CN111904469A - Heart section detection method and system capable of realizing parallel processing - Google Patents

Abstract

The present application relates to a method and system for detecting a cardiac slice that can be processed in parallel. The method includes: acquiring a heart image to be detected; performing target detection on the heart image to be detected, and obtaining corresponding target information, the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the heart image to be detected, and the information to be detected. The slice type of the target cardiac slice corresponding to the cardiac image; according to the target information, determine whether the target cardiac slice is a standard slice. The method can improve the reliability of the detection result of the standard slice of the heart.

Description

Cardiac slice detection method and system capable of parallel processing

technical field

The present application relates to the technical field of prenatal ultrasound examination, and in particular, to a method and system for detecting cardiac slices that can be processed in parallel.

Background technique

The detection of fetal cardiac development is of great significance for the detection of fetal cardiac dysplasia and congenital heart disease. At this stage, the fetal heart section images are mainly obtained by ultrasound examination, and the cardiac section images are detected to obtain the cardiac standard section, and then the fetal development is analyzed based on the cardiac standard section. Current standard cardiac slice detection methods have problems of poor interpretability and low reliability.

SUMMARY OF THE INVENTION

Based on this, it is necessary to provide a cardiac slice detection method, system, computer equipment and storage medium that can improve the reliability of the cardiac standard slice detection results in view of the above technical problems.

A method for detecting a cardiac section, the method comprising:

Obtain the heart image to be detected;

Perform target detection on the to-be-detected cardiac image to obtain corresponding target information, where the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the to-be-detected cardiac image, and the to-be-detected cardiac image The slice type of the corresponding target cardiac slice;

According to the target information, it is determined whether the target cardiac slice is a standard slice.

A cardiac section detection system, the system includes:

an image acquisition module for acquiring the heart image to be detected;

a target detection module, configured to perform target detection on the to-be-detected cardiac image to obtain corresponding target information, where the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the to-be-detected cardiac image, and The slice type of the target cardiac slice corresponding to the cardiac image to be detected;

A standard slice determination module, configured to determine whether the target cardiac slice is a standard slice according to the target information.

A computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

Obtain the heart image to be detected;

Perform target detection on the to-be-detected cardiac image to obtain corresponding target information, where the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the to-be-detected cardiac image, and the to-be-detected cardiac image The slice type of the corresponding target cardiac slice;

According to the target information, it is determined whether the target cardiac slice is a standard slice.

A computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

Obtain the heart image to be detected;

Perform target detection on the to-be-detected cardiac image to obtain corresponding target information, where the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the to-be-detected cardiac image, and the to-be-detected cardiac image The slice type of the corresponding target cardiac slice;

According to the target information, it is determined whether the target cardiac slice is a standard slice.

The above-mentioned cardiac section detection method, system, computer equipment and storage medium obtain the cardiac image to be detected; perform target detection on the cardiac image to be detected to obtain corresponding target information, and the target information includes: the structural type of each target cardiac structure in the cardiac image to be detected and the corresponding confidence level, and the slice type of the target cardiac slice corresponding to the cardiac image to be detected; according to the target information, determine whether the target cardiac slice is a standard slice. Therefore, according to the structure type and corresponding confidence level of each target cardiac structure detected from the cardiac image to be detected, as well as the detected slice type of the target cardiac slice, it is comprehensively determined whether the target cardiac slice is a standard slice, which has better results. Interpretability, thereby increasing the reliability of detection results.

Description of drawings

1 is a schematic flowchart of a method for detecting a cardiac section in one embodiment;

2 is a schematic structural diagram of a target detection model in one embodiment;

Fig. 3 is the labeling schematic diagram of various cardiac slices in one embodiment;

4 is a schematic diagram of the detection results and scoring results of a standard four-chamber slice in one embodiment;

5 is a schematic diagram of the detection results and scoring results of a non-standard four-chamber slice in one embodiment;

Fig. 6 is the schematic diagram of the detection result and scoring result of the standard 3VT section in one embodiment;

Fig. 7 is the schematic diagram of the detection result and scoring result of non-standard 3VT section in one embodiment;

8 is a schematic diagram of the detection results and scoring results of a standard right ventricular outflow tract section in one embodiment;

9 is a schematic diagram of the detection results and scoring results of a non-standard right ventricular outflow tract section in one embodiment;

10 is a schematic diagram of the detection results and scoring results of a standard left ventricular outflow tract section in one embodiment;

11 is a schematic diagram of the detection results and scoring results of a non-standard left ventricular outflow tract section in one embodiment;

12 is a schematic diagram of the segmentation result of each key cardiac structure in the standard view of a four-chamber heart in one embodiment;

13 is a schematic flowchart of a method for detecting a cardiac slice in one embodiment;

14 is a schematic diagram of a parallelized pipeline method in one embodiment;

15 is a structural block diagram of a cardiac slice detection system in one embodiment;

Figure 16 is an internal structure diagram of a computer device in one embodiment;

Figure 17 is a diagram of the internal structure of a computer device in one embodiment.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application more clearly understood, the present application will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application.

In one embodiment, as shown in FIG. 1 , a method for detecting a cardiac slice is provided, including the following steps S102 to S106 .

S102, acquiring a heart image to be detected.

Specifically, the cardiac ultrasound video of the detected object may be acquired, the cardiac ultrasound video may be analyzed, and frame pictures corresponding to various cardiac slices may be obtained as the cardiac image to be detected. When applied to obstetric inspection, the detected object represents the fetus, and the heart image to be detected represents the fetal heart image.

S104, perform target detection on the cardiac image to be detected to obtain corresponding target information, where the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the cardiac image to be detected, and the slice of the target cardiac slice corresponding to the cardiac image to be detected type.

Among them, the target cardiac structure represents the detected cardiac structure, the structure type of the target cardiac structure represents the detected cardiac structure type (such as left atrium, right atrium, etc.), and the confidence level can be understood as the target cardiac structure belongs to the corresponding structure type. probability. The target cardiac slice represents the detected cardiac slice, and the slice type of the target cardiac slice represents the type of the detected cardiac slice (eg, four-chamber slice, 3VT slice, etc.).

Specifically, the types of cardiac views may include a four-chamber view, a 3VT view, a right ventricular outflow tract view, and a left ventricular outflow tract view. Among them, the four-chamber view may include the following types of cardiac structures: left ventricle, left atrium, right ventricle, right atrium, ribs, descending aorta, spine. 3VT views can include the following types of cardiac structures: main pulmonary artery and ductus arteriosus, aortic arch, superior vena cava, trachea, descending aorta, spine. The right ventricular outflow tract view can include the following types of cardiac structures: main pulmonary artery and ductus arteriosus, aortic arch, right ventricle, superior vena cava, descending aorta, spine. Left ventricular outflow tract views can include the following types of cardiac structures: left ventricle, left ventricular outflow tract and aorta, right ventricle, interventricular septum, spine.

S106, according to the target information, determine whether the target cardiac slice is a standard slice.

For any cardiac image to be detected, it is determined whether the target cardiac slice is a standard slice according to the detected structural type and corresponding confidence level of each target cardiac structure, and the detected slice type of the target cardiac slice.

For example, assuming that the slice type of the detected target cardiac slice is a four-chamber slice, it can be determined whether the standard conditions corresponding to the four-chamber slice are satisfied according to the detected structural type and corresponding confidence level of each target cardiac slice. , if it is satisfied, the detected target cardiac slice is determined as the standard four-chamber slice. The standard conditions may include: the detected structure type of each target cardiac structure includes the key cardiac structures (left ventricle, left atrium, right ventricle, right atrium, spine, descending aorta, and ribs) of the four-chamber cardiac view, or The method further includes: the score obtained by scoring according to the structure type of each target cardiac structure and the corresponding confidence level exceeds a certain threshold.

