Automatic intussusception diagnosis method and device based on deep learning

Abstract

The invention discloses an automatic intussusception diagnosis method and device based on deep learning. The method comprises the following steps: constructing an initial intussusception diagnosis model; obtaining a data set, and training the initial intussusception diagnosis model based on the data set to obtain a target intussusception diagnosis model; acquiring an image to be diagnosed; and inputting a to-be-diagnosed image into the target intussusception diagnosis model to obtain an intussusception classification result corresponding to the to-be-diagnosed image. According to the method, various different images are analyzed based on the target intussusception diagnosis model to obtain the corresponding intussusception classification result, so that the application range is wide, and the robustness of the model and the accuracy of the classification result are improved.

Description

Technical Field

[0001] The present invention relates to the fields of medical artificial intelligence, deep learning and computer vision technology, and in particular to a method and device for automatic diagnosis of intussusception based on deep learning. Background Art

[0002] Intussusception is a common abdominal emergency in children. It is important to identify the specific type of intussusception and determine its treatment.

[0003] In the prior art, deep learning can be used to automatically classify and diagnose medical images through ultrasound or X-ray imaging. In the prior art, deep learning is only used to diagnose the type of intussusception based on a certain imaging examination, but the examinations that patients receive may be different in clinical practice, which limits the applicable scenarios of the prior art and reduces its robustness. Summary of the invention

[0004] The present invention aims to solve one of the technical problems in the related art at least to a certain extent.

[0005] To this end, an object of the present invention is to propose an automatic diagnosis method for intussusception based on deep learning, which can analyze a variety of different images based on a target intussusception diagnosis model to obtain corresponding intussusception classification results, thereby having a wide range of applications and improving the robustness of the model and the accuracy of the classification results.

[0006] Another object of the present invention is to provide an automatic diagnosis device for intussusception based on deep learning.

[0007] To achieve the above object, an embodiment of the present invention provides an automatic diagnosis method for intussusception based on deep learning, comprising:

[0008] Construct an initial intussusception diagnostic model;

[0009] Acquiring a data set, and training the initial intussusception diagnostic model based on the data set to obtain a target intussusception diagnostic model;

[0010] Acquiring images to be diagnosed;

[0011] The image to be diagnosed is input into the target intussusception diagnosis model to obtain an intussusception classification result corresponding to the image to be diagnosed.

[0012] The deep learning-based automatic diagnosis method for intussusception in the embodiment of the present invention may also have the following additional technical features:

[0013] Furthermore, the initial intussusception diagnostic model includes a first encoder, a second encoder, a third encoder, a dynamic weighting module and a classifier, wherein the first encoder, the second encoder and the third encoder correspond to different input image types.

[0014] Furthermore, the initial intussusception diagnostic model is trained based on the data set to obtain a target intussusception diagnostic model, including:

[0015] Classifying different types of images of the same case into the same data group, and marking the intussusception type of the data group, to obtain a training data set and a test data set for training the initial intussusception diagnosis model;

[0016] Using the training data set to train the initial intussusception diagnosis model to obtain a trained intussusception diagnosis model;

[0017] The trained intussusception diagnosis model is tested and optimized using the test data set to obtain a target intussusception diagnosis model.

[0018] Furthermore, the initial intussusception diagnosis model is trained using the training data set to obtain a trained intussusception diagnosis model, including:

[0019] Inputting the data components in the training data set into the initial intussusception diagnosis model in batches to obtain a first output result, a second output result, a third output result and a predicted intussusception classification result for each batch of data;

[0020] Based on the first output result, the second output result and the third output result of each batch of data, determining a first loss function value corresponding to each batch of data by using a first loss function;

[0021] Based on the predicted intussusception classification results and the actual intussusception classification results of each batch of data, determining a second loss function value corresponding to each batch of data by a second loss function;

[0022] The first parameters in the first encoder, the second encoder, and the third encoder are updated based on the first loss function value, and the second parameters in the dynamic weighting module and the classifier are updated based on the second loss function value. The above training process is repeated until the number of iterations of the first parameter and the second parameter reaches a preset number or the initial intussusception diagnostic model converges, thereby obtaining a trained intussusception diagnostic model.

