Training method, recognition method and device for alimentary canal perforation segmentation model
By adopting supervised and unsupervised classification and segmentation methods in the diagnosis of digestive tract perforation, combined with image segmentation technology and pseudo-optimization labels, the problem of low recognition accuracy in the existing technology is solved, and the accurate identification and cause of digestive tract perforation is achieved. Analysis improves the accuracy and reference value of diagnosis.
Patent Information
- Application Number
- CN202411181120.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-26
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-08-26
AI Technical Summary
The existing diagnostic methods for perforation of digestive tract are insufficient in identifying the specific distribution location and size of perforation of digestive tract, resulting in low accuracy and inability to effectively distinguish the causes of perforation.
The supervised and unsupervised classification and segmentation method is adopted, and the semi-supervised model training is combined with image segmentation technology to locate the specific distribution position and size of the digestive tract perforation, and analyze the severity according to the perforation situation. The pseudo-optimized label and Cosine distance loss function are used. Optimize network model.
It improves the accuracy of gastrointestinal perforation identification, can ensure the accuracy of the model with less labeling workload, and provides references to the causes of perforation for subsequent analysis, enhancing the reference value of diagnosis.
Smart Images

Figure CN119048863A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of medical image processing, and in particular, to a training method, an identification method and a device for a digestive tract perforation segmentation model. Background Art
[0002] Currently, the diagnosis of gastrointestinal perforation depends largely on imaging examinations. Correctly diagnosing the presence, extent, and cause of perforation is very important for appropriate patient management and surgical planning. With the continuous development of CT (Computed Tomography) imaging technology, it is possible to use thin-slice CT to examine the entire abdomen in one breath hold. In summary, a technique for quickly and accurately segmenting gastrointestinal perforations using CT images is needed to identify gastrointestinal perforations.
[0003] In order to do so, the related technology discloses a method for detecting digestive tract perforation based on deep learning, which uses a network model to determine whether there is free gas in the medical image to determine whether there is digestive tract perforation. Since the Hounsfield unit value (HU value) of gas in medical images is lower than that of human tissue structure, by setting the HU value, it is possible to preliminarily determine whether there is free gas in the image and realize image annotation.
[0004] In the process of implementing the embodiments of the present disclosure, it is found that there are at least the following problems in the related art: Although the related technology can identify free gas, the output of the network model is only 0 and 1, and the specific distribution location of the free gas cannot be given, which provides little reference value for the subsequent diagnosis of gastrointestinal perforation. In addition, due to different causes of perforation, the manifestation of gastrointestinal perforation is also different. Judging whether there is gastrointestinal perforation only by free gas may result in low accuracy.
[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present application, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention
[0006] In order to provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. The summary is not an extensive review, nor is it intended to identify key / critical elements or delineate the scope of protection of these embodiments, but rather serves as a prelude to the detailed description that follows.
[0007] The embodiments of the present disclosure provide a training method, an identification method and a device for a digestive tract perforation segmentation model, which can improve the accuracy of digestive tract perforation identification.
[0008] In some embodiments, a training method for a gastrointestinal perforation segmentation model includes: obtaining a predicted value output by a network model based on gastrointestinal perforation marked data and a label corresponding to the gastrointestinal perforation marked data; obtaining a pseudo-optimized label corresponding to the unlabeled gastrointestinal perforation data based on the unlabeled gastrointestinal perforation data; wherein the change directions of different pseudo-optimized labels are consistent; and training the network model based on the predicted value and the pseudo-optimized label to obtain a gastrointestinal perforation segmentation model.
[0009] Optionally, based on the unlabeled data of digestive tract perforation, a pseudo-optimized label corresponding to the unlabeled data of digestive tract perforation is obtained, including: obtaining a pseudo-initial label output by a network model; obtaining voxels in the pseudo-initial label; and assigning weights to the voxels in the pseudo-initial label according to the confidence of the pseudo-initial label to obtain the pseudo-optimized label.
[0010] Optionally, based on the unlabeled data of digestive tract perforation, a pseudo-optimized label corresponding to the unlabeled data of digestive tract perforation is obtained, including: obtaining a pseudo-initial label output by a network model; calculating the similarity of multiple pseudo-initial labels based on a Cosine distance loss function; and performing regularization constraints on multiple pseudo-initial labels based on the similarity to obtain a pseudo-optimized label.