In the above-mentioned cardiac slice detection method, the cardiac image to be detected is obtained; the target detection is performed on the cardiac image to be detected, and corresponding target information is obtained, and the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the cardiac image to be detected, and The slice type of the target cardiac slice corresponding to the heart image to be detected; according to the target information, it is determined whether the target cardiac slice is a standard slice. Therefore, according to the structure type and corresponding confidence level of each target cardiac structure detected from the cardiac image to be detected, as well as the detected slice type of the target cardiac slice, it is comprehensively determined whether the target cardiac slice is a standard slice, which has better results. Interpretability, thereby increasing the reliability of detection results.

In one embodiment, the step of performing target detection on the heart image to be detected and obtaining corresponding target information may specifically be: using a trained target detection model to perform target detection on the detected heart image to obtain corresponding target information.

As shown in Figure 2, the structure of the target detection model can include a backbone network composed of convolutional layers and residual layers connected in sequence, and three branch networks composed of convolutional layers, upsampling layers and connection layers. Among them, the convolutional layer is filled with SAME method. Specifically, the convolutional layer can be a Darknetconv2D_BN_Leaky-type convolutional layer, which is filled in the SAME method, and then uses BatchNormalization, and uses a leaky linear rectification function (Leaky ReLU) as the activation function.

Specifically, the network architecture of the backbone network is as follows:

The first layer is the input layer, whose input is an image of size 800*800*3;

The second layer is the Darknetconv2D_BN_Leaky type convolutional layer, which receives the image from the input layer. The convolution kernel size of this layer is 3*3, the number is 32, the stride is 1, and the output size is a matrix of 800*800*32, denoted The layer is C1;

The third layer is the Darknetconv2D_BN_Leaky type convolutional layer, which receives the image from the input layer. The convolution kernel size of this layer is 3*3, the number is 64, the stride is 2, and the output size is a matrix of 400*400*64, denoted This layer is C2;

The fourth layer is the residual block (res1). First, this layer receives a matrix of size 400*400*64 output by the third layer and passes through two C1 convolution layers (the number of convolution kernels is 32 and 64 respectively), and the output size is It is a matrix of 400*400*64, and then the matrix is added with the matrix of size 400*400*64 output by the third layer to obtain a matrix of size 400*400*64, record the residual The difference block is R1;

The fifth layer is the residual block (res2). First, this layer receives a matrix of size 400*400*64 output by the fourth layer. After the C2 convolution layer (the number of convolution kernels is 128), the size is 200*200* 128 matrix, and then after two residual blocks R1 (the number of convolution kernels of the two C1 convolutional layers used by the two residual blocks R1 are 64 and 128 respectively), the size is 200*200*128 the size of the matrix;

The sixth layer is the residual block (res8). First, this layer receives a matrix of size 200*200*128 output by the fifth layer. After the C2 convolution layer (the number of convolution kernels is 256), the size is 100*100* 256 matrix, and then after eight residual blocks R1 (the number of convolution kernels of the two C1 convolutional layers used by the eight residual blocks R1 are 128 and 256 respectively), the size is 100*100*256 the size of the matrix;

The seventh layer is the residual block (res8). First, this layer receives a matrix of size 100*100*256 output by the sixth layer. After the C2 convolution layer (the number of convolution kernels is 512), the size is 50*50* 512 matrix, and then after eight residual blocks R1 (the number of convolution kernels of the two C1 convolutional layers used by the eight residual blocks R1 are 256 and 512 respectively), the size is 50*50*512 the size of the matrix;

The eighth layer is the residual block (res4). First, this layer receives a matrix of size 50*50*512 output by the seventh layer. After the C2 convolution layer (the number of convolution kernels is 1024), the size is 25*25* 1024 matrix, and then after eight residual blocks R1 (the number of convolution kernels of the two C1 convolutional layers used by the eight residual blocks R1 are 512 and 1024 respectively), the size is 25*25*1024 matrix.

The network architecture of the branch network is as follows:

The first layer is the Darknetconv2D_BN_Leaky type convolutional layer. This layer receives a matrix of size 25*25*1024 output by the backbone network. The size of the convolution kernel of this layer is 1*1, the number is 512, the step size is 1, and the output size is 25*25*512 matrix;

The second layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 3*3, the number is 1024, the stride is 1, and the output size is a matrix of 25*25*1024;

The third layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 1*1, the number is 512, the stride is 1, and the output size is a matrix of 25*25*512;

The fourth layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 3*3, the number is 1024, the stride is 1, and the output size is a matrix of 25*25*1024;

The fifth layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 1*1, the number is 512, the stride is 1, and the output size is a matrix of 25*25*512;

The sixth layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 3*3, the number is 256, the stride is 1, and the output size is a matrix of 25*25*256;

The seventh layer is the convolution layer. The size of the convolution kernel of this layer is 1*1, the number is 60, the step size is 1, and the output size is a matrix of 25*25*60. The output of this layer is the first branch child. The output of the network, from the first layer to the seventh layer is the network structure of the first branch sub-network;

The eighth layer is the Darknetconv2D_BN_Leaky type convolutional layer. This layer receives a matrix of size 25*25*512 output by the fifth layer. The size of the convolution kernel of this layer is 1*1, the number is 256, the step size is 1, and the output size is 1. is a matrix of 25*25*256;

The ninth layer is an upsampling layer, which receives a matrix with a size of 25*25*256 output by the eighth layer, the upsampling multiple of this layer is 2*2, and the output size is a matrix of 50*50*256;

The tenth layer is the connection layer (Concatenate), which combines the output of the ninth layer with a matrix of size 50*50*256 and the output of the seventh-layer residual block in the backbone network with a size of 50*50*512. Connect to get a matrix of size 50*50*768;

The eleventh layer is the Darknetconv2D_BN_Leaky type convolutional layer. This layer receives a matrix of size 50*50*768 output by the tenth layer. The size of the convolution kernel of this layer is 1*1, the number is 256, the step size is 1, and the output A matrix of size 50*50*256;

The twelfth layer is the Darknetconv2D_BN_Leaky type convolutional layer, the convolution kernel size of this layer is 3*3, the number is 512, the stride is 1, and the output size is a matrix of 50*50*512;

The thirteenth layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 1*1, the number is 256, the stride is 1, and the output size is a matrix of 50*50*256;

The fourteenth layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 3*3, the number is 512, the stride is 1, and the output size is a matrix of 50*50*512;

The fifteenth layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 1*1, the number is 256, the stride is 1, and the output size is a matrix of 50*50*256;

The sixteenth layer is the Darknetconv2D_BN_Leaky type convolution layer, the convolution kernel size of this layer is 3*3, the number is 128, the stride is 1, and the output size is a matrix of 50*50*128;

The seventeenth layer is a convolutional layer. The size of the convolution kernel of this layer is 1*1, the number is 60, the stride is 1, and the output size is a matrix of 50*50*60. The output of this layer is the second branch. The output of the sub-network, from the first layer to the fifth layer and the eighth layer to the seventeenth layer is the network structure of the second branch sub-network;

The eighteenth layer is a Darknetconv2D_BN_Leaky convolutional layer, which receives a matrix of size 50*50*256 output by the fifteenth layer. The size of the convolution kernel of this layer is 1*1, the number is 128, and the step size is 1. Output a matrix of size 50*50*128;

The nineteenth layer is an upsampling layer, which receives a matrix of size 50*50*128 output by the eighteenth layer, the upsampling multiple of this layer is 2*2, and the output size is a matrix of 100*100*128;

The twentieth layer is the connection layer (Concatenate), which outputs a matrix of size 100*100*128 output by the nineteenth layer and a matrix of size 100*100*256 output by the sixth layer of residual blocks in the backbone network Connect to get a matrix of size 100*100*384;

