Intussusception automatic diagnosis method and device based on deep learning

Abstract

The application discloses an intestinal intussusception automatic diagnosis method and device based on deep learning, and relates to the technical field of medical image processing, and aims at providing an intestinal intussusception automatic diagnosis method and device based on deep learning.

Description

Technical Field

[0001] This invention relates to the fields of medical artificial intelligence, deep learning, and computer vision, and in particular to an automatic diagnostic method and apparatus for intussusception based on deep learning. Background Technology

[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 existing technologies, deep learning can be used to automatically classify and diagnose medical images through ultrasound or X-ray examinations. However, these existing technologies rely solely on a specific imaging examination for diagnosing intussusception, which limits the applicability and robustness of these technologies due to the diverse range of clinical examinations patients undergo. Summary of the Invention

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

[0005] Therefore, one objective of this invention is to propose an automatic diagnosis method for intussusception based on deep learning. This method can analyze various images based on a target intussusception diagnosis model to obtain corresponding intussusception classification results, thus having a wide range of applications and improving the robustness of the model and the accuracy of the classification results.

[0006] Another objective of this invention is to propose an automated diagnostic device for intussusception based on deep learning.

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

[0008] Construct an initial diagnostic model for intussusception;

[0009] Obtain the dataset and train the initial intussusception diagnostic model based on the dataset to obtain the target intussusception diagnostic model;

[0010] Acquire the image to be diagnosed;

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

[0012] The deep learning-based automatic diagnosis method for intussusception according to embodiments 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] Further, the step of training the initial intussusception diagnostic model based on the dataset to obtain the target intussusception diagnostic model includes:

[0015] Different types of images from the same case are divided into the same data group, and the intussusception type of the data group is labeled to obtain the training dataset and test dataset for training the initial intussusception diagnostic model.

[0016] The initial intussusception diagnostic model is trained using the training dataset to obtain a trained intussusception diagnostic model.

[0017] The trained intussusception diagnostic model was tested and optimized using the test dataset to obtain the target intussusception diagnostic model.

[0018] Further, the step of training the initial intussusception diagnostic model using the training dataset to obtain the trained intussusception diagnostic model includes:

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

[0020] Based on the first output result, second output result and third output result of each batch of data, the value of the first loss function corresponding to each batch of data is determined by the first loss function.

[0021] Based on the predicted intussusception classification result and the actual intussusception classification result of each batch of data, the value of the second loss function corresponding to each batch of data is determined by the second loss function.

[0022] The first parameters in the first encoder, second encoder, and third encoder are updated based on the first loss function value, and the second parameters in the dynamic weighting module and 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, and the trained intussusception diagnostic model is obtained.

[0023] Further, the dynamic weighting module includes a fully connected layer and a Softmax layer, and the classifier includes a multilayer perceptron; the step of inputting data components from the training dataset into the initial intussusception diagnosis model in batches to obtain the first output result, second output result, third output result, and predicted intussusception classification result for each batch of data includes:

[0024] Different types of images of the same case in the data set are input 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 the first feature vector, second feature vector and third feature vector are respectively confirmed as the first output result, second output result and third output result, wherein the first feature vector, second feature vector and third feature vector have the same dimension;

[0025] The first feature vector, the second feature vector, and the third feature vector are randomly masked to obtain the fourth feature vector, the fifth feature vector, and the sixth feature vector;

[0026] The dynamic weighting module concatenates the fourth, fifth, and sixth feature vectors to obtain the seventh feature vector. The seventh feature vector is then passed through the fully connected layer and the Softmax layer to obtain a weight vector. Based on the weight vector, the fourth, fifth, and sixth feature vectors are weighted and summed to obtain the eighth feature vector.

[0027] The classifier passes the eighth feature vector through the multilayer perceptron to obtain the predicted intussusception classification result.

[0028] Further, the step of randomly masking the first feature vector, the second feature vector, and the third feature vector to obtain the fourth feature vector, the fifth feature vector, and the sixth feature vector includes:

[0029] The first target feature vector that needs to be masked is determined from the first feature vector, the second feature vector, and the third feature vector using a first probability.