[0023] Furthermore, the dynamic weighting module includes a fully connected layer and a Softmax layer, and the classifier includes a multilayer perceptron; the data components in the training data set are input into the initial intussusception diagnosis model in batches to obtain a first output result, a second output result, a third output result and a predicted intussusception classification result of each batch of data, including:

[0024] Inputting different types of images of the same case in the data group into the corresponding first encoder, second encoder and third encoder respectively, obtaining first feature vectors, second feature vectors and third feature vectors corresponding to the different types of images, and confirming the first feature vector, the second feature vector and the third feature vector as the first output result, the second output result and the third output result respectively, wherein the dimensions of the first feature vector, the second feature vector and the third feature vector are the same;

[0025] Randomly masking the first eigenvector, the second eigenvector and the third eigenvector to obtain a fourth eigenvector, a fifth eigenvector and a sixth eigenvector;

[0026] The dynamic weighting module concatenates the fourth eigenvector, the fifth eigenvector and the sixth eigenvector to obtain a seventh eigenvector, passes the seventh eigenvector through the fully connected layer and the Softmax layer to obtain a weight vector, and performs weighted summation of the fourth eigenvector, the fifth eigenvector and the sixth eigenvector based on the weight vector to obtain an eighth eigenvector;

[0027] The classifier passes the eighth feature vector through the multi-layer perceptron to obtain a predicted intussusception classification result.

[0028] Further, the randomly masking the first eigenvector, the second eigenvector and the third eigenvector to obtain a fourth eigenvector, a fifth eigenvector and a sixth eigenvector includes:

[0029] Determine a first target feature vector that needs to be masked among the first feature vector, the second feature vector, and the third feature vector by using a first probability;

[0030] Replacing the first target feature vector with the corresponding sliding average to obtain a first mask feature vector;

[0031] Based on the masked feature vectors and the remaining feature vectors that do not need to be masked, a fourth feature vector, a fifth feature vector and a sixth feature vector are obtained.

[0032] Further, the randomly masking the first eigenvector, the second eigenvector and the third eigenvector to obtain a fourth eigenvector, a fifth eigenvector and a sixth eigenvector includes:

[0033] Determine a second target feature vector and a third target feature vector that need to be masked among the first feature vector, the second feature vector, and the third feature vector by using a second probability;

[0034] The second target feature vector and the third target feature vector are replaced by corresponding sliding averages to obtain a second mask feature vector and a third mask feature vector;

[0035] A fourth eigenvector, a fifth eigenvector and a sixth eigenvector are obtained based on the second mask eigenvector, the third mask eigenvector and the remaining eigenvectors that do not need to be masked.

[0036] To achieve the above object, another embodiment of the present invention provides an automatic diagnosis device for intussusception based on deep learning, the device comprising:

[0037] A building module for constructing an initial intussusception diagnosis model;

[0038] A training module, used for acquiring a data set, and training the initial intussusception diagnostic model based on the data set to obtain a target intussusception diagnostic model;

[0039] An acquisition module, used for acquiring an image to be diagnosed;

[0040] The diagnosis module is used to input the image to be diagnosed into the target intussusception diagnosis model to obtain the intussusception classification result corresponding to the image to be diagnosed.

[0041] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0043] Figure 1 is a flow chart of an automatic diagnosis method for intussusception based on deep learning according to an embodiment of the present invention;

[0044] Figure 2 Schematic diagram of the structure of an automatic diagnosis device for intussusception based on deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0046] The following describes the automatic diagnosis method and device for intussusception based on deep learning proposed according to an embodiment of the present invention with reference to the accompanying drawings.

[0047] First, an automatic diagnosis method for intussusception based on deep learning proposed according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0048] Figure 1 Flow chart of an automatic diagnosis method for intussusception based on deep learning according to an embodiment of the present invention.

[0049] like Figure 1 As shown, the automatic diagnosis method for intussusception based on deep learning includes the following steps:

[0050] Step S1, constructing an initial intussusception diagnosis model;

[0051] In one embodiment of the present invention, the initial intussusception diagnostic model may include a first encoder, a second encoder, a third encoder, a dynamic weighting module and a classifier. In one embodiment of the present invention, the first encoder, the second encoder and the third encoder correspond to different input image types, and the first encoder, the second encoder and the third encoder may all be convolutional neural networks based on ResNet-50. In one embodiment of the present invention, the dynamic weighting module may include a linear fully connected layer and a Softmax layer, and the classifier may include a multi-layer perceptron.