The twenty-first layer is a Darknetconv2D_BN_Leaky convolutional layer, which receives a matrix of size 100*100*384 output by the twenty-first layer. The size of the convolution kernel of this layer is 1*1, the number is 128, and the stride is 1. , the output size is a matrix of 100*100*128;

The twenty-second layer is the Darknetconv2D_BN_Leaky convolutional layer. The convolution kernel size of this layer is 3*3, the number is 256, the stride is 1, and the output size is a matrix of 100*100*256;

The twenty-third layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 1*1, the number is 128, the stride is 1, and the output size is a matrix of 100*100*128;

The twenty-fourth layer is the Darknetconv2D_BN_Leaky convolutional layer, the convolution kernel size of this layer is 3*3, the number is 256, the stride is 1, and the output size is a matrix of 100*100*256;

The twenty-fifth layer is the Darknetconv2D_BN_Leaky type convolutional layer. The convolution kernel size of this layer is 1*1, the number is 128, the stride is 1, and the output size is a matrix of 100*100*128;

The twenty-sixth layer is a Darknetconv2D_BN_Leaky convolutional layer. The convolution kernel size of this layer is 3*3, the number is 64, the stride is 1, and the output size is a matrix of 100*100*64;

The twenty-seventh layer is a convolutional layer. The size of the convolution kernel of this layer is 1*1, the number is 60, the stride is 1, and the output size is a matrix of 100*100*60. The output of this layer is the third one. The output of the branch sub-network, from the first layer to the fifth layer, the eighth layer to the fifteenth layer and the eighteenth to the twenty-seventh layer, is the network structure of the third branch sub-network.

The target detection model can be trained by the following steps: acquiring the heart sample image and the corresponding labeling information, the labeling information includes: the position labeling information and structure labeling type of each cardiac structure in the heart sample image, and the position of the cardiac slice corresponding to the heart sample image Labeling information and slice labeling type; input the cardiac sample image into the target detection model to be trained to obtain the detection information corresponding to the cardiac sample image, the detection information includes: the position detection information of each cardiac structure in the cardiac sample image, the structure detection type and the corresponding detection information Confidence, as well as the slice position detection information, slice detection type and corresponding detection confidence of the cardiac slice corresponding to the heart sample image; based on the detection information and label information, adjust the parameters of the target detection model to be trained until the model training end condition is met, Obtain the trained object detection model.

The cardiac sample images include cardiac images corresponding to various cardiac view types, including four-chamber cardiac view, 3VT view, right ventricular outflow tract view and left ventricular outflow tract view. For any heart sample image, it can be marked by a professional doctor. The specific marking method can be to mark each heart structure in the heart sample image with a rectangular frame, and the position marking information of the heart structure includes the diagonal corners of the corresponding marked frame. Point coordinates, structure annotation type indicates the cardiac structure type indicated by the corresponding annotation box. After labeling each cardiac structure, a rectangular frame can be used to mark the cardiac section. The position labeling information of the cardiac section includes the coordinates of the diagonal points of the corresponding labeling box. The labeling type of the section indicates the heart indicated by the corresponding labeling box. Slice type.

Specifically, FIG. 3 shows a standard four-chamber view, a non-standard four-chamber view, a standard 3VT view and a non-standard 3VT view, a right ventricular outflow tract view, a non-standard right ventricular outflow tract view, a left ventricular Annotated diagram of outflow tract view, non-standard left ventricular outflow tract view.

After the heart sample images are obtained, each heart sample image can be normalized, for example, the pixel range of the image is converted from 0 to 255 to 0 to 1, so as to improve the convergence speed during model training. The normalized heart sample images are divided into training set, validation set and test set according to the ratio of 8:1:1.

Input the training set into the target detection model to be trained, and obtain the detection information corresponding to the cardiac sample image. For any cardiac sample image, the detection information includes the information corresponding to each detection frame in the cardiac sample image, and the peripheral detection frame corresponds to the detected cardiac section, Each detection frame included in the peripheral detection frame corresponds to a detected cardiac structure, the position detection information of the cardiac structure includes the coordinates of the diagonal points of the corresponding detection frame, the structure detection type indicates the cardiac structure type indicated by the corresponding detection frame, and the confidence level indicates The confidence level of the corresponding detection box. The slice position detection information of the cardiac slice includes the coordinates of the diagonal points of the corresponding detection frame, the slice detection type indicates the cardiac slice type indicated by the corresponding detection frame, and the confidence level indicates the confidence level of the corresponding detection frame.

The Adam optimizer is used to iteratively train the target detection model to be trained, the loss value is calculated based on the difference between the detection information and the labeling information, and the parameters of the target detection model to be trained are adjusted until the end condition of the model training is met, thereby obtaining a trained target detection model. It may be that when the preset number of iterations is reached, or the loss value is less than the preset threshold, it is determined that the model training end condition is satisfied. In one embodiment, the initial learning rate (LearningRate) in the iterative training process is set to 0.0001, the batch size (BatchSize) is set to 4, and the number of iterations is set to 50.

In the above embodiment, a target detection model can be used to detect both the cardiac structure contained in the cardiac image and the slice type of the cardiac image, thereby achieving a faster detection speed, shortening the detection time, and improving the detection efficiency.

Use the test set to test the model, test the effect of each cardiac structure in the detected cardiac section, and use the accuracy rate (Precision), the recall rate (Recall), and the average accuracy rate (mAP) as the evaluation criteria to measure the detection results of the model, as follows shown in Table 1. In addition, the precision and recall rate (Recall) of detecting standard and non-standard sections are further tested to measure the detection effect of standard and non-standard sections, as shown in Table 2 below.

Table 1

section Precision Recall mAP Four-chamber view 0.872 0.968 0.958 3VT slice 0.918 0.951 0.964 Right ventricular outflow tract view 0.805 0.96 0.916 Left ventricular outflow tract view 0.902 0.923 0.937

Table 2

section Precision Recall Standard four-chamber view 0.892 0.953 Non-standard four-chamber view 0.911 0.961 Standard 3VT slice 0.957 0.929 Non-standard 3VT slices 0.986 0.932 Standard right ventricular outflow tract view 0.927 0.948 Nonstandard right ventricular outflow tract view 0.972 0.915 Standard left ventricular outflow tract view 0.914 0.968 Nonstandard left ventricular outflow tract view 0.899 0.974

From the above table 1, it can be seen that the four-chamber view, 3VT view, right ventricular outflow tract view and left ventricular outflow tract view have high accuracy and can basically meet the clinical needs; The standard and non-standard aspects corresponding to each aspect have higher accuracy and recall rate, and the detection effect is better.

In one embodiment, the step of determining whether the target cardiac section is a standard section according to the target information may specifically include the following steps: according to the structural type of each target cardiac structure, judging whether the target cardiac section includes the basic cardiac structure corresponding to the section type; When the target cardiac section contains the corresponding basic cardiac structure, the evaluation score of the target cardiac section is calculated according to the structure type and corresponding confidence level of each target cardiac structure; according to the evaluation score, it is determined whether the target cardiac section is a standard section.