[0030] The first target feature vector is replaced by the corresponding moving average to obtain the first mask feature vector;

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

[0032] Further, the step of randomly masking the first feature vector, the second feature vector, and the third feature vector to obtain the fourth feature vector, the fifth feature vector, and the sixth feature vector includes:

[0033] The second target feature vector and the third target feature vector that need to be masked are determined from the first feature vector, the second feature vector and the third feature vector using the second probability.

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

[0035] Based on the second mask feature vector, the third mask feature vector, and the remaining feature vectors that do not require masking, the fourth feature vector, the fifth feature vector, and the sixth feature vector are obtained.

[0036] To achieve the above objectives, another embodiment of the present invention proposes an automated diagnostic device for intussusception based on deep learning, the device comprising:

[0037] The building block is used to construct the initial diagnostic model for intussusception;

[0038] The training module is used to acquire a dataset and train the initial intussusception diagnostic model based on the dataset to obtain the target intussusception diagnostic model.

[0039] The acquisition module is used to acquire the image to be diagnosed.

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

[0041] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

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

[0043] Figure 1 A flowchart of a deep learning-based automatic diagnosis method for intussusception according to an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of a deep learning-based automatic diagnostic device for intussusception according to an embodiment of the present invention. Detailed Implementation

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

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

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

[0048] Figure 1 This is a flowchart of a deep learning-based automatic diagnosis method for intussusception according to an embodiment of the present invention.

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

[0050] Step S1: Construct an initial diagnostic model for intussusception;

[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, second encoder, and third encoder correspond to different input image types, and all three encoders can be ResNet-50-based convolutional neural networks. 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 multilayer perceptron.

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

[0053] Step S2: Obtain the dataset and train the initial intussusception diagnostic model based on the dataset to obtain the target intussusception diagnostic model.

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

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

[0056] Step S21: Different types of images from the same case are divided into the same data group, and the intussusception type of the data group is labeled to obtain the training dataset and test dataset for training the initial intussusception diagnostic model.

[0057] Step S22: Train the initial intussusception diagnostic model using the training dataset to obtain the trained intussusception diagnostic model.

[0058] Step S23: Use the test dataset to test and optimize the trained intussusception diagnostic model to obtain the target intussusception diagnostic model.

[0059] In one embodiment of the present invention, after obtaining the dataset, different types of images of the same case in the dataset can be divided into the same data group, and the intussusception type of the data group can be labeled to obtain labeled data. Where N is the number of data sets, U i X represents the ultrasound image containing signs of intussusception obtained from the ultrasound examination of the i-th case. i Let P represent the X-ray image of the i-th case. i This represents a partial image of the intussusception lesion area in the X-ray image of the i-th case. i Indicates the type of intussusception in the i-th case, where the type of intussusception may include ileocolonic, ileocolonic, teloid, and small bowel intussusception.

[0060] Furthermore, in one embodiment of the present invention, after the images in the dataset are annotated as described above, the annotated data can be divided into a training dataset and a test dataset for training the initial intussusception diagnostic model according to a preset ratio. For example, assuming the preset ratio is 7:3, that is, 70% of the annotated data is randomly divided into a training dataset and 30% into a test dataset.

[0061] Furthermore, in one embodiment of the present invention, the method for training the initial intussusception diagnostic model using a training dataset to obtain a trained intussusception diagnostic model may include the following steps:

[0062] Step 1: Input the data groups in the training dataset into the initial intussusception diagnosis model in batches to obtain the first output result, second output result, third output result and predicted intussusception classification result for each batch of data;

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

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

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

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

[0067] Step a: Input different types of images of the same case in the data set 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 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 first feature vector, second feature vector and third feature vector have the same dimension;

[0068] Step b: Randomly mask the first feature vector, the second feature vector, and the third feature vector to obtain the fourth feature vector, the fifth feature vector, and the sixth feature vector;

[0069] Step c: The dynamic weighting module concatenates the fourth, fifth, and sixth feature vectors to obtain the seventh feature vector. The seventh feature vector is then passed through a fully connected layer and a Softmax layer to obtain a weight vector. Based on the weight vector, the fourth, fifth, and sixth feature vectors are weighted and summed to obtain the eighth feature vector.

[0070] Step d: The classifier passes the eighth feature vector through a multilayer perceptron to obtain the predicted intussusception classification result.

[0071] 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 pairwise combination of ultrasound image, X-ray image and local X-ray image can be compared and learned.