[0011] Optionally, the unlabeled digestive tract perforation data are obtained in the following manner: a plurality of data enhancement transformations are performed on the labeled digestive tract perforation data to obtain a plurality of unlabeled digestive tract perforation data.
[0012] Optionally, the network model is trained according to the predicted values and the pseudo-optimized labels to obtain a digestive tract perforation segmentation model, including: combining the predicted values with multiple pseudo-optimized labels to obtain multiple combined labels; using the multiple combined labels to perform pseudo-supervised loss optimization on the network model to obtain multiple training models; and selecting the training model with the best segmentation effect from the multiple training models as the digestive tract perforation segmentation model.
[0013] Optionally, the pseudo-supervisory loss of the network model is optimized using multiple combined labels to obtain multiple training models, including: calculating the comprehensive losses of the predicted values and pseudo-labels in the multiple combined labels according to the loss function; updating the parameters of the multiple network models according to the comprehensive losses of the predicted values and pseudo-labels in the multiple data groups to obtain multiple training models.
[0014] Optionally, the method for obtaining multiple training models also includes: calculating the loss of predicted values in multiple combination labels respectively according to the loss function; and updating the parameters of the network model respectively according to the loss of the predicted values to obtain multiple training models.
[0015] In some embodiments, a method for identifying gastrointestinal perforation includes: obtaining a medical gastrointestinal image to be identified; inputting the medical gastrointestinal image to be identified into a gastrointestinal perforation segmentation model such as the above-mentioned to obtain a prediction result of gastrointestinal perforation; marking the prediction result in the identified medical gastrointestinal image; and generating a diagnosis report based on the marked medical gastrointestinal image.
[0016] In some embodiments, a device for identifying gastrointestinal perforation includes: a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the above-mentioned training method for the gastrointestinal perforation segmentation model, or the above-mentioned method for identifying gastrointestinal perforation, through the computer program.
[0017] In some embodiments, the medical device includes: a medical device body; and, the identification device for digestive tract perforation as described above, installed on the medical device body.
[0018] The training method, identification method and device for the digestive tract perforation segmentation model provided in the embodiments of the present disclosure can achieve the following technical effects: In the disclosed embodiment, due to the various causes of digestive tract perforation and the different manifestations that may occur in perforations of different tissue structures, a classification segmentation method of supervised segmentation and unsupervised segmentation is adopted, so that different manifestations of digestive tract perforation can be segmented, providing a reference for the subsequent analysis of the possible causes of digestive tract perforation. Part of the data is labeled, and the part of the data should include all possible manifestations of digestive tract perforation, including medical images of different perforation locations in the abdominal cavity, to form a training set for the semi-supervised model. In this way, a large amount of unlabeled data of digestive tract perforation can be used to ensure the accuracy of the model while reducing the workload of labeling. Furthermore, using the image segmentation method, the specific distribution position and perforation size of the digestive tract perforation manifestations can be located, and a reference can be provided for the subsequent analysis of the severity of the digestive tract perforation based on the perforation situation.
[0019] The above general description and the following description are exemplary and explanatory only and are not intended to limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] One or more embodiments are exemplarily described by corresponding drawings, which do not limit the embodiments. Elements with the same reference numerals in the drawings are shown as similar elements, and the drawings do not constitute a scale limitation, and wherein: Figure 1 is a schematic diagram of an implementation environment of a training method for a digestive tract perforation segmentation model provided in an embodiment of the present disclosure; Figure 2 is a schematic diagram of a training method for a digestive tract perforation segmentation model provided by an embodiment of the present disclosure; Figure 3 is a schematic diagram of another training method for a digestive tract perforation segmentation model provided by an embodiment of the present disclosure; Figure 4 is a schematic diagram of another training method for a digestive tract perforation segmentation model provided by an embodiment of the present disclosure; Figure 5 is a schematic diagram of a method for identifying digestive tract perforation provided by an embodiment of the present disclosure; Figure 6 is a schematic diagram of a digestive tract perforation identification device provided by an embodiment of the present disclosure; Figure 7 It is a schematic diagram of the basic structure of an algorithm of a method for detecting digestive tract perforation based on deep learning provided in an embodiment of the present disclosure; Figure 8 is a schematic diagram of a training module of a training method for a digestive tract perforation segmentation model provided by an embodiment of the present disclosure; Fig. 9 This is a CT enhanced image of a patient with duodenal perforation provided by an embodiment of the present disclosure. DETAILED DESCRIPTION
[0021] In order to be able to understand the features and technical contents of the embodiments of the present disclosure in more detail, the implementation of the embodiments of the present disclosure is described in detail below in conjunction with the accompanying drawings. The attached drawings are for reference only and are not used to limit the embodiments of the present disclosure. In the following technical description, for the convenience of explanation, a full understanding of the disclosed embodiments is provided through multiple details. However, one or more embodiments can still be implemented without these details. In other cases, to simplify the drawings, well-known structures and devices can be simplified for display.