In one embodiment, for a heart image to be detected, if the slice type of the detected target cardiac slice is a four-chamber slice, it is determined whether it is a standard four-chamber slice by the following method: according to each detected target heart slice The structure type of the structure, determine whether all the detected target cardiac structures include at least one left ventricle, left atrium, right ventricle, right atrium, spine, descending aorta and at least two ribs, if not, determine the target cardiac section It is a non-standard four-chamber view. If it is, it will be scored according to the structural type of each target cardiac structure and the corresponding confidence. The specific scoring rules are as follows:

Score=confidence(left ventricle)*factor(left ventricle)+confidence(left atrium)*factor(left atrium)+confidence(right ventricle)*factor(right ventricle)+confidence(right atrium)*factor(right atrium)+ confidence(spine)*factor(spine)+confidence(descending aorta)*factor(descending aorta)+confidence(rib)*factor(rib);

Among them, confidence represents the target confidence of each cardiac structure detection frame, the ribs select the two detection frames with the highest confidence, and the average of their confidences is used as the target confidence. Other cardiac structures (left ventricle, left atrium, right ventricle, right Atrium, spine, descending aorta) select a detection frame with the highest confidence, and use its confidence as the target confidence. factor represents the weight of each cardiac structure in the four-chamber view. In one embodiment, the proportions of the left ventricle, left atrium, right ventricle, right atrium, spine, descending aorta, and ribs in the four-chamber view The weights are set to 15, 15, 15, 15, 10, 10, 20, respectively. Score represents the evaluation score. If the evaluation score is higher than the preset score threshold, it is determined that the above four-chamber view is a standard four-chamber view. If it is lower than or equal to the preset score threshold, it is determined that the four-chamber view is a non-standard four-chamber view. Cavity section. In one embodiment, the preset score threshold may be set to 70. FIG. 4 shows a schematic diagram of the detection results and scoring results of the standard four-chamber slice in one embodiment. FIG. 5 shows a schematic diagram of the detection results and scoring results of the non-standard four-chamber slice in one embodiment.

In one embodiment, for a heart image to be detected, if the detected slice type of the target cardiac slice is a 3VT slice, it is determined whether it is a standard 3VT slice by the following method: according to the detected structural type of each target cardiac structure , determine whether all the detected target cardiac structures include at least one aorta and ductus arteriosus, aortic arch, superior vena cava, trachea, descending aorta and spine, if not, determine that the target cardiac view is a non-standard 3VT view, if so , then the scoring is performed according to the structural type and corresponding confidence of each target cardiac structure. The specific scoring rules are as follows:

Score=confidence (aorta and ductus arteriosus)*factor (aorta and ductus arteriosus)+confidence (aortic arch)*factor (aortic arch)+confidence (superior vena cava)*factor (superior vena cava)+confidence (trachea)*factor (trachea)+confidence(descending aorta)*factor(descending aorta)+confidence(spine)*factor(spine);

Among them, confidence represents the target confidence of each cardiac structure detection frame. The aorta and arterial duct, aortic arch, superior vena cava, trachea, descending aorta and spine all select the detection frame with the highest confidence and take its confidence as the target confidence. Spend. factor represents the weight of each cardiac structure in the 3VT view. In one embodiment, the weights of the aorta and the ductus arteriosus, the aortic arch, the superior vena cava, the trachea, the descending aorta, and the spine in the 3VT view are respectively set as 20, 20, 15, 20, 15, 10. Score represents the evaluation score. If the evaluation score is higher than the preset score threshold, it is determined that the above-mentioned 3VT slice is a standard 3VT slice, and if it is lower than or equal to the preset score threshold, it is determined that the above-mentioned 3VT slice is a non-standard 3VT slice. In one embodiment, the preset score threshold may be set to 70. FIG. 6 shows a schematic diagram of the detection results and scoring results of standard 3VT slices in one embodiment. FIG. 7 shows a schematic diagram of the detection results and scoring results of non-standard 3VT slices in one embodiment.

In one embodiment, for a heart image to be detected, if the detected slice type of the target cardiac slice is a right ventricular outflow tract slice, it is determined whether it is a standard right ventricular outflow tract slice by the following method: The structure type of the target cardiac structure, determine whether all the detected target cardiac structures include at least one aorta and ductus arteriosus, right ventricle, spine, aortic arch, descending aorta and superior vena cava, if not, determine the target cardiac section It is a non-standard right ventricular outflow tract view. If it is, it will be scored according to the structural type of each target cardiac structure and the corresponding confidence. The specific scoring rules are as follows:

Score=confidence(aorta and ductus arteriosus)*factor(aorta and ductus arteriosus)+confidence(right ventricle)*factor(right ventricle)+confidence(spine)*factor(spine)+confidence(aortic arch)*factor(aortic arch) )+confidence(descending aorta)*factor(descending aorta)+confidence(superior vena cava)*factor(superior vena cava);

Among them, confidence represents the target confidence of each cardiac structure detection frame. The aorta and ductus arteriosus, right ventricle, spine, aortic arch, descending aorta, and superior vena cava all select the detection frame with the highest confidence, and take its confidence as the target. Confidence. The factor represents the weight of each cardiac structure in the right ventricular outflow tract view. In one embodiment, the aorta and the ductus arteriosus, the right ventricle, the spine, the aortic arch, the descending aorta, and the superior vena cava are in the right ventricular outflow tract view. The occupied weights are set to 15, 20, 20, 15, 15, and 15, respectively. Score represents the evaluation score. If the evaluation score is higher than the preset score threshold, it is determined that the above RV outflow tract section is a standard RV outflow tract section. If it is lower than or equal to the preset score threshold, it is determined that the above RV outflow tract section is Nonstandard right ventricular outflow tract view. In one embodiment, the preset score threshold may be set to 70. FIG. 8 shows a schematic diagram of the detection results and scoring results of a standard right ventricular outflow tract section in one embodiment. FIG. 9 shows a schematic diagram of the detection results and scoring results of a non-standard right ventricular outflow tract section in one embodiment.

In one embodiment, for a heart image to be detected, if the slice type of the detected target cardiac slice is a left ventricular outflow tract slice, it is determined whether it is a standard left ventricular outflow tract slice by the following method: Structural type of the target cardiac structure, determine whether all detected target cardiac structures include at least one left ventricle, right ventricle, left ventricular outflow tract, aorta, spine and ventricular septum, if not, determine that the target cardiac section is non-standard The left ventricular outflow tract section, if yes, will be scored according to the structural type of each target cardiac structure and the corresponding confidence. The specific scoring rules are as follows:

Score=confidence(left ventricle)*factor(left ventricle)+confidence(right ventricle)*factor(right ventricle)+confidence(left ventricular outflow tract and aorta)*factor(left ventricular outflow tract and aorta)+confidence( spine)*factor(spine)+confidence(ventricular septum)*factor(ventricular septum);

Among them, confidence represents the target confidence level of each cardiac structure detection frame. The left ventricle, right ventricle, left ventricular outflow tract, aorta, spine, and interventricular septum are selected for the detection frame with the highest confidence level, and its confidence level is used as the target confidence level. . factor represents the weight of each cardiac structure in the left ventricular outflow tract view. In one embodiment, the left ventricle, right ventricle, left ventricular outflow tract, and the aorta, spine, and ventricular septum in the left ventricular outflow tract view are all weights. The weights are set to 25, 25, 20, 15, and 15, respectively. Score represents the evaluation score. If the evaluation score is higher than the preset score threshold, it is determined that the above-mentioned left ventricular outflow tract view is a standard left ventricular outflow tract view. Nonstandard left ventricular outflow tract view. In one embodiment, the preset score threshold may be set to 70. Figure 10 shows a schematic diagram of the detection results and scoring results of a standard left ventricular outflow tract section in one embodiment. FIG. 11 shows a schematic diagram of the detection results and scoring results of a non-standard left ventricular outflow tract section in one embodiment.

In the above embodiment, based on the detailed structure scoring mechanism, the specific score value of whether each target cardiac section detected is a standard section is obtained, which has high interpretability in ultrasound prenatal medicine, and improves the detection results of the standard cardiac section. credibility.

In one embodiment, the target information obtained by performing target detection on the heart image to be detected further includes position information of each target cardiac structure, and when it is determined that the target cardiac slice is a standard slice, the following steps are also included: according to the position of each target cardiac structure The target cardiac structure image corresponding to each target cardiac structure is extracted from the to-be-detected cardiac image; each target cardiac structure image is segmented to obtain contour information of the target cardiac structure in each target cardiac structure image.