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

[0073] Furthermore, in one embodiment of the present invention, the first loss function described above allows different encoders to learn aligned features for different images, thereby improving the fusion and utilization effect of different image features during intussusception classification.

[0074] In one embodiment of the present invention, to improve the robustness of the model to missing inputs, a random mask is used to randomly mask one feature with probability p1 when the feature vectors of the three input images are given. With the remaining probability (1-p1), two features are randomly masked with probability p2, where p1 and p2 are two hyperparameters. Furthermore, in another embodiment of the present invention, the method for masking feature vectors may include: recording a moving average value for the feature vectors of each image in the same batch, and using the corresponding moving average value to replace the original value to mask a certain feature.

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

[0076] Step 11: Determine the first target feature vector that needs to be masked from the first feature vector, the second feature vector, and the third feature vector using the first probability;

[0077] Step 12: Replace the first target feature vector with the corresponding moving average value to obtain the 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, fifth, and sixth feature vectors.

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

[0080] Step 1: Determine the second and third target feature vectors that need to be masked from the first, second, and third feature vectors using the second probability.

[0081] Step 2: Replace the second target feature vector and the third target feature vector with their corresponding moving averages to obtain the second mask feature vector and the 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 can 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 can be any two of the first feature vector, the second feature vector, and the third feature vector.

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

[0085] Furthermore, in one embodiment of the present invention, after obtaining the eighth feature vector through the above steps, the eighth feature vector can be passed through the multilayer perceptron in the classifier to obtain the probability vector of the 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 result through the above steps, the second loss function value corresponding to each batch of data can be determined based on the predicted intussusception classification result and the actual intussusception classification result of each batch of data, wherein the second loss function can be the 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 parameters in the first encoder, the second encoder, and the third encoder can be updated based on the first loss function value, and the second parameters 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 diagnosis model converges, and the trained intussusception diagnosis model is obtained.

[0088] In one embodiment of the present invention, after obtaining the trained intussusception diagnostic model through the above steps, the trained intussusception diagnostic model can be tested and optimized using a test dataset to obtain the accuracy performance of the trained intussusception diagnostic model on new samples and evaluate the model's generalization ability. Then, based on the evaluation results of the test dataset, the trained intussusception diagnostic model is adjusted and optimized, for example, by changing the network structure or hyperparameter settings, to improve the model's accuracy, robustness, and performance.

[0089] Step S3: Obtain the 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 invention, when only a patient ultrasound image is available, only the ultrasound image can be entered; when only a patient X-ray image is available, only the X-ray image can be entered; when an X-ray image is available and the location of intussusception can be determined therein, both the X-ray image and a local X-ray image can be entered.

[0092] Step S4: Input the image to be diagnosed into the target intussusception diagnostic model to obtain the intussusception classification result corresponding to the image to be diagnosed.

[0093] In one embodiment of the present invention, after obtaining the image to be diagnosed 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] The intussusception automatic diagnosis method based on deep learning proposed in this embodiment of the invention can analyze various images based on the target intussusception diagnosis model to obtain the corresponding intussusception classification results, thus having a wide range of applications and improving the robustness of the model and the accuracy of the classification results.

[0095] Next, referring to the accompanying drawings, an automatic diagnostic device for intussusception based on deep learning according to an embodiment of the present invention is described.

[0096] Figure 2 This is a schematic diagram of a deep learning-based automatic diagnosis device for intussusception according to an embodiment of the present invention.

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

[0098] Module 201 is used to build an initial diagnostic model for intussusception;

[0099] Training module 202 is used to acquire a dataset and train the initial intussusception diagnostic model based on the dataset to obtain the target intussusception diagnostic model.

[0100] The acquisition module 203 is used to acquire the image to be diagnosed;

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

[0102] 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 types of input images.

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

[0104] Different types of images from the same case were grouped into the same data set, and the intussusception type of the data set was labeled to obtain the training dataset and test dataset for training the initial intussusception diagnostic model.

[0105] The initial intussusception diagnostic model was trained using the training dataset to obtain the trained intussusception diagnostic model.

[0106] The trained intussusception diagnostic model was tested and optimized using a test dataset to obtain the target intussusception diagnostic model.

