TWI792055B - Establishing method of echocardiography judging model with 3d deep learning, echocardiography judging system with 3d deep learning and method thereof - Google Patents

Abstract

The present disclosure relates to an echocardiography judging system with 3D deep learning including an ultrasonic detector and a processor. The processor stores a program including a reference data obtaining module, a reference image pre-processing module, a reference feature selecting module, a model training module and an analyzing module. Therefore, the echocardiography judging system with 3D deep learning can be used to analyze a target echocardiography image through an echocardiography judging model to generate a target echocardiography judgment result corresponding to an infarct related artery.

Description

A method for establishing a cardiac ultrasound judgment model using three-dimensional deep learning, a cardiac ultrasound judgment system and its method using three-dimensional deep learning

The present invention relates to a method, system and method for establishing a medical information judgment model, in particular to a method for establishing a cardiac ultrasound judgment model using three-dimensional deep learning, a cardiac ultrasound judgment system and its method using three-dimensional deep learning.

Among cardiovascular diseases, acute coronary syndrome (ACS) is the most deadly and critical. In recent years, with the popularity of cardiac catheterization, the survival rate of patients has increased significantly. Early understanding of infarct-related arteries (Infarct Related Artery, IRA) is extremely helpful for preoperative preparation of cardiac catheterization and evaluation of strategies to open blood vessels.

However, at present, only patients with ST-elevating myocardial infarction (STEMI) can judge infarction-related vessels through ECG in the acute phase, while patients with non-ST-elevating myocardial infarction (NSTEMI) and unstable angina (UA) can Cardiac catheterization is often required to assess reperfusion planning.

Cardiac ultrasound is widely used clinically, but cardiac ultrasound is often unable to judge regional wall motion abnormality (RWMA), even experienced operators may not be able to correctly judge whether there is an infarction based on the patient's cardiac ultrasound associated blood vessels. Although there is a bright spot at present, Speckle Tracking Echocardiography (STE) can detect infarction-related blood vessels that are difficult to identify with the naked eye at an early stage, but it is not popular because of its high price and time-consuming.

It can be seen from this that if the blood vessels related to infarction can be judged quickly by using cardiac ultrasound in the first time, it is indeed the eager expectation of the public, and it is also the goal and direction that the relevant industry must work hard to develop breakthroughs.

Therefore, for non-ST ascending myocardial infarction and unstable angina, the purpose of the present invention is to provide a cardiac ultrasound judgment system using three-dimensional deep learning, which can compare the cardiac ultrasound judgment results with the cardiac catheter diagnosis results to correct Cardiac ultrasound judgment model to improve the accuracy of cardiac ultrasound judgment model for judging blood vessels related to infarction.

An embodiment according to an aspect of the present invention provides a method for establishing a cardiac ultrasound judgment model using 3D deep learning, which includes the following steps: data acquisition step, image preprocessing step, reference feature selection step, and model training step. The data obtaining step is obtaining reference data, wherein the reference data includes a plurality of reference echocardiographic data and a plurality of diagnostic results of cardiac catheterization corresponding to the reference echocardiographic data. The image preprocessing step is to cut each reference echocardiogram data to obtain a plurality of reference echocardiogram images. The reference feature selection step is to use the reference feature selection module to analyze each reference cardiac ultrasound image, and obtain reference image feature values in each reference cardiac ultrasound image, so as to obtain a plurality of reference image features corresponding to these reference cardiac ultrasound images value. The model training step is to generate a cardiac ultrasound judgment model based on the reference image feature value training, and input the reference cardiac ultrasound data into the cardiac ultrasound judgment model to generate a plurality of cardiac ultrasound corresponding to the reference cardiac ultrasound data Judgment results, and then compare these heart ultrasound judgment results with the cardiac catheter diagnosis results to modify the heart ultrasound judgment model.

In this way, the present invention trains the cardiac ultrasound judgment model through the cardiac ultrasound data, and corrects the cardiac ultrasound judgment model according to the diagnosis result of the cardiac catheterization, and then obtains the optimized cardiac ultrasound judgment model.