For any detected target heart structure, its position information includes the coordinates of the diagonal points of the corresponding detection frame, which are assumed to be represented by (x _min , y _min ) and (x _max , y _max ) respectively, and the coordinates of the diagonal points are outward Expand N pixels around to obtain the coordinates of the diagonal points of the cropping frame, which are represented by (x _min -N, y _min -N) and (x _max +N, y _max +N) respectively, where N can be set according to the actual situation , for example, it can be set to 10. The target cardiac structure is cropped from the to-be-detected cardiac image according to the coordinates of the diagonal points of the cropping frame, and a target cardiac structure image corresponding to the target cardiac structure is obtained.

In one embodiment, the step of segmenting each target cardiac structure image to obtain contour information of the target cardiac structure in each target cardiac structure image may specifically be: for any target cardiac structure image, use the corresponding trained image The segmentation model is used to segment the target cardiac structure image to obtain contour information of the target cardiac structure in each target cardiac structure image.

The structure of the image segmentation model can include convolutional layers, pooling layers, regularization layers, upsampling layers and feature fusion layers. Among them, the convolutional layer is filled with the SAME method, and the linear rectification function (ReLU) is used as the activation function.

Specifically, the architecture of the image segmentation network model is as follows:

The first layer is the input layer, whose input is an image of size 128*128*3;

The second layer is the first convolutional layer, which receives the image from the input layer, the convolution kernel size of this layer is 3*3, the number is 64, the stride is 1, and the output size is a matrix of 128*128*64;

The third layer is the second layer of convolution layer, the size of the convolution kernel of this layer is 3*3, the number is 64, the stride is 1, and the output size is a matrix of 128*128*64;

The fourth layer is the first pooling layer, the pooling size is 2*2, and the output size is a matrix of 64*64*64;

The fifth layer is the third convolutional layer, the convolution kernel size of this layer is 3*3, the number is 128, the stride is 1, and the output size is a matrix of 64*64*128;

The sixth layer is the fourth layer of convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 128, the stride is 1, and the output size is a matrix of 64*64*128;

The seventh layer is the second pooling layer, the pooling size is 2*2, and the output size is a matrix of 32*32*128;

The eighth layer is the fifth convolutional layer, the convolution kernel size of this layer is 3*3, the number is 256, the stride is 1, and the output size is a matrix of 32*32*256;

The ninth layer is the sixth convolution layer. The size of the convolution kernel of this layer is 3*3, the number is 256, the stride is 1, and the output size is a matrix of 32*32*256;

The tenth layer is the third pooling layer, the pooling size is 2*2, and the output size is a matrix of 16*16*256;

The eleventh layer is the seventh layer of convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 512, the stride is 1, and the output size is a matrix of 16*16*512;

The twelfth layer is the eighth convolution layer. The size of the convolution kernel of this layer is 3*3, the number is 512, the stride is 1, and the output size is a matrix of 16*16*512;

The thirteenth layer is the first dropout regularization layer (Dropout), the dropout rate is 0.5, and the output size is a matrix of 16*16*512;

The fourteenth layer is the fourth pooling layer, the pooling size is 2*2, and the output size is a matrix of 8*8*512;

The fifteenth layer is the ninth convolution layer. The size of the convolution kernel of this layer is 3*3, the number is 1024, the stride is 1, and the output size is a matrix of 8*8*1024;

The sixteenth layer is the tenth convolution layer. The size of the convolution kernel of this layer is 3*3, the number is 1024, the stride is 1, and the output size is a matrix of 8*8*1024;

The seventeenth layer is the second layer dropout regularization layer (Dropout), the dropout rate is 0.5, and the output size is a matrix of 8*8*1024;

The eighteenth layer is the first layer upsampling layer (UpSampling), the upsampling factor is 2*2, and the output size is a matrix of 16*16*1024;

The nineteenth layer is the eleventh convolutional layer, the convolution kernel size of this layer is 2*2, the number is 512, the stride is 1, and the output size is a matrix of 16*16*512;

The twentieth layer is the first feature fusion layer. This layer performs feature fusion on the output of the first layer discarding regularization layer and the eleventh layer convolution layer. The axis parameter axis is 3, and the output size is 16*16*1024 the matrix;

The twenty-first layer is the twelfth convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 512, the stride is 1, and the output size is a matrix of 16*16*512;

The twenty-second layer is the thirteenth convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 512, the stride is 1, and the output size is a matrix of 16*16*512;

The twenty-third layer is the second upsampling layer, the upsampling factor is 2*2, and the output size is a matrix of 32*32*512;

The twenty-fourth layer is the fourteenth convolutional layer. The size of the convolution kernel of this layer is 2*2, the number is 256, the stride is 1, and the output size is a matrix of 32*32*256;

The twenty-fifth layer is the second feature fusion layer. This layer performs feature fusion on the outputs of the sixth convolutional layer and the fourteenth convolutional layer. The axis parameter axis is 3, and the output size is 32*32*512 ;

The twenty-sixth layer is the fifteenth convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 256, the stride is 1, and the output size is a matrix of 32*32*256;

The twenty-seventh layer is the sixteenth convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 256, the stride is 1, and the output size is a matrix of 32*32*256;

The twenty-eighth layer is the third upsampling layer, the upsampling factor is 2*2, and the output size is a matrix of 64*64*256;

The twenty-ninth layer is the seventeenth convolutional layer. The size of the convolution kernel of this layer is 2*2, the number is 128, the stride is 1, and the output size is a matrix of 64*64*128;

The thirtieth layer is the third feature fusion layer, which performs feature fusion on the outputs of the fourth convolutional layer and the seventeenth convolutional layer. The axis parameter axis is 3, and the output size is 64*64*256. matrix;

The thirty-first layer is the eighteenth convolution layer. The size of the convolution kernel of this layer is 3*3, the number is 128, the stride is 1, and the output size is a matrix of 64*64*128;

The thirty-second layer is the nineteenth convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 128, the stride is 1, and the output size is a matrix of 64*64*128;

The thirty-third layer is the fourth upsampling layer, the upsampling factor is 2*2, and the output size is a matrix of 128*128*128;

The thirty-fourth layer is the twentieth convolutional layer. The size of the convolution kernel of this layer is 2*2, the number is 64, the stride is 1, and the output size is a matrix of 128*128*64;

The thirty-fifth layer is the fourth feature fusion layer. This layer performs feature fusion on the outputs of the second convolutional layer and the twentieth convolutional layer. The axis parameter axis is 3, and the output size is 128*128*128 the matrix;

The thirty-sixth layer is the twenty-first convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 64, the stride is 1, and the output size is a matrix of 128*128*64;

The thirty-seventh layer is the twenty-second convolutional layer. The size of the convolution kernel of this layer is 3*3, the number is 64, the stride is 1, and the output size is a matrix of 128*128*64;

The thirty-eighth layer is the twenty-third convolution layer. The size of the convolution kernel of this layer is 3*3, the number is 2, the stride is 1, and the output size is a matrix of 128*128*2;

The thirty-ninth layer is the twenty-fourth convolutional layer. The convolution kernel size of this layer is 1*1, the number is 1, the stride is 1, and the output size is a matrix of 128*128*1.

The image segmentation model can be trained by the following steps: acquiring the cardiac structure sample image and corresponding annotation information, the annotation information includes: the outline annotation information and structure annotation type of the cardiac structure in the cardiac structure sample image; input the cardiac structure sample image into the image to be trained The segmentation model is used to obtain segmentation information corresponding to the cardiac structure sample image, and the segmentation information includes: contour segmentation information, structure segmentation type and corresponding segmentation confidence of the cardiac structure in the cardiac structure sample image; based on the segmentation information and label information, adjust the image to be trained The parameters of the segmentation model are obtained until the end condition of the model training is met, and the trained image segmentation model is obtained.