[0022] The terms "first", "second", etc. in the specification and claims of the embodiments of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged where appropriate, so that the embodiments of the embodiments of the present disclosure described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions.
[0023] Unless otherwise stated, the term "plurality" means two or more.
[0024] In the embodiment of the present disclosure, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B indicates: A or B.
[0025] The term "and / or" is a description of the association relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or, A and B.
[0026] The term "correspondence" may refer to an association relationship or a binding relationship. The correspondence between A and B means that there is an association relationship or a binding relationship between A and B.
[0027] The digestive tract is a complex system responsible for food digestion and absorption in the human body. It starts from the mouth, passes through the pharynx, esophagus, stomach, small intestine, large intestine, and finally reaches the anus. It includes not only the digestive organs, but also the digestive juices secreted by the digestive glands, such as saliva, gastric juice, and intestinal juice. Food is broken down into small molecules in the digestive tract to facilitate the body's absorption of nutrients. The small intestine is the main absorption site, while the large intestine is mainly responsible for absorbing water and electrolytes, forming and storing excrement. The entire digestive tract pushes food forward through the peristaltic movement of muscles to complete the digestion and absorption process.
[0028] Gastrointestinal tract perforation is a serious and urgent medical condition. When the wall of the digestive tract (including the esophagus, stomach, small intestine, etc.) ruptures, it is easy for the contents of the digestive tract to leak into the abdominal cavity or other cavities. This rupture is a surgical emergency and requires immediate intervention to prevent contamination of the peritoneal cavity, otherwise it may cause serious complications or even endanger life. Causes of digestive tract perforation include digestive tract ulcers, the growth of digestive tract tumors leading to the destruction of surrounding tissues, etc.
[0029] The diagnosis of gastrointestinal perforation depends largely on imaging examinations. The difficulties of the gastrointestinal perforation segmentation task are as follows: First, the location of the gastrointestinal perforation accounts for a very small proportion of the total image, the diameter of the hole is very small, and the probability of the perforation being directly and accurately identified is very low. Secondly, the perforation manifestations may be different for organs in different locations. For example, the representation of gastrointestinal perforation may be symptoms such as free gas in the abdominal cavity, ascites, intestinal wall thickening, increased mesentery density, and continuous middle section of the intestinal wall. Therefore, it is time-consuming and laborious to annotate each image, which means that the number of labeled samples of the training data will be limited. Furthermore, the unclear representation of the lesion area also increases the difficulty of identifying gastrointestinal perforation. At present, there are some related technologies dedicated to identifying free gas in the abdominal cavity as a representation of gastrointestinal perforation, but they are all based on image classification, and the specific location and size of free gas or perforation are not given, and the other representations of gastrointestinal perforation are not taken into account, giving little reference value for the clinical diagnosis of gastrointestinal perforation. In addition, since a small amount of gas is not obvious in the image, the false negative rate of the classification model is very high. like Fig. 9 As shown, small amounts of extraluminal air between the anterior peritoneal surface of the liver and the mid-abdomen and peritoneal folds can be easily overlooked.