Other examples of the aforementioned embodiments are as follows: the aforementioned cardiac ultrasound judgment model includes a sequentially connected dense convolutional neural network (DenseNet) model, a fully-connected neural network (Fully-connected neural network, FCNN) model and a normalized Exponential function (Softmax) model.

Other examples of the foregoing embodiments are as follows: the aforementioned dense convolutional neural network model includes a plurality of dense convolutional neural network modules (DenseNet Block, DNB), a plurality of maximum pooling modules (Max pooling) and flat modules ( Flatten).

Other examples of the foregoing embodiments are as follows: each of the aforementioned dense convolutional neural network modules includes a plurality of dense convolutional neural network units, and these dense convolutional neural network units are cascaded to each other in sequence, and each dense convolutional neural network The network unit includes a three-dimensional convolutional layer (Conv), a linear rectification unit training layer (ReLU) and a batch normalization layer (Batch Normalization, BN).

An embodiment according to another aspect of the present invention provides a cardiac ultrasound diagnosis system using three-dimensional deep learning, which includes an ultrasound detector and a processor. The ultrasonic detector is used to generate the target cardiac ultrasonic image of the subject. The processor is electrically connected to the ultrasound detector, and the processor stores a program. When the program is executed by the processor, the program judges whether the subject has an Infarct Related Artery (IRA) according to the target cardiac ultrasound image. The program includes a reference data acquisition module, a reference image preprocessing module, a reference feature selection module, a model training module and an analysis module. The reference data obtaining module is used to obtain reference data, and the reference data includes a plurality of reference echocardiographic data and a plurality of diagnostic results of cardiac catheterization corresponding to the reference echocardiographic data. The reference image preprocessing module is used for cutting each reference echocardiogram data to obtain a plurality of reference echocardiogram images. The reference feature selection module is used to obtain reference image feature values in each reference echocardiogram image after analyzing each reference echocardiogram image, so as to obtain a plurality of reference image feature values corresponding to the reference echocardiogram images. The model training module is based on the training of these reference image feature values to generate a cardiac ultrasound judgment model, and input these reference cardiac ultrasound data into the cardiac ultrasound judgment model to generate a plurality of hearts corresponding to these reference cardiac ultrasound data Ultrasonic judgment results, and then compare the cardiac ultrasound judgment results with the cardiac catheter diagnosis results to modify the cardiac ultrasound judgment model. The analysis module analyzes the target cardiac ultrasound image according to the cardiac ultrasound judgment model to generate the target cardiac ultrasound judgment result corresponding to the infarction-related blood vessels.

In this way, the present invention trains a cardiac ultrasound judgment model through cardiac ultrasound data, and then analyzes target cardiac ultrasound images of the subject to accurately determine whether the subject has infarct-related blood vessels.

Other examples of the aforementioned embodiments are as follows: the aforementioned cardiac ultrasound judgment model includes a sequentially connected dense convolutional neural network model, a fully connected neural network model, and a normalized exponential function model.

Other examples of the foregoing embodiments are as follows: the foregoing dense convolutional neural network model includes a plurality of dense convolutional neural network modules, a plurality of maximum pooling modules and a flattening module.

An embodiment according to yet another aspect of the present invention provides a method for judging cardiac ultrasound using 3D deep learning, which includes the following steps. Provide the cardiac ultrasound judgment model mentioned in the preceding paragraph. Provide the subject's target cardiac ultrasound image. The target cardiac ultrasound image is analyzed according to the cardiac ultrasound judgment model to generate a target cardiac ultrasound judgment result. Whether the subject has an infarct-related blood vessel is judged by using the judgment result of the target cardiac ultrasound.

Therefore, the present invention analyzes the subject's target cardiac ultrasound image through the cardiac ultrasound judgment model, and then can accurately determine whether the subject has infarct-related blood vessels.

