a more accurate processing result can be obtained;
specifically, the position of the blood vessel in the obtained carotid artery ultrasound image can be represented by a blood vessel prediction rectangular box;
specifically, the automatically generating a sampling gate on the carotid artery ultrasonic image based on the obtained blood vessel angle, blood vessel position and blood vessel width comprises: setting the center of the sampling gate at the central point of the position of the blood vessel in the carotid artery ultrasonic image so that the sampling gate is positioned in the middle area of the blood vessel prediction rectangular frame; the angle of the sampling gate is the obtained angle of the blood vessel; the width of the sampling gate is smaller than or equal to the width of the obtained blood vessel, for example, the width of the sampling gate is 1/3-1/2 of the width of the blood vessel in the obtained carotid artery ultrasonic image, so that the blood flow can be sampled in an optimal area;
the automated procedure of the carotid artery ultrasound automatic doppler method is described below with reference to an example;
firstly, obtaining a carotid artery ultrasonic image, wherein fig. 1 shows an obtained carotid artery ultrasonic long-axis image, deep black carotid artery blood vessels (approximately in a transverse trend) can be seen in an upper middle area of the image, then, randomly extracting a preset number of image blocks on the carotid artery ultrasonic image according to a preset extraction standard and recording position information of each image block on the carotid artery ultrasonic image; the preset extraction criteria include: the aspect ratio of the image block is one or more preset aspect ratios, such as 2: 1. 3: 1. 4: 1. 5: 1, etc., and/or the extraction length is a preset length and the width is a preset width; FIG. 2 exemplarily shows three extracted image blocks; due to random extraction and a large number, at least one image block can be ensured to contain a certain section of the complete carotid artery blood vessel; then respectively inputting the image blocks of the preset number into a convolutional neural network classification model and a convolutional neural network regression model; outputting, for each image block, by a convolutional neural network classification model, a first prediction output vector comprising negative sample image block prediction probabilities and a plurality of vessel angle prediction probabilities for the image block, e.g., [0.02,0.8,0.1,.. a.., 0,0.01], the first prediction output vector including 14 classes, the negative sample image block as the 0 th class, 13 vessel angles of-30 °, -25 °, -20 °, -15 °, -10 °, -5 °,0 °,5 °,10 °,15 °,20 °, 25 °, 30 °, as the remaining 13 classes; in the first prediction output vector, the prediction probability of a negative sample image block is 0.02, the prediction probability of a-30-degree blood vessel angle is 0.8, the prediction probability of a-25-degree blood vessel angle is 0.1.... 25-degree blood vessel angle is 0, the prediction probability of a 30-degree blood vessel angle is 0.01, and the sum of all prediction probabilities is 1; assuming that the maximum prediction probability in other first prediction output vectors is less than 0.8, the image block of which the first prediction output vector is [0.02,0.8, 0.1.... 0,0.01] can be determined; then, the-30-degree blood vessel angle corresponding to the category (2 nd category) with the blood vessel angle prediction probability of 0.8 in the first prediction output vector of the image block is used as the blood vessel angle in the obtained carotid artery ultrasonic image; outputting a second prediction output vector containing the predicted blood vessel position and the predicted blood vessel width for each image block through the convolutional neural network regression model; after the image blocks are determined, restoring the image block with the highest blood vessel angle prediction probability to the carotid artery ultrasonic image, and obtaining the blood vessel position in the carotid artery ultrasonic image by combining the predicted blood vessel position in the image block, wherein the predicted blood vessel width in the image block is used as the blood vessel width in the obtained carotid artery ultrasonic image; in this embodiment, the predicted blood vessel position in the image block is represented by a blood vessel prediction rectangular box, and then the blood vessel position in the obtained carotid artery ultrasound image is also represented by a blood vessel prediction rectangular box; two rectangular boxes are shown in FIG. 3, the predicted vessel position in the image block having been transformed into the carotid ultrasound image, where the larger one represents the vessel prediction rectangular box and the smaller one represents the sample gate; a second prediction output vector output by the convolutional neural network regression model adopts a 5-dimensional vector to represent the predicted blood vessel position and the predicted blood vessel width in the image block; the first 4 dimensions in the second prediction output vector represent the position information of the blood vessel prediction rectangular frame, and the 5 th dimension represents the predicted blood vessel width information; the second prediction output vector can be represented in two ways, for example, in a second prediction output vector [100, 200,20], a first value and a second value represent the coordinates of the upper left corner of the blood vessel prediction rectangular frame, a third value and a fourth value represent the coordinates of the lower right corner of the blood vessel prediction rectangular frame, and a fifth value represents the predicted blood vessel width of 20 pixels; for another example, in another second prediction output vector [150, 50,50,20], a first value and a second value represent coordinates of the center of the blood vessel prediction rectangular frame, a third value and a fourth value represent the length and width of the blood vessel prediction rectangular frame, and a fifth value represents that the predicted blood vessel width is 20 pixels;