[0030] In the related art, considering that the hole at the perforation location is very small and difficult to find, most implementation methods determine whether there is digestive tract perforation by judging whether there is free gas in the medical image. Figure 7 The basic structure diagram of the algorithm of the method for detecting digestive tract perforation based on deep learning is shown. The image is input into the deep classification network for training and testing. After stacking operations such as convolution, pooling, and full connection, the network model is constructed, and the manually labeled data set is aligned for training. The model will output the classification result, and finally realize the recognition of digestive tract perforation. In the training stage, all training data are input into the network model Network, and the network model calculates the prediction result, that is, 0 or 1, indicating the presence or absence of free gas. The prediction result and the corresponding label are simultaneously substituted into the loss function to calculate the error value. This error value is back-propagated to each layer in the network model Network, and the corresponding parameters are updated to achieve the purpose of reducing the error. In the testing stage, the test data is input into the trained Network model to obtain the network prediction result. Based on the obtained classification result, the recognition result of whether the image contains free gas is obtained.
[0031] The disclosed embodiment aims to realize an accurate digestive tract perforation segmentation technology. Figure 1 FIG. 1 is a schematic diagram of an implementation environment of a training method for a digestive tract perforation segmentation model provided in an embodiment of the present disclosure. Figure 1 As shown, the implementation environment of the above method may include a processor 600 and a database 604 .
[0032] The database 604 stores a large number of medical digestive tract images, and a training data set can be constructed after image annotation. The processor 600 can train a digestive tract perforation segmentation model based on the training data set provided by the database 604, thereby realizing the analysis of digestive tract perforation.
[0033] Combination Figure 2 As shown, the embodiment of the present disclosure provides a training method for a digestive tract perforation segmentation model, comprising: S201, the processor obtains the predicted value output by the network model according to the digestive tract perforation marking data and the label corresponding to the digestive tract perforation marking data.
[0034] S202, the processor obtains a pseudo-optimization label corresponding to the unlabeled digestive tract perforation data according to the unlabeled digestive tract perforation data; wherein the change directions of different pseudo-optimization labels are consistent.
[0035] S203: The processor trains the network model according to the predicted values and the pseudo-optimization labels to obtain a digestive tract perforation segmentation model.
[0036] In the training method for the digestive tract perforation segmentation model provided in the embodiment of the present disclosure, due to the various causes of digestive tract perforation and the different representations that may appear in perforations of different tissue structures, a classification segmentation method of supervised segmentation and unsupervised segmentation is adopted, so that different representations of digestive tract perforation can be segmented, providing a reference for the subsequent analysis of the possible causes of digestive tract perforation. Part of the data is labeled, and the part of the data should include all possible representations of digestive tract perforation, including medical images of different perforation locations in the abdominal cavity, to form a training set for the semi-supervised model. In this way, a large amount of unlabeled digestive tract perforation data can be used to ensure the accuracy of the model while reducing the workload of labeling. Furthermore, using the image segmentation method, the specific distribution position and perforation size of the digestive tract perforation representation can be located, and a reference can be provided for the subsequent analysis of the severity of the digestive tract perforation based on the perforation situation.
[0037] Combination Figure 8 As shown, in the embodiment of the present disclosure, the processor uses the digestive tract perforation labeled data and the digestive tract perforation unlabeled data as network input through the input module 80. The model training is performed through the network training module 81, and the network training module 81 includes a supervised segmentation module 811, an unsupervised segmentation module 812 and a consistency constraint module 813. Finally, the model generation module 82 selects a model with the best segmentation effect from the multiple models obtained through training.
[0038] Optionally, based on the unlabeled data of digestive tract perforation, a pseudo-optimized label corresponding to the unlabeled data of digestive tract perforation is obtained, including: obtaining a pseudo-initial label output by a network model; obtaining voxels in the pseudo-initial label; and assigning weights to the voxels in the pseudo-initial label according to the confidence of the pseudo-initial label to obtain the pseudo-optimized label.
[0039] Combination Figure 3 As shown, the embodiment of the present disclosure provides another training method for a digestive tract perforation segmentation model, comprising: S301, the processor obtains the predicted value output by the network model according to the digestive tract perforation mark data and the label corresponding to the digestive tract perforation mark data.
[0040] S302: The processor obtains a pseudo initial label output by the network model.
[0041] S303: The processor obtains voxels in the pseudo initial label.