Other examples of the aforementioned embodiments are as follows: the aforementioned cardiac ultrasound judgment model includes a sequentially connected dense convolutional neural network model, a fully connected neural network model, and a normalized exponential function model.

Other examples of the foregoing embodiments are as follows: the foregoing dense convolutional neural network model includes a plurality of dense convolutional neural network modules, a plurality of maximum pooling modules and a flattening module.

Several embodiments of the present invention will be described below with reference to the drawings. For the sake of clarity, many practical details are included in the following narrative. It should be understood, however, that these practical details should not be used to limit the invention. That is, in some embodiments of the present invention, these practical details are unnecessary. In addition, for the sake of simplifying the drawings, some commonly used structures and elements will be shown in a simple and schematic way in the drawings; and repeated elements may be denoted by the same reference numerals.

FIG. 1 is a flow chart showing the steps of a method 100 for establishing a cardiac ultrasound judgment model using 3D deep learning according to an embodiment of the present invention. As shown in FIG. 1 , the method 100 for establishing a cardiac ultrasound judgment model using 3D deep learning includes step 110 , step 120 , step 130 and step 140 . The established cardiac ultrasound judgment model can be used to analyze the target cardiac ultrasound image, so as to facilitate the subsequent judgment of whether there is an infarct related artery (Infarct Related Artery, IRA) and generate the target cardiac ultrasound judgment result.

Step 110 is to perform a data acquisition step. The data acquisition step is to obtain reference data. The reference data includes a plurality of reference echocardiographic data and a plurality of diagnostic results of cardiac catheterization corresponding to the reference echocardiographic data.

Step 120 is an image pre-processing step. The image pre-processing step is to cut each reference echocardiogram data to obtain a plurality of reference echocardiogram images.

Step 130 is to perform a reference feature selection step. The reference feature selection step is to use the reference feature selection module to analyze each reference cardiac ultrasound image, and obtain reference image feature values from each reference cardiac ultrasound image, so as to obtain the corresponding reference cardiac ultrasound images. A plurality of reference image feature values of the ultrasound image.

Step 140 is a model training step. The model training step is to generate a cardiac ultrasound judgment model based on the reference image feature value training, and input the reference cardiac ultrasound data to the cardiac ultrasound judgment model to generate a corresponding reference heart. The multiple cardiac ultrasound judgment results of the ultrasound data, and then compare these cardiac ultrasound judgment results with the cardiac catheter diagnosis results to modify the cardiac ultrasound judgment model.

The cardiac ultrasound judgment model may include a dense convolutional neural network (DenseNet) model, a fully-connected neural network (Fully-connected neural network, FCNN) model and a normalized exponential function (Softmax )Model. Preferably, the dense convolutional neural network model may include a plurality of dense convolutional neural network modules (DenseNet Block, DNB), a plurality of maximum pooling modules (Max pooling) and a flat module (Flatten), and Each dense convolutional neural network module may contain a plurality of dense convolutional neural network units, and these dense convolutional neural network units are cascaded to each other in sequence, and each dense convolutional neural network unit contains a three-dimensional convolutional layer (Conv ), linear rectification unit training layer (ReLU) and batch normalization layer (Batch Normalization, BN).

FIG. 2 is a schematic block diagram of a cardiac ultrasound judgment system 200 using three-dimensional deep learning according to another embodiment of the present invention. As shown in FIG. 2 , the cardiac ultrasound diagnosis system 200 using 3D deep learning includes an ultrasound detector 210 and a processor 220 .

The ultrasound detector 210 can be a cardiac ultrasound detector, which is used to generate a target cardiac ultrasound image of the subject.

The processor 220 is electrically connected to the ultrasound detector 210. The processor 220 stores a program. When the program is executed by the processor 220, the program judges whether the subject has infarct-related blood vessels according to the target cardiac ultrasound image. The program includes a reference data acquisition module 221 , a reference image preprocessing module 222 , a reference feature selection module 223 , a model training module 224 and an analysis module 225 .