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

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

[0109] Based on the first, second, and third output results of each batch of data, the value of the first loss function corresponding to each batch of data is determined by the first loss function.

[0110] Based on the predicted intussusception classification results and the actual intussusception classification results of each batch of data, the value of the second loss function corresponding to each batch of data is determined by the second loss function.

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

[0112] Furthermore, the dynamic weighting module includes a fully connected layer and a Softmax layer, and the classifier includes a multilayer perceptron; the training module 202 is specifically used to: input different types of images of the same case in the data set 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 to 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 first feature vector, second feature vector and third feature vector have the same dimension;

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

[0114] The dynamic weighting module concatenates the fourth, fifth, and sixth feature vectors to obtain the seventh feature vector. The seventh feature vector is then passed through a fully connected layer and a Softmax layer to obtain a weight vector. Based on the weight vector, the fourth, fifth, and sixth feature vectors are weighted and summed to obtain the eighth feature vector.

[0115] The classifier passes the eighth feature vector through a multilayer perceptron to obtain the predicted classification result of intussusception.

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

[0117] The first target feature vector that needs to be masked is determined from the first feature vector, the second feature vector, and the third feature vector using the first probability.

[0118] The first target feature vector is replaced by the corresponding moving average to obtain the first mask feature vector;

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

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

[0121] The second target feature vector and the third target feature vector that need to be masked are determined from the first feature vector, the second feature vector and the third feature vector using the second probability.

[0122] The second target feature vector and the third target feature vector are replaced by their corresponding moving averages to obtain the second mask feature vector and the 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 require masking, the fourth feature vector, the fifth feature vector, and the sixth feature vector are obtained.

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

[0125] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0126] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0127] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to 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 includes: Construct an initial diagnostic model for intussusception; Obtain the dataset and train the initial intussusception diagnostic model based on the dataset to obtain the target intussusception diagnostic model; Acquire the image to be diagnosed; The image to be diagnosed is input into the target intussusception diagnosis model to obtain the intussusception classification result corresponding to the image to be diagnosed. The step of training the initial intussusception diagnostic model based on the dataset to obtain the target intussusception diagnostic model includes: Different types of images from the same case are divided into the same data group, and the intussusception type of the data group is labeled to obtain the training dataset and test dataset for training the initial intussusception diagnostic model. The initial intussusception diagnostic model is trained using the training dataset to obtain a trained intussusception diagnostic model. The initial intussusception diagnostic model includes a first encoder, a second encoder, a third encoder, a dynamic weighting module, and a classifier. The first encoder, second encoder, and third encoder correspond to different image types. Different types of images from the same case in the dataset are input into the corresponding first encoder, second encoder, and third encoder, respectively, to obtain first feature vectors, second feature vectors, and third feature vectors corresponding to different image types. The first feature vector, second feature vector, and third feature vector are then identified as the first output result, second output result, and third output result, respectively. The results and third output result are as follows: the first feature vector, the second feature vector, and the third feature vector have the same dimension; the first feature vector, the second feature vector, and the third feature vector are randomly masked to obtain the fourth feature vector, the fifth feature vector, and the sixth feature vector; the fourth feature vector, the fifth feature vector, and the sixth feature vector are concatenated to obtain the seventh feature vector; the seventh feature vector is passed through a fully connected layer and a Softmax layer to obtain a weight vector; and the fourth feature vector, the fifth feature vector, and the sixth feature vector are weighted and summed based on the weight vector to obtain the eighth feature vector; the eighth feature vector is passed through a multilayer perceptron to obtain the predicted intussusception classification result. The trained intussusception diagnostic model was tested and optimized using the test dataset to obtain the target intussusception diagnostic model. The step of randomly masking the first, second, and third feature vectors to obtain the fourth, fifth, and sixth feature vectors includes: The first target feature vector that needs to be masked is determined from the first feature vector, the second feature vector, and the third feature vector by a first probability, wherein the first target feature vector is any one of the first feature vector, the second feature vector, and the third feature vector; The first target feature vector is replaced by the corresponding moving average to obtain the first mask feature vector; Based on the first mask feature vector and the remaining feature vectors that do not need to be masked, the fourth feature vector, the fifth feature vector, and the sixth feature vector are obtained; The step of training the initial intussusception diagnostic model using the training dataset to obtain the trained intussusception diagnostic model includes: The data components in the training dataset are input into the initial intussusception diagnosis model in batches to obtain the first output result, second output result, third output result and predicted intussusception classification result for each batch of data; Based on the first output result, second output result and third output result of each batch of data, the value of the first loss function corresponding to each batch of data is determined by the first loss function, wherein comparative learning is performed by the first loss function. Based on the predicted intussusception classification result and the actual intussusception classification result of each batch of data, the value of the second loss function corresponding to each batch of data is determined by the second loss function. The first parameters in the first encoder, second encoder, and third encoder are updated based on the first loss function value, and the second parameters in the dynamic weighting module and 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, and the trained intussusception diagnostic model is obtained. The dynamic weighting module includes a fully connected layer and a Softmax layer, and the classifier includes a multilayer perceptron.