finally, automatically generating a sampling gate on the carotid artery ultrasonic image based on the obtained blood vessel angle, blood vessel position and blood vessel width; the sampling gate position is set within the obtained blood vessel position (blood vessel prediction rectangular box), as shown in fig. 3 where the smaller rectangular box (sampling gate) is located within the larger rectangular box (blood vessel prediction rectangular box); certainly, the blood vessel prediction rectangular frame does not necessarily need to be displayed, and can also be hidden, and only four position values need to be represented; the angle of the sampling gate is the obtained blood vessel angle, and the width of the sampling gate is smaller than or equal to the predicted blood vessel width; preferably, the central point of the sampling gate is set as the central point of the position of the blood vessel in the obtained carotid artery ultrasonic image, so that the sampling gate is positioned in the middle area of the blood vessel prediction rectangular frame, and the blood flow sampling is more accurate; the width of the sampling gate is preferably 1/3 of the width of the blood vessel in the obtained carotid artery ultrasound long-axis image, so that the blood flow can be sampled in an optimal area;
how to train to obtain a convolutional neural network classification model and a convolutional neural network regression model is described below; the convolutional neural network classification model and the convolutional neural network regression model are obtained by training in the following mode:
acquiring a sample carotid artery ultrasonic image, and collecting the sample carotid artery ultrasonic image which is as comprehensive as possible, such as sample carotid artery ultrasonic images of different image depths and different collection objects; the sample carotid artery ultrasonic image is marked with blood vessel angle information, a blood vessel position mark and blood vessel width information; specifically, the collected sample carotid artery ultrasonic image can be marked by an image marking person, and blood vessel angle information is marked on the sample carotid artery ultrasonic image, for example, the sample carotid artery ultrasonic image is totally classified into 13, namely-30 degrees, -25 degrees, -20 degrees, -15 degrees, -10 degrees, -5 degrees, -0 degrees, 5 degrees, 10 degrees, 15 degrees, 20 degrees, 25 degrees and 30 degrees; marking a blood vessel position mark on the sample carotid artery ultrasonic image, namely marking a section of complete carotid artery blood vessel in the sample carotid artery ultrasonic image by using a blood vessel marking rectangular frame; calculating the distance between two corresponding points on two long sides of the blood vessel marking rectangular frame and marking the distance as the width of the blood vessel;
then randomly extracting a preset number of sample image blocks from the sample carotid artery ultrasound image, and recording the position information of each sample image block on the carotid artery ultrasound image to obtain a sample image block corresponding to each sample carotid artery ultrasound image; preferably, a preset number of image blocks may be extracted from the sample carotid ultrasound image according to a preset extraction standard, where the preset extraction standard includes: the aspect ratio of the sample image block is one or more preset aspect ratios, such as 2: 1. 3: 1. 4: 1. 5: 1, etc., and/or the extraction length is a preset length and the width is a preset width; the length and the width of the extracted sample image block are respectively required to be smaller than the length and the width of the original sample carotid artery ultrasonic image;
dividing the sample image block into a positive sample image block and a negative sample image block according to a preset rule; the blood vessel angle of the positive sample image block is the blood vessel angle marked by the sample carotid artery ultrasound image corresponding to the positive sample image block; the dividing the sample image block into a positive sample image block and a negative sample image block according to a predetermined rule specifically includes: calculating the intersection ratio of each sample image block and the corresponding blood vessel position mark in the sample carotid artery ultrasonic image, and taking the sample image block with the intersection ratio larger than a set threshold value of 0.8 as a positive sample image block, or taking the sample image block as a negative sample image block; the intersection ratio is the ratio of the intersection and the union of the blood vessels; the number of the positive sample image blocks is far less than that of the negative sample image blocks, which is not beneficial to the training of the convolutional neural network classification model and/or the convolutional neural network regression model, so that the negative sample image blocks with the set number are randomly reserved before the convolutional neural network classification model and/or the convolutional neural network regression model are trained, and the rest of the negative sample image blocks are discarded; for example, the number of positive sample image blocks is set to be N1, and 2N 1-3N 1 negative sample image blocks are randomly reserved;
then training the convolutional neural network classification model according to the positive sample image blocks and the negative sample image blocks; training the convolutional neural network regression model according to the positive sample image blocks and the negative sample image blocks;
(1) the specific process of training the convolutional neural network classification model according to the positive sample image blocks and the negative sample image blocks is as follows:
inputting the positive sample image blocks and the negative sample image blocks into a convolutional neural network classification model; setting a corresponding sample class vector for each sample image block in a convolutional neural network classification model, wherein the sample class vector comprises a negative sample class and a plurality of predefined blood vessel angle classes of the sample image block; for example, the 0 th class in the sample class vector is the negative sample class, and the remaining 13 classes are the-30 °, -25 °, -20 °, -15 °, -10 °, -5 °,0 °,5 °,10 °,15 °,20 °, 25 °, 30 vessel angle classes; the sample class vector [0,0,0,0,0,0,1,0,0,0,0,0,0,0] indicates that this is a positive sample image block of-5 ° blood vessel angle, and the sample class vector [1,0,0,0,0,0,0,0,0,0,0, 0] indicates that this is a negative sample image block;
fixing a positive sample image block and a negative sample image block to a uniform pixel size, processing the positive sample image block and the negative sample image block by a first layer group consisting of a plurality of convolution layers, a batch normalization layer and an activation function layer, connecting a global pooling layer, and finally connecting a full-connection layer, wherein the layers in the convolutional neural network classification model are connected through weight parameters, and outputting a first prediction output vector, the value of each dimension in the first prediction output vector represents the prediction probability of the input sample image block corresponding to the dimension, and the prediction probability comprises the blood vessel angle prediction probability of the sample image block; for example, the numerical value of the first dimension in the first prediction output vector represents the prediction probability of the negative sample class corresponding to the sample image block, and the numerical value of the second dimension in the first prediction output vector represents the prediction probability of the-30-degree blood vessel angle corresponding to the sample image block; when the step length of the convolution layer in the first layer group is selected to be 1, the convolution layer needs to be matched with the largest pooling layer for use, namely the first layer group is structurally composed of the convolution layer, the largest pooling layer, a batch normalization layer and an activation function layer; in the training process, the learning rate of 0.001 is adopted, and the saturation, the hue and the brightness of the input image block samples are randomly changed;
after the prediction probability is obtained, calculating a first loss function in the convolutional neural network classification model, wherein the formula is as follows:
Loss＝α(1-y’)r(-logy’) (1)
wherein, alpha and r are hyper-parameters and take values of 0.25 and 2; y' represents the maximum prediction probability in the first prediction output vector; alpha (1-y')rIs the weight of the entire first loss function, if y' is larger, it takes less weight in the first loss function; on the contrary, if the prediction probability is smaller, the specific gravity of the sample image block in the first loss function is larger, so that the influence of a large number of simple samples on the neural network can be effectively inhibited, and the convolutional neural network classification model is focused on samples with higher resolution difficulty; the (-logy ') is the error loss generated by a sample image block, the value range of y' is 0-1, when the prediction probability is smaller, the generated error loss is larger, namely, the convolutional neural network classification model can finally judge the blood vessel angle of the sample image block with higher probability;
finally, calculating a partial derivative value of the first loss function relative to each weight parameter, and reversely transmitting the partial derivative value back to the convolutional neural network classification model for updating and iteration to obtain a trained convolutional neural network classification model;
(2) the specific process of training the convolutional neural network regression model according to the positive sample image blocks and the negative sample image blocks is as follows:
inputting the positive sample image blocks and the negative sample image blocks into a convolutional neural network regression model;
fixing the positive sample image block and the negative sample image block to a uniform pixel size, processing by a second layer group consisting of a plurality of convolution layers, a batch normalization layer and an activation function layer, connecting with a global pooling layer, and finally connecting with a full-connection layer to output a second prediction output vector, wherein the second prediction output vector comprises blood vessel position information and blood vessel width information predicted by the sample image block, for example, a 5-dimensional second prediction output vector, wherein the first 4-dimensional second prediction output vector represents the position information of a blood vessel prediction rectangular frame, and the 5 th dimension represents the blood vessel width information; the bottom layer convolution layer can extract the edge characteristics of the blood vessel, the middle layer convolution layer can extract the texture characteristics of the blood vessel, and the high layer convolution layer can extract the global shape characteristics of the blood vessel;
calculating a second loss function in the convolutional neural network regression model, wherein the formula is as follows:
wherein out represents a second prediction output vector of the convolutional neural network regression model, and truth represents a real blood vessel position and a blood vessel width (that is, the position information of a blood vessel marking rectangular frame marked in the carotid artery ultrasound long-axis image sample and the blood vessel width), and the second loss function calculates the square error of each dimension in the second prediction output vector to guide the neural network training;
and reversely transmitting the function value of the second loss function back to the convolution neural network regression model for updating and iteration to obtain the trained convolution neural network regression model.