The reference data acquisition module 221 is used to obtain reference data, which includes a plurality of reference echocardiographic data and a plurality of cardiac catheterization results corresponding to the reference echocardiographic data.

The reference image pre-processing module 222 is used for cutting each reference echocardiogram data to obtain a plurality of reference echocardiogram images.

The reference feature selection module 223 is used to obtain reference image feature values in each reference echocardiogram image after analyzing each reference echocardiogram image, so as to obtain a plurality of reference image feature values corresponding to these reference echocardiogram images.

The model training module 224 is based on the training of these reference image feature values to generate a cardiac ultrasound judgment model, and input these reference cardiac ultrasound data into the cardiac ultrasound judgment model to generate a plurality of corresponding to these reference cardiac ultrasound data Cardiac ultrasound judgment results, and then these cardiac ultrasound judgment results are compared with these cardiac catheter diagnosis results to modify the cardiac ultrasound judgment model.

The analysis module 225 analyzes the target cardiac ultrasound image according to the cardiac ultrasound judgment model to generate a target cardiac ultrasound judgment result corresponding to the infarct-related blood vessel.

FIG. 3 is a flow chart showing the steps of a cardiac ultrasound judgment method 300 using 3D deep learning according to another embodiment of the present invention. The cardiac ultrasound judgment method 300 using 3D deep learning includes step 310 , step 320 , step 330 and step 340 .

Step 310 is to provide a cardiac ultrasound judgment model. The cardiac ultrasound judgment model is established through the aforementioned steps 110 to 140 .

Step 320 is to provide the target ultrasound image of the heart, that is, to provide the target ultrasound image of the heart of the subject.

Step 330 is to generate a target ultrasonic cardiac judgment result, that is, the analysis module 225 analyzes the target ultrasonic cardiac image according to the ultrasonic cardiac judgment model to generate a target ultrasonic cardiac judgment result.

Step 340 is to determine whether there is an infarction-related vessel, that is, to determine whether the subject has an infarction-related vessel by using the target cardiac ultrasound determination result.

[Test example]

1. Reference materials

The reference data used in the present invention include multiple reference echocardiographic data of multiple patients with acute coronary syndrome (ACS) and multiple cardiac catheterization results after diagnosis by doctors, and can be classified into the following three groups:

(1) ACS: Coronary artery obstruction was seen in cardiac catheterization, and regional wall motion abnormalities (RWMA) were also seen in cardiac ultrasound.

(2) Insignificant: Coronary artery obstruction was seen in cardiac catheterization, but regional wall motion abnormalities were not seen in cardiac ultrasound.

(3) Control: Cardiac catheterization did not see coronary artery obstruction, and cardiac ultrasound did not see regional wall motion abnormalities.

It is worth noting that a reference echocardiographic data and a cardiac catheterization result were defined as a group of samples. There are 9000 sample groups in the experiment, which are further classified into 7500 training sample groups, 1000 verification sample groups and 500 test sample groups.

2. Image pre-processing

After obtaining the reference data, the image pre-processing step is performed firstly, which is to cut each reference echocardiogram data to obtain a plurality of reference echocardiogram images.