[0052] For example, in one embodiment of the present invention, the first encoder is an ultrasound image encoder UE, and the corresponding input image type is an ultrasound image, the second encoder is an X-ray picture encoder XE, and the corresponding input image type is an X-ray picture, and the third encoder is a local X-ray picture encoder PE, and the corresponding input image type is a local X-ray picture.

[0053] Step S2, acquiring a data set, and training an initial intussusception diagnosis model based on the data set to obtain a target intussusception diagnosis model;

[0054] In one embodiment of the present invention, an initial intussusception diagnostic model is obtained through the above steps, a data set can be acquired, and the initial intussusception diagnostic model is trained based on the data set to obtain a target intussusception diagnostic model.

[0055] Specifically, in one embodiment of the present invention, the method for training the initial intussusception diagnosis model based on the data set to obtain the target intussusception diagnosis model may include the following steps:

[0056] Step S21, grouping different types of images of the same case into the same data group, and labeling the intussusception type of the data group, to obtain a training data set and a test data set for training an initial intussusception diagnosis model;

[0057] Step S22, training the initial intussusception diagnosis model using the training data set to obtain a trained intussusception diagnosis model;

[0058] Step S23, using the test data set to test and tune the trained intussusception diagnosis model to obtain a target intussusception diagnosis model.

[0059] In one embodiment of the present invention, after the data set is obtained, different types of images of the same case in the data set can be divided into the same data group, and the intussusception type of the data group can be labeled to obtain the labeled data: Where N is the number of data sets, U i represents the ultrasound image of the ith case with signs of intussusception obtained by ultrasound examination, X i represents the X-ray image of the i-th case, P i represents the local image of the intussusception lesion area in the X-ray image of the i-th case, l i represents the intussusception type of the i-th case, wherein the intussusception type may include the back-knot type, the back-knot type, the knot type, and the small intestinal intussusception type.

[0060] And, in one embodiment of the present invention, after the above data annotation is performed on the images in the data set, the annotated data can be divided into a training data set and a test data set for training the initial intussusception diagnosis model according to a preset ratio. For example, assuming that the preset ratio is 7:3, that is, 70% of the annotated data is randomly divided into a training data set and 30% of the data is divided into a test data set.

[0061] Further, in one embodiment of the present invention, the method of training the initial intussusception diagnostic model using the training data set to obtain the trained intussusception diagnostic model may include the following steps:

[0062] Step 1, inputting the data components in the training data set into the initial intussusception diagnosis model in batches, and obtaining the first output result, the second output result, the third output result and the predicted intussusception classification result of each batch of data;

[0063] Step 2: Based on the first output result, the second output result and the third output result of each batch of data, determine the first loss function value corresponding to each batch of data through the first loss function;

[0064] Step 3, based on the predicted intussusception classification results and the actual intussusception classification results of each batch of data, determining the second loss function value corresponding to each batch of data through the second loss function;

[0065] Step 4: Update the first parameter in the first encoder, the second encoder, and the third encoder based on the first loss function value, and update the second parameter in the dynamic weighting module and the classifier based on the second loss function value, and repeat the above training process until the number of iterations of the first parameter and the second parameter reaches a preset number or the initial intussusception diagnostic model converges, thereby obtaining a trained intussusception diagnostic model.

[0066] In one embodiment of the present invention, the method of inputting the data components in the training data set into the initial intussusception diagnosis model in batches to obtain the first output result, the second output result, the third output result and the predicted intussusception classification result of each batch of data may include the following steps:

[0067] Step a, inputting different types of images of the same case in the data group into the corresponding first encoder, second encoder and third encoder respectively, obtaining first feature vectors, second feature vectors and third feature vectors corresponding to the different types of images, and confirming the first feature vector, second feature vector and third feature vector as the first output result, second output result and third output result respectively, wherein the dimensions of the first feature vector, the second feature vector and the third feature vector are the same;

[0068] Step b, randomly masking the first eigenvector, the second eigenvector and the third eigenvector to obtain a fourth eigenvector, a fifth eigenvector and a sixth eigenvector;

[0069] Step c, the dynamic weighting module concatenates the fourth eigenvector, the fifth eigenvector and the sixth eigenvector to obtain a seventh eigenvector, passes the seventh eigenvector through a fully connected layer and a Softmax layer to obtain a weight vector, and performs weighted summation on the fourth eigenvector, the fifth eigenvector and the sixth eigenvector based on the weight vector to obtain an eighth eigenvector;

[0070] Step d: The classifier passes the eighth eigenvector through a multi-layer perceptron to obtain a predicted intussusception classification result.