[0042] S304: The processor assigns weights to voxels in the pseudo initial label according to the confidence of the pseudo initial label to obtain a pseudo optimized label.
[0043] S305: The processor trains the network model according to the predicted values and the pseudo-optimization labels to obtain a digestive tract perforation segmentation model.
[0044] Pseudo-optimized labels refer to using the model's own prediction results as labels when no real labels are available. The information of unlabeled data is used to predict the data through the model to generate pseudo labels, and the pseudo labels are used together with the real labels to further train the model to improve the generalization ability and performance of the model.
[0045] When generating pseudo labels, the category with the highest model prediction probability is usually selected as the pseudo label, and different strategies may be used to ensure the reliability of the pseudo labels, such as probability-based thresholding or sorting-based ranking. In addition, a common problem of pseudo labeling methods is that noise and errors may be introduced, so pseudo labels need to be carefully selected and may be combined with uncertainty estimation to improve their quality.
[0046] In practical applications, pseudo-optimized labels can significantly improve the prediction performance of the model on unlabeled data, especially when labeled data is scarce. For example, in image classification tasks, pseudo-label learning can help the model learn richer feature representations, thereby improving classification accuracy. In addition, pseudo-label methods are also widely used in other fields such as text classification and recommendation systems.
[0047] In the disclosed embodiment, the input of the network model is unlabeled data of digestive tract perforation. The output of the network is a pseudo-initial label. However, since the pseudo-initial label may contain potential noise, the pseudo-supervision process is corrected by using an uncertainty estimate. Uncertainty estimation introduces adaptive voxel weights in pseudo-supervised loss. Adaptive voxel weights are a technique used in tasks such as medical image segmentation, especially in semi-supervised or unsupervised learning scenarios. The weights in the loss function are dynamically adjusted according to the prediction confidence of the voxel (the counterpart of the pixel point in the image in three-dimensional space). Among them, voxels with higher confidence have higher weights, while voxels with lower confidence have lower weights. By reducing the weights of voxels with low confidence, the negative effects caused by incorrect predictions or noise can be reduced. Thereby reducing the impact of label noise and improving the robustness of segmentation.
[0048] Optionally, based on the unlabeled data of digestive tract perforation, a pseudo-optimized label corresponding to the unlabeled data of digestive tract perforation is obtained, including: obtaining a pseudo-initial label output by a network model; calculating the similarity of multiple pseudo-initial labels based on a Cosine distance loss function; and performing regularization constraints on multiple pseudo-initial labels based on the similarity to obtain a pseudo-optimized label.
[0049] Combination Figure 4 As shown, the embodiment of the present disclosure provides another training method for a digestive tract perforation segmentation model, comprising: S401, the processor obtains the predicted value output by the network model according to the digestive tract perforation mark data and the label corresponding to the digestive tract perforation mark data.
[0050] S402: The processor obtains a pseudo initial label output by the network model.
[0051] S403: The processor calculates the similarity of multiple pseudo initial labels according to the Cosine distance loss function.
[0052] S404: The processor performs regularization constraints on multiple pseudo initial labels according to the similarity to obtain pseudo optimized labels.
[0053] S405: The processor trains the network model according to the predicted values and the pseudo-optimization labels to obtain a digestive tract perforation segmentation model.
[0054] In the disclosed embodiment, the Cosine distance loss function is used to minimize the dissimilarity between pseudo-initial labels to ensure that the prediction results obtained from the same original image through different enhancement methods are consistent. Cosine distance loss is a loss function that measures the angular difference between two vectors. It is commonly used in machine learning and deep learning, especially in scenarios where the direction of the vector rather than the size needs to be considered. The calculation of the Cosine loss function is based on the cosine similarity between vectors, and its value ranges from -1 to 1, where 1 means that the vectors are exactly the same, 0 means that the vectors are orthogonal, and -1 means that the vectors are completely opposite.
[0055] The Cosine distance loss function can be calculated according to the following formula:
[0056] Among them, cosine distance loss is the Cosine distance loss, and a and b are two vectors.