The cardiac structure sample image is an image including a corresponding cardiac structure. For example, when the cardiac structure is a left ventricle, the corresponding cardiac structure sample image is an image including a left ventricular structure. For any cardiac structure sample image, it can be marked by a professional physician, and the specific marking method can be to mark the outline and type of the cardiac structure in the cardiac structure sample image as the outline annotation information and structure annotation type. Input the cardiac structure sample image into the image segmentation model to be trained, and obtain the segmentation information corresponding to the cardiac structure sample image, wherein the outline segmentation information of the cardiac structure represents the outline information output by the image segmentation model, and the structure segmentation type represents the structure type output by the image segmentation model, The segmentation confidence represents the confidence in the output of the image segmentation model. Based on the segmentation information and the labeling information, the parameters of the image segmentation model to be trained are adjusted until the end condition of the model training is satisfied, and the trained image segmentation model is obtained.

Stochastic gradient descent (SGD) is used to iteratively train the image segmentation model to be trained, and the loss value is calculated based on the difference between the segmentation information and the annotation information, and the parameters of the image segmentation model to be trained are adjusted until the end condition of the model training is met. Thereby, a trained image segmentation model is obtained. It may be that when the preset number of iterations is reached, or the loss value is less than the preset threshold, it is determined that the model training end condition is satisfied. In one embodiment, the initial learning rate (LearningRate) in the iterative training process is set to 0.002, the batch size (BatchSize) is set to 32, the impulse ξ is set to 0.8, and the number of iterations is set to 60.

In the above embodiment, using the specific positions of the target cardiac structures in the detection results of the target detection model to perform semantic segmentation on the structures in each target detection frame, compared to directly segmenting multiple structures in the entire cardiac image, it can be Improve the accuracy of segmentation results.

In one embodiment, after obtaining the contour information of each target cardiac structure, the following step may be further included: measuring each target cardiac structure according to the contour information of each target cardiac structure.

As shown in FIG. 12, it shows a schematic diagram of the segmentation result of each key cardiac structure (left atrium, left ventricle, right atrium, right ventricle, spine, descending aorta, rib) in the standard view of the four-chamber heart in one embodiment. Specifically, the measurement method can be determined according to the type of structure. For example, for the left atrium, the contour circumference and area can be measured according to the left atrial contour information obtained by segmentation; for ribs, the contour can be measured according to the rib contour information obtained by segmentation. length. Once the measurements are obtained, the development of individual cardiac structures can be assessed based on the measurements.

In one embodiment, as shown in FIG. 13 , a method for detecting a cardiac slice is provided, including the following steps S1301 to S1308 .

S1301, acquiring a heart image to be detected.

S1302, using the trained target detection model, perform target detection on the heart image to be detected, and obtain corresponding target information, where the target information includes: position information, structure type and corresponding confidence level of each target cardiac structure in the heart image to be detected, and The slice type of the target cardiac slice corresponding to the cardiac image to be detected.

S1303, according to the structure type of each target cardiac structure, determine whether the target cardiac slice contains the basic cardiac structure corresponding to the slice type, and if so, go to step S1304.

S1304, according to the structure type of each target cardiac structure and the corresponding confidence level, calculate and obtain the evaluation score of the target cardiac slice.

S1305, determine whether the evaluation score is higher than the preset score threshold, and if so, determine that the target cardiac slice is a standard slice, and proceed to step S1306.

S1306 , according to the position information of each target cardiac structure, extract a target cardiac structure image corresponding to each target cardiac structure from the cardiac image to be detected.

S1307 , using the trained image segmentation model, segment each target cardiac structure image, and obtain contour information of the target cardiac structure in each target cardiac structure image.

S1308, measure each target cardiac structure according to the contour information of each target cardiac structure.

For the specific limitations of the foregoing steps S1301 to S1308, reference may be made to the foregoing embodiments. In this embodiment, a target detection model can be used to detect both the cardiac structure contained in the cardiac image and the slice type of the cardiac image, thereby achieving faster detection speed, shortening detection time, and improving detection efficiency. Based on the detailed structure scoring mechanism, the specific score value of whether each target cardiac section detected is a standard section can be obtained, which has high interpretability in prenatal ultrasound and improves the reliability of the detection results of the standard cardiac section. Using the specific position of each target cardiac structure in the detection result of the target detection model, semantic segmentation is performed on the structure in each target detection frame, which can improve the accuracy of segmentation results compared to directly segmenting multiple structures in the entire cardiac image. .

In one embodiment, as shown in FIG. 14 , the processes of image preprocessing, target detection, prediction based on scoring rules, image segmentation, and structure measurement in the foregoing embodiment can be performed by using a parallelized pipeline method. Specifically, the parallel pipeline consists of five stages: image preprocessing, target detection, prediction based on scoring rules, image segmentation, and structure measurement. When the first frame of image is in the structure measurement stage, the second frame of image is in the image segmentation stage, and the third The frame image is in the prediction stage based on scoring rules, the fourth frame image is in the target detection stage, and the fifth frame image is in the image preprocessing stage. These five frame images can be processed at the same time, thus realizing a five-stage pipeline. Since the different steps in different image frames do not interfere with each other, the parallel pipeline method can be realized. Accordingly, the processing speed can be greatly accelerated, and the real-time acquisition and measurement of the standard slice of the heart can be guaranteed to a certain extent.

It should be understood that, although the steps in the flowcharts of FIGS. 1 and 13 are sequentially displayed according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, the execution of these steps is not strictly limited to the order, and these steps may be performed in other orders. Moreover, at least a part of the steps in FIGS. 1 and 13 may include multiple steps or multiple stages. These steps or stages are not necessarily executed and completed at the same time, but may be executed at different times. The execution of these steps or stages The order is also not necessarily sequential, but may be performed alternately or alternately with other steps or at least a portion of the steps or phases within the other steps.

In one embodiment, as shown in FIG. 15, a cardiac slice detection system 1500 is provided, including: an image acquisition module 1510, a target detection module 1520 and a standard slice determination module 1530, wherein:

The image acquisition module 1510 is used for acquiring the heart image to be detected.

The target detection module 1520 is used to perform target detection on the heart image to be detected, and obtain corresponding target information. The target information includes: the structural type and corresponding confidence level of each target cardiac structure in the heart image to be detected, and the corresponding confidence level of the heart image to be detected. The slice type of the target cardiac slice.

The standard slice determination module 1530 is configured to determine whether the target cardiac slice is a standard slice according to the target information.

In one embodiment, the target detection module 1520 is specifically configured to use a trained target detection model to perform target detection on the heart image to be detected to obtain corresponding target information.

In one embodiment, the target detection module 1520 includes a first training unit, which is used for training to obtain a target detection model, and is specifically used for: acquiring a heart sample image and corresponding labeling information, where the labeling information includes: each cardiac structure in the heart sample image Location annotation information and structure annotation type, as well as the location annotation information and slice annotation type of the cardiac slice corresponding to the cardiac sample image; input the cardiac sample image into the target detection model to be trained to obtain the detection information corresponding to the cardiac sample image, and the detection information includes: heart The position detection information, structure detection type and corresponding detection confidence of each cardiac structure in the sample image, as well as the slice position detection information, slice detection type and corresponding detection confidence of the cardiac slice corresponding to the cardiac sample image; based on the detection information and annotations information, adjust the parameters of the target detection model to be trained until the end condition of the model training is met, and obtain the trained target detection model.