[0071] Among them, in one embodiment of the present invention, after obtaining the first output result, the second output result and the third output result through the above steps, any two combinations of ultrasound images, X-ray images and local X-ray images can be compared and studied.

[0072] For example, in one embodiment of the present invention, it is assumed that the ultrasound image and the X-ray image of the same case are The image is passed through the corresponding encoder to obtain the first eigenvector u i =UE(U i ) and the second eigenvector x i =XE(X i ), where u i and x i are all 2048-dimensional feature vectors. And, comparative learning can be performed through the first loss function, constraining the paired ultrasound image features to be similar to the X-ray image features, and the unpaired features to be dissimilar, where the first loss function is L contrast =L(U,X)+L(U,P)+L(P,X), Where exp() is an exponential function with a natural constant as the base, and cos() is the cosine value of the angle between two vectors. The calculation formulas of L(U, P) and L(P, X) are similar to those of L(U, X) above, and the embodiments of the present invention are not described in detail here.

[0073] Furthermore, in one embodiment of the present invention, the first loss function can allow different encoders to learn aligned features for different images, so as to improve the fusion and utilization effect of different image features during intussusception classification.

[0074] Among them, in one embodiment of the present invention, in order to improve the robustness of the model to missing inputs, when inputting the feature vectors of three images, the random mask randomly masks one of the features with a probability of p1, and in the case of the remaining (1-p1) probability, randomly masks two of the features with a probability of p2, where p1 and p2 are two hyperparameters. And, in one embodiment of the present invention, the method of masking the feature vector may include: recording a corresponding sliding average for the feature vector of each image in the same batch, and using the corresponding sliding average instead of the original value to mask a certain feature.

[0075] Based on this, in one embodiment of the present invention, the method of randomly masking the first eigenvector, the second eigenvector and the third eigenvector to obtain the fourth eigenvector, the fifth eigenvector and the sixth eigenvector may include the following steps:

[0076] Step 11: determining a first target feature vector that needs to be masked among the first feature vector, the second feature vector, and the third feature vector by using the first probability;

[0077] Step 12: Replace the first target feature vector with the corresponding sliding average to obtain a first mask feature vector;

[0078] Step 13: Based on the masked feature vectors and the remaining feature vectors that do not need to be masked, obtain the fourth feature vector, the fifth feature vector and the sixth feature vector.

[0079] And, in another embodiment of the present invention, the method of randomly masking the first eigenvector, the second eigenvector and the third eigenvector to obtain the fourth eigenvector, the fifth eigenvector and the sixth eigenvector may include the following steps:

[0080] Step 1: Determine the second target feature vector and the third target feature vector that need to be masked among the first feature vector, the second feature vector and the third feature vector by using the second probability;

[0081] Step 2: Replace the second target feature vector and the third target feature vector by corresponding sliding averages to obtain a second mask feature vector and a third mask feature vector;

[0082] Step 3: Based on the second mask feature vector, the third mask feature vector and the remaining feature vectors that do not need to be masked, obtain the fourth feature vector, the fifth feature vector and the sixth feature vector.

[0083] In one embodiment of the present invention, the first target feature vector may be any one of the first feature vector, the second feature vector and the third feature vector, and the second target feature vector and the third target feature vector may be any two of the first feature vector, the second feature vector and the third feature vector.

[0084] And, in one embodiment of the present invention, after obtaining the fourth eigenvector, the fifth eigenvector and the sixth eigenvector after random masking through the above steps, the fourth eigenvector, the fifth eigenvector and the sixth eigenvector can be concatenated to obtain the seventh eigenvector, and the seventh eigenvector is passed through the fully connected layer and the Softmax layer of the dynamic weighting module to obtain a three-dimensional weight vector w, and then w is used to perform weighted summation on the fourth eigenvector, the fifth eigenvector and the sixth eigenvector to obtain the eighth eigenvector, wherein the eighth eigenvector has the same dimension as the fourth eigenvector.