[0057] The Cosine distance loss function is not affected by the length of the vector and only focuses on the directional difference. In neural networks, the Cosine distance loss can be used in regularization layers to encourage the model output to be consistent in direction with the expected output vector, rather than simply reducing the Euclidean distance. For example, in the consistency constraint of the digestive tract perforation segmentation model, the Cosine distance loss may be used to ensure that the prediction results obtained by different data augmentation methods are consistent in direction, thereby improving the stability and robustness of the model.
[0058] Optionally, the unlabeled digestive tract perforation data are obtained in the following manner: a plurality of data enhancement transformations are performed on the labeled digestive tract perforation data to obtain a plurality of unlabeled digestive tract perforation data.
[0059] In the disclosed embodiment, two different pixel intensity-based data enhancements are performed on the input digestive tract perforation labeled data to obtain digestive tract perforation unlabeled data, and the digestive tract perforation labeled data and the digestive tract perforation unlabeled data constitute the input of the network. The data enhancement methods include random intensity scaling, random intensity shifting, and Gaussian noise.
[0060] Optionally, the network model is trained according to the predicted values and the pseudo-optimized labels to obtain a digestive tract perforation segmentation model, including: combining the predicted values with multiple pseudo-optimized labels to obtain multiple combined labels; using the multiple combined labels to perform pseudo-supervised loss optimization on the network model to obtain multiple training models; and selecting the training model with the best segmentation effect from the multiple training models as the digestive tract perforation segmentation model.
[0061] In the disclosed embodiments, pseudo-supervised loss optimization is a semi-supervised learning method that uses unlabeled data (pseudo-labels) to train the model while combining a small amount of labeled data. It is particularly suitable for situations where labeled data is scarce but unlabeled data is abundant.
[0062] Use the current model to predict the unlabeled data of gastrointestinal perforation and generate pseudo labels. Combine the pseudo labels with the model prediction results and design a loss function to optimize the model. Include supervised loss and pseudo-supervised loss. Estimate the uncertainty of the generated pseudo labels to identify labels that may be noisy or erroneous. Assign different weights to each sample in the loss function based on the uncertainty estimate. From the multiple models trained, select the one with the best segmentation effect as the gastrointestinal perforation segmentation model.
[0063] Optionally, the pseudo-supervisory loss of the network model is optimized using multiple combined labels to obtain multiple training models, including: calculating the comprehensive losses of the predicted values and pseudo-labels in the multiple combined labels according to the loss function; updating the parameters of the multiple network models according to the comprehensive losses of the predicted values and pseudo-labels in the multiple data groups to obtain multiple training models.
[0064] In the disclosed embodiment, supervised loss and pseudo-supervised loss are combined to form a comprehensive loss function. The comprehensive loss function is used to train and update the model. The process may include iteratively recalculating pseudo labels and updating model parameters. The pseudo-supervised loss optimization method allows the training model to learn from unlabeled data, while improving the robustness of the training model by combining predictions from multiple data groups. Through adaptive weight allocation, the training model can pay more attention to areas that are easier to predict correctly, thereby improving the overall segmentation accuracy and model performance.
[0065] Optionally, the method for obtaining multiple training models also includes: calculating the loss of predicted values in multiple combination labels respectively according to the loss function; and updating the parameters of the network model respectively according to the loss of the predicted values to obtain multiple training models.
[0066] During the training process, an iterative optimization strategy is adopted to repeat the loss calculation and parameter update until the model performance no longer improves on the validation set or reaches the predetermined number of iterations. Implementing this solution requires building a flexible training framework to adapt to the characteristics of different combinations. First, the data is preprocessed and enhanced to improve the model's adaptability to new situations. Then, a multi-task learning architecture is designed to allow the model to learn from labeled and unlabeled data simultaneously. In each iteration, the model uses cross entropy loss and a custom consistency loss to optimize parameters to ensure that the prediction results of different combinations are semantically consistent. In this way, the model can self-correct during the training process and gradually improve its understanding of complex data distributions. Finally, the optimal solution is extracted from multiple training models through ensemble learning or model selection strategies to achieve the best performance.
[0067] Combination Figure 5 As shown, the embodiment of the present disclosure provides a method for identifying digestive tract perforation, comprising: S501: The processor obtains a medical digestive tract image to be identified.