The embodiment of the invention adopts the second neural network model, namely the convolutional neural network regression model, and adopts the other loss function, because the numerical range of the position and the width of the blood vessel is far larger than each prediction probability value output by the convolutional neural network classification model, if the numerical range of the position and the width of the blood vessel is mixed into one neural network model to realize, the loss function is caused to be dominated by the error of the position and the width of the blood vessel, and the accurate prediction of the angle of the blood vessel is not easy to carry out.
In a second aspect, embodiments of the present invention also provide an ultrasound apparatus, including: a processor and a memory; the processor and the memory are communicated with each other, and the memory is stored with a computer program; the processor is configured to run the computer program, which when run performs the steps of the method as described above; the Processor may be a CPU, or other general-purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other Programmable logic device; the memory may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, a hard disk, or a solid state disk; the memory may also comprise a combination of memories of the kind described above.
In a third aspect, embodiments of the present invention also propose a storage medium having a computer program stored therein, the computer program being configured to perform the steps of the method as described hereinbefore when executed. The storage medium may be a magnetic Disk, an optical Disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a Flash Memory (Flash Memory), a Hard Disk (Hard Disk Drive, abbreviated as HDD) or a Solid State Drive (SSD), etc.; the storage medium may also comprise a combination of memories of the kind described above.
Finally, it should be noted that the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to examples, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.
Claims (11)
1. A carotid artery ultrasound automatic Doppler method, characterized in that the method comprises:
acquiring a carotid artery ultrasonic image;
randomly extracting a preset number of image blocks on the carotid artery ultrasonic image;
inputting the preset number of image blocks into a convolutional neural network classification model to obtain a blood vessel angle in the carotid artery ultrasonic image;
inputting the preset number of image blocks into a convolutional neural network regression model to obtain the position and width of the blood vessel in the carotid artery ultrasonic image;
and automatically generating a sampling gate on the carotid artery ultrasonic image based on the obtained blood vessel angle, blood vessel position and blood vessel width.
2. The carotid artery ultrasound automatic doppler method as claimed in claim 1, wherein said randomly extracting a preset number of image patches on the carotid artery ultrasound image comprises:
extracting a preset number of image blocks on the carotid artery ultrasonic image according to a preset extraction standard, wherein each image block records position information on the carotid artery ultrasonic image; the preset extraction criteria include: the aspect ratio of the image block is one or more preset aspect ratios, and/or the extraction length is a preset length and the extraction width is a preset width.
3. The carotid artery ultrasound automatic doppler method according to claim 1 or 2, wherein the inputting the image blocks of the preset number into a convolutional neural network classification model to obtain the blood vessel angle in the carotid artery ultrasound image specifically comprises:
outputting a first prediction output vector containing a plurality of blood vessel angle prediction probabilities for each image block through the convolutional neural network classification model;
determining image blocks with highest blood vessel angle prediction probability in all first prediction output vectors;
and taking the blood vessel angle corresponding to the category with the highest blood vessel angle prediction probability in the first prediction output vector of the image block with the highest blood vessel angle prediction probability as the blood vessel angle in the carotid artery ultrasonic image.