[0085] Furthermore, in one embodiment of the present invention, after obtaining the eighth eigenvector through the above steps, the eighth eigenvector can be passed through a multi-layer perceptron in a classifier to obtain probability vectors of each category to which the image belongs, and the category with the largest probability vector value is determined as the predicted intussusception classification result.

[0086] Furthermore, in one embodiment of the present invention, after obtaining the predicted intussusception classification results through the above steps, the second loss function value corresponding to each batch of data can be determined through a second loss function based on the predicted intussusception classification results and the actual intussusception classification results of each batch of data, wherein the second loss function can be a cross entropy loss function.

[0087] Furthermore, in one embodiment of the present invention, after obtaining the first loss function value and the second loss function value through the above steps, the first parameter in the first encoder, the second encoder, and the third encoder can be updated based on the first loss function value, and the second parameter in the dynamic weighting module and the classifier can be updated based on the second loss function value. The above training process is repeated until the number of iterations of the first parameter and the second parameter reaches a preset number or the initial intussusception diagnostic model converges, thereby obtaining a trained intussusception diagnostic model.

[0088] Among them, in one embodiment of the present invention, after the trained intussusception diagnosis model is obtained through the above steps, the trained intussusception diagnosis model can be tested and tuned using a test data set to obtain the accuracy performance of the trained intussusception diagnosis model on new samples and evaluate the generalization ability of the model. Then, according to the evaluation results of the test data set, the trained intussusception diagnosis model is adjusted and optimized, for example, the network structure or hyperparameter settings are changed to improve the accuracy, robustness and performance of the model.

[0089] Step S3, obtaining an image to be diagnosed;

[0090] In one embodiment of the present invention, the image to be diagnosed may be an ultrasound image, an X-ray image, a partial X-ray image, or any non-empty subset thereof.

[0091] For example, in one embodiment of the present invention, when only an ultrasound image of the patient is available, only the ultrasound image may be input; when only an X-ray image of the patient is available, only the X-ray image may be input; when an X-ray image is available and the location of intussusception can be determined therein, both the X-ray image and the local X-ray image may be input.

[0092] Step S4, inputting the image to be diagnosed into the target intussusception diagnosis model to obtain the intussusception classification result corresponding to the image to be diagnosed.

[0093] In one embodiment of the present invention, after the image to be diagnosed is obtained through the above steps, the image to be diagnosed can be input into the target intussusception diagnosis model to obtain the intussusception classification result corresponding to the image to be diagnosed.

[0094] According to the deep learning-based automatic diagnosis method for intussusception proposed in an embodiment of the present invention, a variety of different images can be analyzed based on the target intussusception diagnosis model to obtain corresponding intussusception classification results, thereby having a wide range of applications and improving the robustness of the model and the accuracy of the classification results.

[0095] Next, an automatic diagnosis device for intussusception based on deep learning proposed in accordance with an embodiment of the present invention is described with reference to the accompanying drawings.

[0096] Figure 2 Schematic diagram of the structure of an automatic diagnosis device for intussusception based on deep learning according to an embodiment of the present invention.

[0097] like Figure 2 As shown, the deep learning-based intussusception automatic diagnosis device 10 includes: a construction module 201, a training module 202, an acquisition module 203 and a diagnosis module 204, wherein

[0098] A construction module 201 is used to construct an initial intussusception diagnosis model;

[0099] A training module 202 is used to obtain a data set, and train an initial intussusception diagnosis model based on the data set to obtain a target intussusception diagnosis model;

[0100] An acquisition module 203 is used to acquire an image to be diagnosed;

[0101] The diagnosis module 204 is used to input the image to be diagnosed into the target intussusception diagnosis model to obtain the intussusception classification result corresponding to the image to be diagnosed.

[0102] Furthermore, the initial intussusception diagnosis model includes a first encoder, a second encoder, a third encoder, a dynamic weighting module and a classifier, wherein the first encoder, the second encoder and the third encoder correspond to different input image types.

[0103] Furthermore, the training module 202 is specifically used for:

[0104] Different types of images of the same case are divided into the same data group, and the intussusception types of the data group are labeled to obtain a training data set and a test data set for training an initial intussusception diagnosis model;

[0105] The initial intussusception diagnosis model is trained using the training data set to obtain a trained intussusception diagnosis model;

[0106] The trained intussusception diagnosis model was tested and optimized using the test data set to obtain the target intussusception diagnosis model.