[0068] S502, the processor inputs the medical digestive tract image to be identified into the digestive tract perforation segmentation model as described above to obtain a prediction result of digestive tract perforation.
[0069] S503: The processor marks the prediction result in the identified medical digestive tract image.
[0070] S504: The processor generates a diagnosis report based on the annotated medical digestive tract image.
[0071] In the disclosed embodiment, the digestive tract perforation segmentation model obtained in the training phase is used to test the image to obtain the prediction results of each digestive tract perforation representation. The predicted multi-classification segmentation results are visualized and displayed on the original image, which can clearly show the location distribution and size of the digestive tract perforation representation. Different categories of digestive tract perforation representations are displayed in different colors, providing doctors with more direct reference information for diagnosis based on clinical symptoms.
[0072] The diagnosis report includes whether there is gastrointestinal perforation, the cause of gastrointestinal perforation, the number of perforation locations, and the volume (area) of each location. The cause of gastrointestinal perforation is preliminarily determined based on the segmentation results. If the segmentation results contain free gas or effusion, it is considered that there is gastrointestinal perforation; if the segmentation results also contain the categories of tumors and other diseases, it is further considered that the cause of gastrointestinal perforation is tumors and other diseases; if the segmentation results also contain the category of foreign body ingestion, it is considered that the cause of gastrointestinal perforation is due to foreign body ingestion; in addition, according to the segmentation results, there may be multiple causes of gastrointestinal perforation at the same time.
[0073] Combination Figure 6 As shown, the embodiment of the present disclosure provides a device 60 for identifying digestive tract perforation, including a processor 600 and a memory 601. Optionally, the device 60 may also include a communication interface 602 and a bus 603. The processor 600, the communication interface 602, and the memory 601 may communicate with each other through the bus 603. The communication interface 602 may be used for information transmission. The processor 600 may call the logic instructions in the memory 601 to execute the training method for the digestive tract perforation segmentation model of the above embodiment, or the above-mentioned method for identifying digestive tract perforation.
[0074] In addition, the logic instructions in the memory 601 described above can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.
[0075] The memory 601 is a computer-readable storage medium that can be used to store software programs and computer executable programs, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor 600 executes the function application and data processing by running the program instructions / modules stored in the memory 601, that is, the training method for the digestive tract perforation segmentation model in the above embodiment, or the above-mentioned method for identifying digestive tract perforation.
[0076] The memory 601 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory 601 may include a high-speed random access memory and may also include a non-volatile memory.
[0077] The disclosed embodiment provides a medical device, including: a medical device body, and the above-mentioned identification device 60 for digestive tract perforation. The identification device 60 for digestive tract perforation is installed on the medical device body. The installation relationship described here is not limited to placement inside the medical device body, but also includes installation connections with other components of the medical device, including but not limited to physical connection, electrical connection or signal transmission connection, etc. It can be understood by those skilled in the art that the identification device 60 for digestive tract perforation can be adapted to a feasible medical device body, thereby realizing other feasible embodiments.
[0078] The technical solution of the embodiment of the present disclosure can be embodied in the form of a software product, which is stored in a storage medium and includes one or more instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiment of the present disclosure. The aforementioned storage medium may be a non-transient storage medium, such as: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, and other media that can store program codes.
[0079] The above description and the accompanying drawings fully illustrate the embodiments of the present disclosure so that those skilled in the art can practice them. Other embodiments may include structural, logical, electrical, process and other changes. The embodiments represent only possible changes. Unless explicitly required, separate components and functions are optional, and the order of operations may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the words used in this application are only used to describe the embodiments and are not used to limit the claims. As used in the description of the embodiments and the claims, unless the context clearly indicates, the singular forms of "a", "an" and "the" are intended to include plural forms as well. Similarly, the term "and / or" as used in this application refers to any and all possible combinations of one or more associated listings. In addition, when used in this application, the term "comprise" and its variants "comprises" and / or comprising refer to the presence of stated features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or groups thereof. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method or device including the elements. In this article, each embodiment may focus on the differences from other embodiments, and the same and similar parts between the embodiments may refer to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method part disclosed in the embodiments, then the relevant parts can refer to the description of the method part.