Three, the cardiac ultrasonic judgment model of the present invention

FIG. 4 is a schematic diagram showing the structure of the cardiac ultrasound judgment model 400 of the present invention. As shown in FIG. 4 , the cardiac ultrasound judgment model 400 is mainly composed of a dense convolutional neural network model 500 , a fully connected neural network model 600 and a normalized exponential function model 700 connected in sequence. The dense convolutional neural network model 500 may include 4 dense convolutional neural network modules 510 , 4 max pooling modules 520 and 1 flattening module 530 . First, the reference cardiac ultrasound image 401 is processed by the dense convolutional neural network module 510 and the maximum pooling module 520 four times, and flattened into a one-dimensional array by the flattening module 530, and then through the fully connected neural network The model 600 synthesizes the eigenvalues through the weight matrix, and finally inputs them into the normalized exponential function model 700 for operation to generate three kinds of deep learning probabilities 800, wherein the deep learning probabilities 800 are respectively abnormal probability, normal probability and suspected abnormal probability. The deep learning probability 800 is classified into three kinds of cardiac ultrasound judgment results 900 through cross entropy (Cross Entropy), and the cardiac ultrasound judgment results 900 are respectively abnormal, normal and suspected abnormal. The resulting echocardiogram The judgment result 900 can be compared with the diagnosis result of cardiac catheterization to modify the cardiac ultrasound judgment model 400 . Among them, the fully connected neural network model 600 is a connection mode of neurons, and its characteristic is that the neurons of the upper layer are connected with all the neurons of the lower layer, and the normalized exponential function model 700 is mainly used in the processing of deep learning On multi-classification problems, and it converts the output value of multi-classification into relativistic probability, making it easier to understand and compare.

Please refer to FIG. 4 and FIG. 5 together, wherein FIG. 5 is a schematic diagram of the architecture of the dense convolutional neural network module 510 in FIG. 4 . The dense convolutional neural network module 510 includes 4 dense convolutional neural network units 511, 512, 513, 514, which are input to the input terminal 515 with reference to the cardiac ultrasound image 401, and through the 4 dense convolutional neural network units 511 , 512 , 513 , 514 are sequentially cascaded and calculated, and then output to the output terminal 516 .

In detail, each dense convolutional neural network unit 511 , 512 , 513 , 514 may include a three-dimensional convolutional layer 540 , a linear rectification unit training layer 550 and a batch normalization layer 560 . The output of each dense convolutional neural network unit 511 , 512 , 513 , 514 will be sent to the cascaded unit C for summing up, and used as the input of the next one, and so on, and finally output to the output terminal 516 . In particular, the padding method is used to enlarge the output image of the three-dimensional convolutional layer 540 while maintaining the image size (Size) of the entire dense convolutional neural network module 510, that is, the image passes through each dense convolutional neural network unit After 511, 512, 513, and 514 operations, the size of the graph remains unchanged, only the number of channels (Channel) changes.

In addition, the cardiac ultrasound judgment model 400 of the present invention uses a linear The rectification unit training layer 550 is used as an activation function (Activation Function). Compared with the traditional neural network activation function, the linear rectification unit training layer 550 has the principle of imitating biology, more efficient gradient descent and backpropagation, and avoids gradient disappearance and Simplify the calculation process and other advantages.

Furthermore, the main function of the loss function (Loss Function) used by the cardiac ultrasound judgment model 400 of the present invention is to provide a deep learning network to evaluate the degree of error between the training result and the target result, and it is shown in the formulas (1) and (2) ) as shown:

為網路參數，Y _r 為訓練結果，

為目標結果，Ω_Reg為正規化項(regularization term)，λ為正規化因子(regularization factor)。 For the calculation of the loss function, the gradient descent method is used to achieve the purpose of network self-adjustment and learning. in

is the network parameter, Y _r is the training result,

is the target result, Ω _Reg is the regularization term, and λ is the regularization factor.

Next, the network parameter w _{l +1} of the cardiac ultrasound judgment model 400 of the present invention is recursively updated by using the Mini Batch Gradient Descent Moment Estimation (Mini Batch Gradient Descent Moment Estimation) technique, and it is as in the formula ( 3) As shown:

是一個值很小的常數主要用於避免式子(3)中第二項出現分母為零的現象發生。在更新網路參數時本發明的心臟超音波判斷模 型400考慮了數量為B個小批次(Mini Batch)，因此在下列式子(4)與(5)中出現一次式梯度下降法(Gradient Descent)▽L ^r (w ^l )與二次式梯度下降法[▽L ^r (w ^l )]²的平均數，其平均數量為小批次的數量B。此外，β₁與β₂為衰減率(decay rate)。每次心臟超音波判斷模型400從測試樣本組中，隨機抽取數量為B個樣本組進行深度學習的方式稱為小批次。 Where η is the learning rate (Learning Rate),

is a constant with a very small value, which is mainly used to avoid the phenomenon that the denominator of the second term in the formula (3) is zero. Cardiac ultrasound judgment model 400 of the present invention considered quantity to be B mini-batches (Mini Batch) when updating network parameter, therefore appears in following formula (4) and (5) Gradient descending method (Gradient Descent)▽ L ^r ( w ^l ) and the average of the quadratic gradient descent method [▽ L ^r ( w ^l )] ² , the average number of which is the number B of the small batch. In addition, β ₁ and β ₂ are decay rates. The mode in which the cardiac ultrasound judgment model 400 randomly selects B sample groups from the test sample group for deep learning is called a small batch.

Equations (4) and (5) show that the cardiac ultrasound judgment model 400 not only uses the small batch technique, but also considers the gradient descent value of each group of small batches before adjusting the network parameters. The previous adjustment experience will be used as the reference basis for this adjustment, and it has the intelligent feature of dynamically adjusting network parameters based on previous experience and this small batch generation.

Please refer to FIG. 4 to FIG. 6 together, wherein FIG. 6 is a schematic diagram of the architecture of the batch normalization layer 560 in FIG. 5 . The cardiac ultrasound judgment model 400 of the present invention uses the batch normalization layer 560 to reduce the time required for training, prevent the problem of disappearing gradient descent and reduce the phenomenon of overfitting caused by parameter initialization. During the learning process, the cardiac ultrasound judgment model 400 randomly selects B sample groups (small batches) from the test sample group for learning. Please refer to the following formula (6) and formula (7) ^to calculate the output value A 1 and A 2 of each neuron from the input value X ¹ , X ² , X ³ and the weight W ¹ , W ² , ^{W 3} , the mean value μ and standard deviation σ of A ³ , and then normalize all the output values A ¹ , A ² , A ³ with the mean value μ and standard deviation σ to get the values à ¹ , à ² , à ³ (that is, the formula The value of (6) à ⁱ ), but we hope that when the small batch enters the activation function, the mean μ and standard deviation σ are different from the other small batch, so the parameter β and parameter γ are introduced to adjust this phenomenon and obtain the value  ¹ ,  ² ,  ³ (that is, the value  ⁱ of formula (7)).

 ⁱ =γ˙à ⁱ +β (7)。

 ⁱ = γ˙ à ⁱ + β (7).

4. Model training steps

7500 training sample groups and 1000 verification sample groups are put into the cardiac ultrasound judgment model 400 for training. According to the training results, if the goal of Loss less than 2% cannot be met, the deep learning parameter correction program will be executed and retrained until the goal is met. Further modifiable parameters include: initial parameters, learning rate, regularization factor, decay rate, number of neurons in each layer, number of neuron levels, small batch size, training sample group and validation sample group, and continue to monitor and parameterize Adjust until condition improves.

When the training result satisfies that the Loss is less than 2% and there is no over-fitting to achieve convergence and obtain the optimized cardiac ultrasound judgment model 400, the aforementioned cardiac ultrasound judgment model 400 outputs a plurality of cardiac ultrasound judgment results, and each cardiac The ultrasonic judgment results are compared with the corresponding diagnostic results of each cardiac catheterization to verify the accuracy of the cardiac ultrasonic judgment model 400. If the comparison result fails, return to the model training step and execute the deep learning parameter correction procedure.

In summary, the present invention has the following advantages: one, the present invention utilizes Three-dimensional deep learning is used to train the cardiac ultrasound judgment model, and the cardiac ultrasound judgment results are compared with the cardiac catheter diagnosis results to correct the cardiac ultrasound judgment model, and finally an optimized cardiac ultrasound judgment model is obtained. Second, the present invention can analyze the target cardiac ultrasound image in real time through the cardiac ultrasound judgment model to generate the target cardiac ultrasound judgment result corresponding to the infarct-related blood vessel, thereby improving the problem that the human eye cannot identify regional abnormal wall motion, and achieving Prepare early for subsequent cardiac catheterization.