[0107] Furthermore, the training module 202 is specifically used for:

[0108] Inputting the data components in the training data set into the initial intussusception diagnosis model in batches to obtain a first output result, a second output result, a third output result and a predicted intussusception classification result for each batch of data;

[0109] Based on the first output result, the second output result and the third output result of each batch of data, determining the first loss function value corresponding to each batch of data by using the first loss function;

[0110] Based on the predicted intussusception classification results and the actual intussusception classification results of each batch of data, determining the second loss function value corresponding to each batch of data through the second loss function;

[0111] The first parameter in the first encoder, the second encoder, and the third encoder is updated based on the first loss function value, and the second parameter in the dynamic weighting module and the classifier is updated based on the second loss function value. The above training process is repeated until the number of iterations of the first parameter and the second parameter reaches a preset number or the initial intussusception diagnostic model converges, thereby obtaining a trained intussusception diagnostic model.

[0112] Further, the dynamic weighting module includes a fully connected layer and a Softmax layer, and the classifier includes a multi-layer perceptron; the training module 202 is specifically used to: input different types of images of the same case in the data group into the corresponding first encoder, second encoder and third encoder, respectively, to obtain the first feature vector, second feature vector and third feature vector corresponding to the different types of images, and confirm the first feature vector, second feature vector and third feature vector as the first output result, second output result and third output result, respectively, wherein the dimensions of the first feature vector, the second feature vector and the third feature vector are the same;

[0113] The first eigenvector, the second eigenvector and the third eigenvector are randomly masked to obtain a fourth eigenvector, a fifth eigenvector and a sixth eigenvector;

[0114] The dynamic weighting module concatenates the fourth eigenvector, the fifth eigenvector and the sixth eigenvector to obtain a seventh eigenvector, passes the seventh eigenvector through a fully connected layer and a Softmax layer to obtain a weight vector, and performs weighted summation of the fourth eigenvector, the fifth eigenvector and the sixth eigenvector based on the weight vector to obtain an eighth eigenvector;

[0115] The classifier passes the eighth eigenvector through a multi-layer perceptron to obtain the predicted intussusception classification result.

[0116] Furthermore, the training module 202 is also used for:

[0117] Determine a first target feature vector that needs to be masked among the first feature vector, the second feature vector, and the third feature vector by using the first probability;

[0118] Replace the first target feature vector with the corresponding sliding average to obtain a first mask feature vector;

[0119] Based on the masked eigenvectors and the remaining eigenvectors that do not need to be masked, a fourth eigenvector, a fifth eigenvector and a sixth eigenvector are obtained.

[0120] Furthermore, the training module 202 is also used for:

[0121] Determine a second target feature vector and a third target feature vector that need to be masked among the first feature vector, the second feature vector, and the third feature vector by using a second probability;

[0122] The second target feature vector and the third target feature vector are replaced by corresponding sliding averages to obtain a second mask feature vector and a third mask feature vector;

[0123] Based on the second mask feature vector, the third mask feature vector and the remaining feature vectors that do not need to be masked, a fourth feature vector, a fifth feature vector and a sixth feature vector are obtained.

[0124] The deep learning-based automatic intussusception diagnosis device proposed in an embodiment of the present invention can analyze a variety of different images based on a target intussusception diagnosis model to obtain corresponding intussusception classification results, thereby having a wide range of applications and improving the robustness of the model and the accuracy of the classification results.

[0125] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0126] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0127] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for automatic diagnosis of intussusception based on deep learning, characterized in that: The method comprises: Construct an initial intussusception diagnostic model; Acquiring a data set, and training the initial intussusception diagnostic model based on the data set to obtain a target intussusception diagnostic model; Acquiring images to be diagnosed; The image to be diagnosed is input into the target intussusception diagnosis model to obtain an intussusception classification result corresponding to the image to be diagnosed.

2. The method according to claim 1, characterized in that The initial intussusception diagnostic model includes a first encoder, a second encoder, a third encoder, a dynamic weighting module and a classifier, wherein the first encoder, the second encoder and the third encoder correspond to different input image types.