In one embodiment, the standard slice determination module 1530 includes: a judgment unit, a scoring unit, and a determination unit. The judgment unit is used for judging whether the target cardiac slice contains the basic cardiac structure corresponding to the slice type according to the structure type of each target cardiac structure. The scoring unit is used for calculating the evaluation score of the target cardiac section according to the structure type of each target cardiac structure and the corresponding confidence when the target cardiac section includes the corresponding basic cardiac structure. The determining unit is used for determining whether the target cardiac slice is a standard slice according to the evaluation score.

In one embodiment, the slice type of the target cardiac slice is a four-chamber slice, and the judging unit is specifically configured to: when each target cardiac structure includes at least one left ventricle, left atrium, right ventricle, right atrium, spine, descending aorta and at least two ribs, it is determined that the target cardiac section contains the basic cardiac structure corresponding to the four-chamber cardiac section. The scoring unit is specifically used for: for the target heart structure whose structure type is any one of left ventricle, left atrium, right ventricle, right atrium, spine, and descending aorta, select the highest confidence level from the corresponding confidence level, as Corresponding target confidence; for the target heart structure whose structure type is rib, the highest two confidences are selected from the corresponding confidences, and the average of the two is used as the corresponding target confidence; based on the left ventricle, left atrium, right The sum of the products of the target confidence levels corresponding to the ventricle, right atrium, spine, descending aorta, and ribs and the preset weights was used to obtain the evaluation score of the four-chamber view.

In one embodiment, the slice type of the target cardiac slice is a 3VT slice, and the judging unit is specifically configured to: when each target cardiac structure includes at least one aorta and ductus arteriosus, aortic arch, superior vena cava, trachea, descending aorta, and spine When , it is determined that the target cardiac slice contains the basic cardiac structure corresponding to the 3VT slice. The scoring unit is specifically used to select the highest confidence level from the corresponding confidence levels for the target heart structure of any structure type: aorta and ductus arteriosus, aortic arch, superior vena cava, trachea, descending aorta, and spine. , as the corresponding target confidence; based on the sum of the products of the corresponding target confidence and the preset weight of the aorta and ductus arteriosus, aortic arch, superior vena cava, trachea, descending aorta, and spine, the evaluation score of the 3VT slice was obtained.

In one embodiment, the slice type of the target cardiac slice is a right ventricular outflow tract slice, and the judging unit is specifically configured to: when each target cardiac structure includes at least one aorta and ductus arteriosus, right ventricle, spine, aortic arch, descending aorta and superior vena cava, it is determined that the target cardiac view contains the basic cardiac structure corresponding to the right ventricular outflow tract view. The scoring unit is specifically used for: for the target heart structure of any one of the structure types of aorta and ductus arteriosus, right ventricle, spine, aortic arch, descending aorta, and superior vena cava, select the highest confidence level from the corresponding confidence levels. The right ventricular outflow tract section is obtained based on the sum of the products of the corresponding target confidence levels of the aorta and the ductus arteriosus, the right ventricle, the spine, the aortic arch, the descending aorta, and the superior vena cava and the preset weights evaluation score.

In one embodiment, the slice type of the target cardiac slice is a left ventricular outflow tract slice, and the judging unit is specifically configured to: when each target cardiac structure includes at least one left ventricle, right ventricle, left ventricular outflow tract and aorta, spine and ventricle During the interval, it is determined that the target cardiac view contains the basic cardiac structure corresponding to the left ventricular outflow tract view. The scoring unit is specifically used to select the highest confidence level from the corresponding confidence levels for the target cardiac structure whose structure type is any one of left ventricle, right ventricle, left ventricular outflow tract, aorta, spine, and interventricular septum. As the corresponding target confidence; based on the sum of the products of the corresponding target confidences of the left ventricle, right ventricle, left ventricular outflow tract, aorta, spine, and interventricular septum and the preset weight, the evaluation score of the left ventricular outflow tract section is obtained.

In one embodiment, the target information further includes: position information of each target cardiac structure in the cardiac image to be detected. The system also includes: an extraction module and a segmentation module. The extraction module is used for extracting the target cardiac structure image corresponding to each target cardiac structure from the to-be-detected cardiac image according to the position information of each target cardiac structure when the target cardiac slice is a standard slice. The segmentation module is used for segmenting each target cardiac structure image to obtain contour information of the target cardiac structure in each target cardiac structure image.

In one embodiment, the segmentation module is specifically used to: for any target cardiac structure image, use a corresponding trained image segmentation model to segment the target cardiac structure image to obtain the contour of the target cardiac structure in each target cardiac structure image information.

In one embodiment, the segmentation module includes a second training unit, which is used for training to obtain an image segmentation model, and is specifically used for: acquiring a cardiac structure sample image and corresponding labeling information, where the labeling information includes: the outline of the cardiac structure in the cardiac structure sample image Labeling information and structure labeling type; input the cardiac structure sample image into the image segmentation model to be trained to obtain the segmentation information corresponding to the cardiac structure sample image, and the segmentation information includes: the outline segmentation information of the cardiac structure in the cardiac structure sample image, the structure segmentation type and corresponding based on the segmentation information and labeling information, adjust the parameters of the image segmentation model to be trained until the end condition of the model training is satisfied, and obtain the trained image segmentation model.

In one embodiment, the system further includes: a measurement module, configured to measure each target cardiac structure according to the contour information of each target cardiac structure.

For the specific limitation of the cardiac slice detection system, reference may be made to the above limitation on the cardiac slice detection method, which will not be repeated here. Each module in the above-mentioned cardiac slice detection system may be implemented in whole or in part by software, hardware and combinations thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

In one embodiment, a computer device is provided, and the computer device may be a server, and its internal structure diagram may be as shown in FIG. 16 . The computer device includes a processor, memory, and a network interface connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium, an internal memory. The nonvolatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by a processor, implements a cardiac slice detection method.

In one embodiment, a computer device is provided, and the computer device may be a terminal, and its internal structure diagram may be as shown in FIG. 17 . The computer equipment includes a processor, memory, a communication interface, a display screen, and an input device connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium, an internal memory. The nonvolatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used for wired or wireless communication with an external terminal, and the wireless communication can be realized by WIFI, operator network, NFC (Near Field Communication) or other technologies. The computer program, when executed by a processor, implements a cardiac slice detection method. The display screen of the computer equipment may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment may be a touch layer covered on the display screen, or a button, a trackball or a touchpad set on the shell of the computer equipment , or an external keyboard, trackpad, or mouse.

Those skilled in the art can understand that the structure shown in FIG. 16 or FIG. 17 is only a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied. A computer device may include more or fewer components than those shown in the figures, or combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided, including a memory and a processor, where a computer program is stored in the memory, and the processor implements the steps in the foregoing method embodiments when the processor executes the computer program.

In one embodiment, there is provided a computer-readable storage medium having a computer program stored thereon, the computer program being implemented when executed by a processor.

In one embodiment, there is provided a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the foregoing method embodiments.

It should be understood that the terms "first", "second", etc. in the above embodiments are only used for description purposes, and cannot be interpreted as indicating or implying relative importance or implicitly indicating the number of indicated technical features.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage In the medium, when the computer program is executed, it may include the processes of the above-mentioned method embodiments. Wherein, any reference to memory, storage, database or other media used in the various embodiments provided in this application may include at least one of non-volatile and volatile memory. The non-volatile memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, and the like. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, the RAM may be in various forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM).

The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, all It is considered to be the range described in this specification.

The above-mentioned embodiments only represent several embodiments of the present application, and the descriptions thereof are specific and detailed, but should not be construed as a limitation on the scope of the invention patent. It should be noted that, for those skilled in the art, without departing from the concept of the present application, several modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the scope of protection of the patent of the present application shall be subject to the appended claims.