Although the present invention has been disclosed above in terms of implementation, it is not intended to limit the present invention. Anyone skilled in this art can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection of the present invention The scope shall be defined by the appended patent application scope.

100: Establishment method of cardiac ultrasound judgment model using three-dimensional deep learning

110, 120, 130, 140: steps

200: Cardiac Ultrasound Judgment System Using 3D Deep Learning

210: Ultrasonic detector

220: Processor

221:Refer to the information to obtain the module

222: Refer to image preprocessing module

223: Reference feature selection module

224:Model training module

225: Analysis module

300: Cardiac ultrasound judgment method using three-dimensional deep learning

310, 320, 330, 340: steps

400: Cardiac Ultrasound Judgment Model

401: Refer to cardiac ultrasound images

500: Dense Convolutional Neural Network Models

510:Dense Convolutional Neural Network Module

511,512,513,514: Dense Convolutional Neural Network Units

515: input terminal

516: output terminal

520: Maximum pooling module

530: flat module

540: Three-dimensional convolution layer

550: Linear rectification unit training layer

560: Batch normalization layer

600: Fully connected neural network model

700:Normalized exponential function model

800: Deep Learning Probability

900: Judgment result of cardiac ultrasound

X ¹ , X ² , X ³ : input value

W ¹ , W ² , W ³ : weights

C: cascade unit

μ: average value

σ: standard deviation

β, γ: parameters

à ¹ , à ² , à ³ ,  ¹ ,  ² ,  ³ : numeric values

In order to make the above and other objects, features, advantages and embodiments of the present invention more clearly understood, the accompanying drawings are described as follows: Figure 1 is a flow chart showing the steps of a method for establishing a cardiac ultrasound judgment model using three-dimensional deep learning according to an embodiment of the present invention; Fig. 2 is a schematic block diagram of a heart ultrasound judgment system using three-dimensional deep learning according to another embodiment of the present invention; FIG. 3 is a flow chart showing the steps of a cardiac ultrasound judgment method using three-dimensional deep learning according to another embodiment of the present invention; Figure 4 is a schematic diagram showing the structure of the cardiac ultrasound judgment model of the present invention; FIG. 5 is a schematic diagram showing the architecture of the dense convolutional neural network module in FIG. 4; and FIG. 6 is a schematic diagram showing the architecture of the batch normalization layer in FIG. 5 .

200: Cardiac Ultrasound Judgment System Using 3D Deep Learning

210: Ultrasonic detector

220: Processor

221:Refer to the information to obtain the module

222: Refer to image preprocessing module

223: Reference feature selection module

224:Model training module

225: Analysis module

Claims (7)