3. The method according to claim 2, characterized in that The initial intussusception diagnosis model is trained based on the data set to obtain a target intussusception diagnosis model, including: Classifying different types of images of the same case into the same data group, and marking the intussusception type of the data group, to obtain a training data set and a test data set for training the initial intussusception diagnosis model; Using the training data set to train the initial intussusception diagnosis model to obtain a trained intussusception diagnosis model; The trained intussusception diagnosis model is tested and optimized using the test data set to obtain a target intussusception diagnosis model.

4. The method according to claim 3, characterized in that The method of training the initial intussusception diagnosis model using the training data set to obtain a trained intussusception diagnosis model includes: Inputting the data components in the training data set into the initial intussusception diagnosis model in batches to obtain a first output result, a second output result, a third output result and a predicted intussusception classification result for each batch of data; Based on the first output result, the second output result and the third output result of each batch of data, determining a first loss function value corresponding to each batch of data by using a first loss function; Based on the predicted intussusception classification results and the actual intussusception classification results of each batch of data, determining a second loss function value corresponding to each batch of data by a second loss function; The first parameters in the first encoder, the second encoder, and the third encoder are updated based on the first loss function value, and the second parameters in the dynamic weighting module and the classifier are updated based on the second loss function value. The above training process is repeated until the number of iterations of the first parameter and the second parameter reaches a preset number or the initial intussusception diagnostic model converges, thereby obtaining a trained intussusception diagnostic model.

5. The method according to claim 4, characterized in that The dynamic weighting module includes a fully connected layer and a Softmax layer, and the classifier includes a multi-layer perceptron; the data components in the training data set are input into the initial intussusception diagnosis model in batches to obtain the first output result, the second output result, the third output result and the predicted intussusception classification result of each batch of data, including: Inputting different types of images of the same case in the data group into the corresponding first encoder, second encoder and third encoder respectively, obtaining first feature vectors, second feature vectors and third feature vectors corresponding to the different types of images, and confirming the first feature vector, the second feature vector and the third feature vector as the first output result, the second output result and the third output result respectively, wherein the dimensions of the first feature vector, the second feature vector and the third feature vector are the same; Randomly masking the first eigenvector, the second eigenvector and the third eigenvector to obtain a fourth eigenvector, a fifth eigenvector and a sixth eigenvector; The dynamic weighting module concatenates the fourth eigenvector, the fifth eigenvector and the sixth eigenvector to obtain a seventh eigenvector, passes the seventh eigenvector through the fully connected layer and the Softmax layer to obtain a weight vector, and performs weighted summation of the fourth eigenvector, the fifth eigenvector and the sixth eigenvector based on the weight vector to obtain an eighth eigenvector; The classifier passes the eighth feature vector through the multi-layer perceptron to obtain a predicted intussusception classification result.

6. The method according to claim 5, characterized in that The step of randomly masking the first eigenvector, the second eigenvector, and the third eigenvector to obtain a fourth eigenvector, a fifth eigenvector, and a sixth eigenvector includes: Determine a first target feature vector that needs to be masked among the first feature vector, the second feature vector, and the third feature vector by using a first probability; Replacing the first target feature vector with the corresponding sliding average to obtain a first mask feature vector; Based on the masked feature vectors and the remaining feature vectors that do not need to be masked, a fourth feature vector, a fifth feature vector and a sixth feature vector are obtained.

7. The method according to claim 5, characterized in that The step of randomly masking the first eigenvector, the second eigenvector, and the third eigenvector to obtain a fourth eigenvector, a fifth eigenvector, and a sixth eigenvector includes: Determine a second target feature vector and a third target feature vector that need to be masked among the first feature vector, the second feature vector, and the third feature vector by using a second probability; The second target feature vector and the third target feature vector are replaced by corresponding sliding averages to obtain a second mask feature vector and a third mask feature vector; A fourth eigenvector, a fifth eigenvector and a sixth eigenvector are obtained based on the second mask eigenvector, the third mask eigenvector and the remaining eigenvectors that do not need to be masked.

8. A method for automatic diagnosis of intussusception based on deep learning, characterized in that: The method comprises: A building module for constructing an initial intussusception diagnosis model; A training module, used for acquiring a data set, and training the initial intussusception diagnostic model based on the data set to obtain a target intussusception diagnostic model; An acquisition module, used for acquiring an image to be diagnosed; The diagnosis module is used to input the image to be diagnosed into the target intussusception diagnosis model to obtain the intussusception classification result corresponding to the image to be diagnosed.

9. An electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images