Claims (10)

1. a cardiac section detection method, is characterized in that, described method comprises: Obtain the heart image to be detected; Perform target detection on the to-be-detected cardiac image to obtain corresponding target information, where the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the to-be-detected cardiac image, and the to-be-detected cardiac image The slice type of the corresponding target cardiac slice; According to the target information, it is determined whether the target cardiac slice is a standard slice. 2. The method according to claim 1, wherein, performing target detection on the to-be-detected heart image to obtain corresponding target information, comprising: Using the trained target detection model, perform target detection on the to-be-detected heart image to obtain corresponding target information; The method for obtaining the target detection model by training includes: Obtain a cardiac sample image and corresponding annotation information, where the annotation information includes: position annotation information and structure annotation type of each cardiac structure in the cardiac sample image, and position annotation information and slice of the cardiac slice corresponding to the cardiac sample image label type; Inputting the heart sample image into the target detection model to be trained, to obtain detection information corresponding to the heart sample image, the detection information includes: position detection information of each cardiac structure in the heart sample image, structure detection type and corresponding Detection confidence, as well as slice position detection information, slice detection type, and corresponding detection confidence of the cardiac slice corresponding to the cardiac sample image; Based on the detection information and the label information, the parameters of the target detection model to be trained are adjusted until the model training end condition is satisfied, and a trained target detection model is obtained. 3. The method according to claim 1, wherein, according to the target information, determining whether the target cardiac slice is a standard slice, comprising: According to the structure type of each target cardiac structure, determine whether the target cardiac slice contains the basic cardiac structure corresponding to the slice type; When the target cardiac section includes a corresponding basic cardiac structure, calculating the evaluation score of the target cardiac section according to the structure type and corresponding confidence level of each of the target cardiac structures; According to the evaluation score, it is determined whether the target cardiac view is a standard view. 4 . The method according to claim 3 , wherein, according to the structure type of each target cardiac structure, it is determined whether the target cardiac slice contains the basic cardiac structure corresponding to the slice type, including the following items: 5 . any one of: the first item: The slice type of the target cardiac slice is a four-chamber slice, and when each of the target cardiac structures includes at least one left ventricle, left atrium, right ventricle, right atrium, spine, descending aorta, and at least two ribs, it is determined. The target cardiac view includes the basic cardiac structure corresponding to the four-chamber view; According to the structure type and corresponding confidence level of each target cardiac structure, the evaluation score of the target cardiac section is calculated, including: For the target heart structure whose structure type is any one of left ventricle, left atrium, right ventricle, right atrium, spine, and descending aorta, select the highest confidence level from the corresponding confidence level as the corresponding target confidence level; For the target heart structure whose structure type is rib, the two highest confidence levels are selected from the corresponding confidence levels, and the average of the two is used as the corresponding target confidence level; Based on the sum of the products of the target confidence levels corresponding to the left ventricle, the left atrium, the right ventricle, the right atrium, the spine, the descending aorta, and the rib and a preset weight, the evaluation score of the four-chamber view is obtained; second section: The section type of the target cardiac section is a 3VT section. When each of the target cardiac structures includes at least one aorta and ductus arteriosus, aortic arch, superior vena cava, trachea, descending aorta and spine, the target cardiac section is determined. Contains the basic cardiac structure corresponding to the 3VT section; According to the structure type and corresponding confidence level of each target cardiac structure, the evaluation score of the target cardiac section is calculated, including: For the target heart structure of any one of the aorta and ductus arteriosus, aortic arch, superior vena cava, trachea, descending aorta, and spine, select the highest confidence level from the corresponding confidence levels as the corresponding target confidence level Spend; Based on the sum of the products of the corresponding target confidence levels of the aorta and the ductus arteriosus, the aortic arch, the superior vena cava, the trachea, the descending aorta, and the spine and the preset weight, the evaluation score of the 3VT slice is obtained; the third item: The slice type of the target cardiac slice is the right ventricular outflow tract slice. When each of the target cardiac structures includes at least one aorta and ductus arteriosus, the right ventricle, the spine, the aortic arch, the descending aorta and the superior vena cava, it is determined that the The target cardiac view contains the basic cardiac structure corresponding to the right ventricular outflow tract view; According to the structure type and corresponding confidence level of each target cardiac structure, the evaluation score of the target cardiac section is calculated, including: For the target heart structure of any one of the aorta and ductus arteriosus, right ventricle, spine, aortic arch, descending aorta, and superior vena cava, select the highest confidence level from the corresponding confidence levels as the corresponding target Confidence; Based on the sum of the products of the target confidence levels corresponding to the aorta and the ductus arteriosus, the right ventricle, the spine, the aortic arch, the descending aorta, and the superior vena cava and the preset weight, the evaluation score of the right ventricular outflow tract section is obtained; Fourth item: The slice type of the target cardiac slice is the left ventricular outflow tract slice. When each of the target cardiac structures includes at least one left ventricle, right ventricle, left ventricular outflow tract, aorta, spine and ventricular septum, the target heart is determined. The view contains the basic cardiac structure corresponding to the left ventricular outflow tract view; According to the structure type and corresponding confidence level of each target cardiac structure, the evaluation score of the target cardiac section is calculated, including: For the target heart structure whose structure type is left ventricle, right ventricle, left ventricular outflow tract, aorta, spine, and interventricular septum, select the highest confidence level from the corresponding confidence level as the corresponding target confidence level ; Based on the sum of the products of the target confidence levels corresponding to the left ventricle, the right ventricle, the left ventricular outflow tract, the aorta, the spine, and the interventricular septum and a preset weight, the evaluation score of the left ventricular outflow tract section is obtained. 5. The method according to any one of claims 1 to 4, wherein the target information further comprises: position information of each target cardiac structure in the cardiac image to be detected; The method also includes: When the target cardiac section is a standard section, extract the target cardiac structure image corresponding to each target cardiac structure from the to-be-detected cardiac image according to the position information of each target cardiac structure; Each of the target cardiac structure images is segmented to obtain contour information of the target cardiac structure in each of the target cardiac structure images. 6. The method according to claim 5, wherein, segmenting each of the target cardiac structure images to obtain contour information of the target cardiac structure in each of the target cardiac structure images, comprising: For any of the target cardiac structure images, use a corresponding trained image segmentation model to segment the target cardiac structure images to obtain contour information of the target cardiac structure in each of the target cardiac structure images; The method for obtaining the image segmentation model by training includes: Obtain a cardiac structure sample image and corresponding annotation information, where the annotation information includes: outline annotation information and a structure annotation type of the cardiac structure in the cardiac structure sample image; Input the cardiac structure sample image into the image segmentation model to be trained, and obtain the segmentation information corresponding to the cardiac structure sample image, and the segmentation information includes: the contour segmentation information of the cardiac structure in the cardiac structure sample image, the structure segmentation type and the Corresponding segmentation confidence; Based on the segmentation information and the labeling information, the parameters of the image segmentation model to be trained are adjusted until the model training end condition is satisfied, and a trained image segmentation model is obtained. 7. The method according to claim 5, wherein the method further comprises: Each of the target cardiac structures is measured based on the contour information of each of the target cardiac structures. 8. A cardiac section detection system, wherein the system comprises: an image acquisition module for acquiring the heart image to be detected; a target detection module, configured to perform target detection on the to-be-detected cardiac image to obtain corresponding target information, where the target information includes: the structural type and corresponding confidence level of each target cardiac structure in the to-be-detected cardiac image, and The slice type of the target cardiac slice corresponding to the cardiac image to be detected; A standard slice determination module, configured to determine whether the target cardiac slice is a standard slice according to the target information. 9. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein the processor implements the method according to any one of claims 1 to 7 when the processor executes the computer program. step. 10. A computer-readable storage medium on which a computer program is stored, characterized in that, when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images