2. An automated diagnostic device for intussusception based on deep learning, characterized in that, The device includes: The building block is used to construct the initial diagnostic model for intussusception; The training module is used to acquire a dataset and train the initial intussusception diagnostic model based on the dataset to obtain the target intussusception diagnostic model. The acquisition module is used to acquire the image to be diagnosed. The diagnostic module is used to input the image to be diagnosed into the target intussusception diagnostic model to obtain the intussusception classification result corresponding to the image to be diagnosed; 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 types of input images; The training module is specifically used for: Different types of images from the same case are divided into the same data group, and the intussusception type of the data group is labeled to obtain the training dataset and test dataset for training the initial intussusception diagnostic model. The initial intussusception diagnostic model is trained using the training dataset to obtain a trained intussusception diagnostic model. The initial intussusception diagnostic model includes a first encoder, a second encoder, a third encoder, a dynamic weighting module, and a classifier. The first encoder, second encoder, and third encoder correspond to different image types. Different types of images from the same case in the dataset are input into the corresponding first encoder, second encoder, and third encoder, respectively, to obtain first feature vectors, second feature vectors, and third feature vectors corresponding to different image types. The first feature vector, second feature vector, and third feature vector are then identified as the first output result, second output result, and third output result, respectively. The results and third output result are as follows: the first feature vector, the second feature vector, and the third feature vector have the same dimension; the first feature vector, the second feature vector, and the third feature vector are randomly masked to obtain the fourth feature vector, the fifth feature vector, and the sixth feature vector; the fourth feature vector, the fifth feature vector, and the sixth feature vector are concatenated to obtain the seventh feature vector; the seventh feature vector is passed through a fully connected layer and a Softmax layer to obtain a weight vector; and the fourth feature vector, the fifth feature vector, and the sixth feature vector are weighted and summed based on the weight vector to obtain the eighth feature vector; the eighth feature vector is passed through a multilayer perceptron to obtain the predicted intussusception classification result. The trained intussusception diagnostic model was tested and optimized using the test dataset to obtain the target intussusception diagnostic model. The training module is further configured to: The first target feature vector that needs to be masked is determined from the first feature vector, the second feature vector, and the third feature vector by a first probability, wherein the first target feature vector is any one of the first feature vector, the second feature vector, and the third feature vector; The first target feature vector is replaced by the corresponding moving average to obtain the first mask feature vector; Based on the first mask feature vector and the remaining feature vectors that do not need to be masked, the fourth feature vector, the fifth feature vector, and the sixth feature vector are obtained; The training module is also used for: The data components in the training dataset are input into the initial intussusception diagnosis model in batches to obtain the first output result, second output result, third output result and predicted intussusception classification result for each batch of data; Based on the first output result, second output result and third output result of each batch of data, the value of the first loss function corresponding to each batch of data is determined by the first loss function, wherein comparative learning is performed by the first loss function. Based on the predicted intussusception classification result and the actual intussusception classification result of each batch of data, the value of the second loss function corresponding to each batch of data is determined by the second loss function. The first parameters in the first encoder, second encoder, and third encoder are updated based on the first loss function value, and the second parameters in the dynamic weighting module and 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, and the trained intussusception diagnostic model is obtained. The dynamic weighting module includes a fully connected layer and a Softmax layer, and the classifier includes a multilayer perceptron.

3. 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 to enable the at least one processor to perform the method of claim 1.

4. 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, it implements the method as described in claim 1.

Images