4. The carotid artery ultrasound automatic doppler method as claimed in claim 3, wherein said inputting the preset number of image blocks into a convolutional neural network regression model to obtain the position and width of the blood vessel in the carotid artery ultrasound image comprises:
outputting a second prediction output vector containing the predicted blood vessel position and the predicted blood vessel width for each image block through the convolutional neural network regression model;
restoring the image block with the highest blood vessel angle prediction probability to the carotid artery ultrasonic image, and obtaining the blood vessel position and the blood vessel width in the carotid artery ultrasonic image by combining the predicted blood vessel position and the blood vessel width in the image block.
5. The carotid artery ultrasound automatic doppler method of claim 1, wherein said automatically generating a sample gate on the carotid artery ultrasound image based on the obtained vessel angle, vessel position and vessel width comprises:
setting the center of the sampling gate at the center point of the position of the blood vessel in the carotid artery ultrasonic image; the angle of the sampling gate is the obtained angle of the blood vessel; the width of the sampling gate is less than or equal to the width of the obtained blood vessel.
6. The carotid artery ultrasound automatic Doppler method according to claim 1 or 2,
the convolutional neural network classification model and the convolutional neural network regression model are obtained by training in the following mode:
acquiring a sample carotid artery ultrasonic image, wherein the sample carotid artery ultrasonic image is marked with blood vessel angle information, a blood vessel position mark and blood vessel width information;
randomly extracting a preset number of sample image blocks from the sample carotid artery ultrasonic image, and recording the position information of each sample image block on the carotid artery ultrasonic image;
dividing the sample image blocks into positive sample image blocks and negative sample image blocks according to a preset rule; the blood vessel angle of the positive sample image block is the blood vessel angle marked by the sample carotid artery ultrasound image corresponding to the positive sample image block;
training the convolutional neural network classification model according to the positive sample image blocks and the negative sample image blocks;
and training the convolutional neural network regression model according to the positive sample image blocks and the negative sample image blocks.
7. The carotid artery ultrasound automatic doppler method of claim 6, wherein said dividing said sample image blocks into positive sample image blocks and negative sample image blocks according to a predetermined rule comprises:
calculating the intersection and comparison of each sample image block and the corresponding blood vessel of the blood vessel position mark in the sample carotid artery ultrasonic image, wherein the intersection and comparison is the ratio of the intersection to the union;
and taking the sample image blocks with the intersection ratio larger than a set threshold value as positive sample image blocks, and otherwise, taking the sample image blocks as negative sample image blocks.
8. The carotid artery ultrasound automatic doppler method of claim 6, wherein said training of said convolutional neural network classification model from said positive sample patch and said negative sample patch comprises:
processing a positive sample image block and a negative sample image block by a first layer group consisting of a plurality of convolution layers, a batch normalization layer and an activation function layer, connecting a global pooling layer, and finally connecting a full-connection layer, wherein the layers in the convolutional neural network classification model are connected through weight parameters, and outputting a first prediction output vector, the numerical value of each dimension in the first prediction output vector represents the prediction probability of the input sample image block corresponding to the dimension, and the prediction probability comprises the blood vessel angle prediction probability of the sample image block;
calculating a first loss function in the convolutional neural network classification model;
and calculating a partial derivative value of the first loss function relative to each weight parameter, and reversely transmitting the partial derivative value back to the convolutional neural network classification model for updating and iteration to obtain the trained convolutional neural network classification model.
9. The carotid artery ultrasound automatic Doppler method of claim 6,
the training of the convolutional neural network regression model according to the positive sample image blocks and the negative sample image blocks comprises:
processing the positive sample image block and the negative sample image block by a second layer group consisting of a plurality of convolution layers, a batch normalization layer and an activation function layer, connecting the positive sample image block and the negative sample image block with a global pooling layer, and finally connecting the positive sample image block and the negative sample image block with a full-connection layer to output a second prediction output vector, wherein the second prediction output vector comprises a blood vessel position and a blood vessel width predicted on the sample image block;
calculating a second loss function in the convolutional neural network regression model;
and reversely transmitting the function value of the second loss function back to the convolution neural network regression model for updating and iteration to obtain the trained convolution neural network regression model.
10. An ultrasound device, comprising:
a memory storing a computer program;
a processor for running the computer program which when run performs the steps of the carotid ultrasound automatic doppler method according to any of claims 1 to 9.