A method for establishing a cardiac ultrasound judgment model using three-dimensional deep learning, including the following steps: performing a data acquisition step, which is to obtain a reference data, the reference data includes a plurality of reference cardiac ultrasound data and the corresponding reference cardiac ultrasound data A plurality of cardiac catheterization diagnosis results of the sound wave data; an image preprocessing step, which is to cut each of the reference heart ultrasound data, to obtain a plurality of reference heart ultrasound images; a reference feature selection step, which is to use a reference After the feature selection module analyzes each of the reference echocardiographic images, a reference image feature value is obtained from each of the reference echocardiographic images, so as to obtain a plurality of reference image feature values corresponding to the reference echocardiographic images; and Carry out a model training step, which is to generate a cardiac ultrasound judgment model based on the reference image feature value training, and input the reference cardiac ultrasound data to the cardiac ultrasound judgment model to generate corresponding reference cardiac ultrasound Multiple cardiac ultrasound judgment results of the data, and then compare these cardiac ultrasound judgment results with these cardiac catheterization diagnosis results to modify the cardiac ultrasound judgment model, wherein the cardiac ultrasound judgment model includes a sequentially connected A dense convolutional neural network (DenseNet) model, a fully-connected neural network (Fully-connected neural network, FCNN) model and a normalized exponential function (Softmax) model; wherein, in the model training step, a depth Learning a parameter correction program and retraining the cardiac ultrasound judgment model until one of the loss functions of the cardiac ultrasound judgment model is less than 2%. A method for establishing a cardiac ultrasound judgment model using three-dimensional deep learning as described in claim 1, wherein the dense convolutional neural network model includes a plurality of dense convolutional neural network modules (DenseNet Block, DNB), a plurality of maximum Pooling module (Max pooling) and a flat module (Flatten). A method for establishing a cardiac ultrasound judgment model using three-dimensional deep learning as described in claim 2, wherein each of the dense convolutional neural network modules includes a plurality of dense convolutional neural network units, and these dense convolutional neural networks The units are cascaded to each other in sequence, and each dense convolutional neural network unit includes a three-dimensional convolutional layer (Conv), a linear rectification unit training layer (ReLU), and a batch of normalization layers (Batch Normalization, BN). A heart ultrasound judgment system using three-dimensional deep learning, including: an ultrasound detector, used to generate a target heart ultrasound image of a subject; and a processor, electrically connected to the ultrasound detector, the The processor stores a program. When the program is executed by the processor, the program judges whether the subject has an infarct-related artery (Infarct Related Artery, IRA) according to the target cardiac ultrasound image. The program includes: a The reference data acquisition module is used to obtain a reference data, which includes a plurality of reference echocardiographic data and corresponding parameters Diagnosis results of multiple cardiac catheterizations based on cardiac ultrasound data; a reference image preprocessing module, which is used to cut each reference cardiac ultrasound data to obtain multiple reference cardiac ultrasound images; a reference feature selection module , which is used to obtain a reference image feature value in each of the reference echocardiogram images after analyzing each of the reference echocardiogram images, so as to obtain a plurality of reference image feature values corresponding to the reference echocardiogram images; A model training module, which generates a cardiac ultrasound judgment model based on the reference image feature value training, and inputs the reference cardiac ultrasound data into the cardiac ultrasound judgment model to generate the corresponding reference cardiac ultrasound Multiple cardiac ultrasound judgment results of the data, and then compare these cardiac ultrasound judgment results with these cardiac catheter diagnosis results to modify the cardiac ultrasound judgment model, the model training module executes a deep learning parameter correction program and Retrain the heart ultrasound judgment model until a loss function of the heart ultrasound judgment model is less than 2%, wherein the heart ultrasound judgment model includes a dense convolutional neural network (DenseNet) model connected in sequence, a A fully-connected neural network (Fully-connected neural network, FCNN) model and a normalized exponential function (Softmax) model; and an analysis module, which analyzes the target cardiac ultrasound image based on the cardiac ultrasound judgment model A target cardiac ultrasonic judgment result corresponding to one of the blood vessels related to the infarction is generated. Cardiac ultrasound of three-dimensional deep learning as described in claim 4 A wave judgment system, wherein the dense convolutional neural network model includes a plurality of dense convolutional neural network modules (DenseNet Block, DNB), a plurality of maximum pooling modules (Max pooling) and a flat module (Flatten). A cardiac ultrasound judgment method using three-dimensional deep learning, comprising the following steps: providing a cardiac ultrasound judgment model as in claim 1; providing a target cardiac ultrasound image of a subject; according to the cardiac ultrasound judgment model Analyzing the target ultrasound image of the heart to generate a target ultrasound diagnosis result; and using the target ultrasound diagnosis result to determine whether the subject has an Infarct Related Artery (IRA). The cardiac ultrasound judgment method using three-dimensional deep learning as described in claim 6, wherein the dense convolutional neural network model includes a plurality of dense convolutional neural network modules (DenseNet Block, DNB), a plurality of maximum pooling models group (Max pooling) and a flat module (Flatten).

Images