[0080] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software may depend on the specific application and design constraints of the technical solution. The technicians may use different methods for each specific application to implement the described functions, but such implementations should not be considered to exceed the scope of the embodiments of the present disclosure. The technicians may clearly understand that, for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above may refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.
[0081] In the embodiments disclosed herein, the disclosed methods and products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units can be only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to implement this embodiment. In addition, each functional unit in the embodiment of the present disclosure may be integrated in a processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit.
[0082] The flowchart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to the embodiment of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. In the description corresponding to the flowchart and the block diagram in the accompanying drawings, the operations or steps corresponding to different boxes can also occur in a different order from the order disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified functions or actions, or may be implemented by a combination of dedicated hardware and computer instructions.
Claims
1. A training method for a digestive tract perforation segmentation model, characterized in that: include: Obtaining a predicted value output by the network model according to the digestive tract perforation labeled data and a label corresponding to the digestive tract perforation labeled data; According to the unlabeled data of digestive tract perforation, a pseudo-optimization label corresponding to the unlabeled data of digestive tract perforation is obtained; wherein the change directions of different pseudo-optimization labels are consistent; According to the predicted values and pseudo-optimized labels, the network model is trained to obtain the digestive tract perforation segmentation model.
2. The training method according to claim 1, characterized in that: According to the unlabeled data of digestive tract perforation, the pseudo-optimized labels corresponding to the unlabeled data of digestive tract perforation are obtained, including: Get the pseudo initial label output by the network model; Get the voxels in the pseudo initial label; According to the confidence of the pseudo initial label, the voxels in the pseudo initial label are weighted to obtain the pseudo optimized label.
3. The training method according to claim 1, characterized in that: According to the unlabeled data of digestive tract perforation, the pseudo-optimized labels corresponding to the unlabeled data of digestive tract perforation are obtained, including: Get the pseudo initial label output by the network model; According to the Cosine distance loss function, the similarity of multiple pseudo initial labels is calculated; According to the similarity, regularization constraints are applied to multiple pseudo initial labels to obtain pseudo optimized labels.
4. The training method according to claim 1, characterized in that: Obtain unlabeled data of digestive tract perforation as follows: A variety of data augmentation transformations are performed on the digestive tract perforation labeled data to obtain multiple digestive tract perforation unlabeled data.
5. The training method according to claim 1, characterized in that: According to the predicted values and pseudo-optimized labels, the network model is trained to obtain the digestive tract perforation segmentation model, including: Combine the predicted values with multiple pseudo-optimized labels respectively to obtain multiple combined labels; Use multiple combined labels to optimize pseudo-supervisory losses on the network model to obtain multiple training models; The training model with the best segmentation effect is selected from multiple training models as the digestive tract perforation segmentation model.
6. The training method according to claim 5, characterized in that: Use multiple combined labels to optimize the pseudo-supervisory loss of the network model and obtain multiple training models, including: According to the loss function, the comprehensive losses of the predicted values and pseudo labels in multiple combined labels are calculated respectively; According to the comprehensive losses of the predicted values and pseudo labels in the multiple data groups, the parameters of the multiple network models are updated respectively to obtain multiple training models.
7. The training method according to claim 5, characterized in that: Methods for obtaining multiple training models also include: According to the loss function, the loss of the predicted values in multiple combined labels is calculated respectively; According to the loss of the predicted value, the parameters of the network model are updated respectively to obtain multiple training models.
8. A method for identifying digestive tract perforation, characterized in that: include: Acquire a medical digestive tract image to be identified; Inputting the medical digestive tract image to be identified into the digestive tract perforation segmentation model according to any one of claims 1 to 7 to obtain a prediction result of digestive tract perforation; annotating prediction results in identified medical digestive tract images; Generate a diagnosis report based on annotated medical digestive tract images.
9. A device for identifying digestive tract perforation, comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to execute through the computer program: the training method for the digestive tract perforation segmentation model as described in any one of claims 1 to 7, or the method for identifying digestive tract perforation as described in claim 8.
10. A medical device, characterized in that: include: Medical device body; and, The digestive tract perforation identification device as described in claim 9 is installed on the medical device body.
Citation Information
Patent Citations
Ear-nose-throat examination image segmentation method and system
CN118470315A