11. A storage medium characterized in that,
the storage medium has stored therein a computer program configured to perform the steps of the carotid artery ultrasound automatic doppler method according to any of claims 1-9 when executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111119622.2A CN114202504A (en) | 2021-09-24 | 2021-09-24 | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111119622.2A CN114202504A (en) | 2021-09-24 | 2021-09-24 | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114202504A true CN114202504A (en) | 2022-03-18 |
Family
ID=80646119
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111119622.2A Pending CN114202504A (en) | 2021-09-24 | 2021-09-24 | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114202504A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113838028A (en) * | 2021-09-24 | 2021-12-24 | 无锡祥生医疗科技股份有限公司 | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium |
| CN116934768A (en) * | 2023-08-16 | 2023-10-24 | 中国人民解放军总医院 | Method and system for improving blood vessel segmentation accuracy in CTA image mode |
-
2021
- 2021-09-24 CN CN202111119622.2A patent/CN114202504A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113838028A (en) * | 2021-09-24 | 2021-12-24 | 无锡祥生医疗科技股份有限公司 | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium |
| CN113838028B (en) * | 2021-09-24 | 2024-08-06 | 无锡祥生医疗科技股份有限公司 | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium |
| CN116934768A (en) * | 2023-08-16 | 2023-10-24 | 中国人民解放军总医院 | Method and system for improving blood vessel segmentation accuracy in CTA image mode |
| CN116934768B (en) * | 2023-08-16 | 2024-05-10 | 中国人民解放军总医院 | Method and system for improving blood vessel segmentation accuracy in CTA image mode |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7407790B2 (en) | Ultrasound system with artificial neural network for guided liver imaging | |
| US20170367685A1 (en) | Method for processing 3d image data and 3d ultrasonic imaging method and system | |
| US11207055B2 (en) | Ultrasound Cardiac Doppler study automation | |
| CN112529894B (en) | Thyroid nodule diagnosis method based on deep learning network | |
| CN110678933B (en) | Ultrasound clinical feature detection and associated devices, systems, and methods | |
| CN111768366A (en) | Ultrasonic imaging system, BI-RADS classification method and model training method | |
| Nurmaini et al. | Accurate detection of septal defects with fetal ultrasonography images using deep learning-based multiclass instance segmentation | |
| JP2016531709A (en) | Image analysis technology for diagnosing disease | |
| CN114202504A (en) | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium | |
| CN104143047B (en) | The automatic tissue scaling method of intravascular ultrasound gray scale image | |
| CN112381164B (en) | Ultrasound image classification method and device based on multi-branch attention mechanism | |
| JP2013542046A (en) | Ultrasound image processing system and method | |
| BR112020014733A2 (en) | COMPUTER IMPLEMENTED METHOD FOR OBTAINING ANATOMICAL MEASUREMENTS ON AN ULTRASOUND IMAGE, COMPUTER PROGRAM MEDIA, IMAGE ANALYSIS DEVICE AND ULTRASOUND IMAGING METHOD | |
| CN112638273A (en) | Biometric measurement and quality assessment | |
| CN117338340A (en) | Ultrasonic imaging method and equipment | |
| CN111275706A (en) | Shear wave elastic imaging-based ultrasound omics depth analysis method and system | |
| CN117197594A (en) | Deep neural network-based heart shunt classification system | |
| Wulaning Ayu et al. | Pixel Classification Based on Local Gray Level Rectangle Window Sampling for Amniotic Fluid Segmentation. | |
| CN115760851B (en) | Ultrasonic image data processing equipment, system and computer readable storage medium based on machine learning | |
| US20230329674A1 (en) | Ultrasound imaging | |
| CN111265234A (en) | Method and system for judging properties of lung mediastinal lymph nodes | |
| US20220249060A1 (en) | Method for processing 3d image data and 3d ultrasonic imaging method and system | |
| Zhang et al. | Improving the quality of fetal heart ultrasound imaging with multihead enhanced self-attention and contrastive learning | |
| CN113838028B (en) | Carotid artery ultrasonic automatic Doppler method, ultrasonic equipment and storage medium | |
| Zhang et al. | Advances in the Application of Artificial Intelligence in Fetal 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 |