Training methods, recognition methods, and devices for segmentation models of gastrointestinal perforation

By training a gastrointestinal perforation segmentation model using supervised and unsupervised segmentation methods, and utilizing pseudo-optimized labels and the Cosine distance loss function, the low accuracy of gastrointestinal perforation identification in existing technologies is addressed. This achieves accurate segmentation of perforation location and size, thereby improving diagnostic accuracy and efficiency.

CN119048863BActive Publication Date: 2026-11-17NEUSOFT MEDICAL SYST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411181120.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-26
Publication Date
2026-11-17
Estimated Expiration
2044-08-26

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately identify the specific location and size of gastrointestinal perforations, and relying solely on the presence of free gas results in low accuracy, failing to provide sufficient diagnostic reference.

Method used

We employ both supervised and unsupervised segmentation methods for classification and segmentation. We train a gastrointestinal perforation segmentation model using pseudo-optimized labels and the Cosine distance loss function. Combined with data augmentation techniques, we form a semi-supervised model to improve the model's accuracy and robustness.

Benefits of technology

It enables accurate segmentation of gastrointestinal perforations, pinpointing their specific location and size, providing a reference for subsequent analysis, and improving the accuracy and efficiency of diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048863B_ABST
    Figure CN119048863B_ABST
Patent Text Reader

Abstract

This application relates to the field of medical image processing, and discloses a training method, recognition method, and apparatus for a gastrointestinal perforation segmentation model. The training method includes: obtaining predicted values ​​output by a network model based on labeled gastrointestinal perforation data and corresponding labels; obtaining pseudo-optimized labels corresponding to unlabeled gastrointestinal perforation data; wherein the different pseudo-optimized labels change in the same direction; and training the network model based on the predicted values ​​and pseudo-optimized labels to obtain a gastrointestinal perforation segmentation model. This method can identify the specific distribution location and characteristics of gastrointestinal perforations, and also improves the accuracy of gastrointestinal perforation recognition.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of medical image processing, and more specifically, to a training method, recognition method, and apparatus for a segmentation model of gastrointestinal perforation. Background Technology

[0002] Currently, the diagnosis of gastrointestinal perforation largely depends on imaging examinations. Accurate diagnosis of the presence, extent, and cause of perforation is crucial for appropriate patient management and surgical planning. With the continuous development of CT (Computed Tomography) imaging technology, it has become possible to examine the entire abdomen using thin-slice CT scans during a single breath-hold. Therefore, there is a need for a technique that uses CT images for rapid and accurate segmentation of gastrointestinal perforations to identify them.

[0003] To address this, a deep learning-based method for detecting gastrointestinal perforation has been disclosed. This method uses a network model to determine the presence of free gas in medical images to identify the presence of gastrointestinal perforation. Since the Henlein unit (HU) value of gas in medical images is lower than that of human tissue structures, setting the HU value can provide a preliminary assessment of the presence of free gas in the image and enable image annotation.

[0004] In the process of implementing the embodiments of this disclosure, at least the following problems were found in the related art: While related technologies can identify free gas, the network model outputs only 0s and 1s, failing to provide the specific distribution location of the free gas and thus offering limited reference value for subsequent diagnosis of gastrointestinal perforation. Furthermore, since the causes of perforation vary, the characteristics of gastrointestinal perforation also differ; relying solely on free gas to determine the presence of gastrointestinal perforation may lead to low accuracy.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] To provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. This summary is not intended as a general commentary, nor is it intended to identify key / important components or describe the scope of protection of these embodiments, but rather as a prelude to the detailed description that follows.

[0007] This disclosure provides a training method, recognition method, and apparatus for a gastrointestinal perforation segmentation model, which can improve the accuracy of gastrointestinal perforation recognition.

[0008] In some embodiments, the training method for the gastrointestinal perforation segmentation model includes: obtaining the predicted value output by the network model based on the gastrointestinal perforation labeled data and the labels corresponding to the gastrointestinal perforation labeled data; obtaining the pseudo-optimized labels corresponding to the unlabeled gastrointestinal perforation data based on the unlabeled gastrointestinal perforation data; wherein the different pseudo-optimized labels change in the same direction; and training the network model based on the predicted value and the pseudo-optimized labels to obtain the gastrointestinal perforation segmentation model.

[0009] Optionally, based on the unlabeled data of gastrointestinal perforation, pseudo-optimized labels are obtained, including: obtaining pseudo-initial labels output by the network model; obtaining voxels in the pseudo-initial labels; and assigning weights to the voxels in the pseudo-initial labels based on the confidence level of the pseudo-initial labels to obtain pseudo-optimized labels.

[0010] Optionally, based on the unlabeled data of gastrointestinal perforation, pseudo-optimized labels are obtained corresponding to the unlabeled data of gastrointestinal perforation, including: obtaining pseudo-initial labels output by the network model; calculating the similarity of multiple pseudo-initial labels according to the Cosine distance loss function; and applying regularization constraints to the multiple pseudo-initial labels based on the similarity to obtain pseudo-optimized labels.

[0011] Optionally, unlabeled data of gastrointestinal perforation can be obtained by performing various data augmentation transformations on labeled data of gastrointestinal perforation to obtain multiple unlabeled data of gastrointestinal perforation.

[0012] Optionally, the network model is trained based on the predicted values ​​and pseudo-optimized labels to obtain a gastrointestinal perforation segmentation model, including: combining the predicted values ​​with multiple pseudo-optimized labels to obtain multiple combined labels; using the multiple combined labels to optimize the network model with pseudo-supervised loss to obtain multiple training models; and selecting the training model with the best segmentation effect among the multiple training models as the gastrointestinal perforation segmentation model.

[0013] Optionally, multiple training models are obtained by optimizing the pseudo-supervised loss of the network model using multiple combined labels, including: calculating the combined loss of the predicted values ​​and pseudo labels in multiple combined labels according to the loss function; and updating the parameters of multiple network models according to the combined loss of the predicted values ​​and pseudo labels in multiple data sets to obtain multiple training models.

[0014] Optionally, the method for obtaining multiple training models further includes: calculating the loss of the predicted values ​​in multiple combined labels according to the loss function; and updating the parameters of the network model according to the loss of the predicted values ​​to obtain multiple training models.

[0015] In some embodiments, a method for identifying gastrointestinal perforation includes: acquiring a medical gastrointestinal image to be identified; inputting the medical gastrointestinal image to be identified into a gastrointestinal perforation segmentation model as described above to obtain a prediction result of gastrointestinal perforation; annotating the prediction result in the identified medical gastrointestinal image; and generating a diagnostic report based on the annotated medical gastrointestinal image.

[0016] In some embodiments, the 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, via the computer program, the training method for the segmentation model of gastrointestinal perforation described above, or the identification method for gastrointestinal perforation described above.

[0017] In some embodiments, the medical device includes: a medical device body; and, as described above, an identification device for gastrointestinal perforation, mounted on the medical device body.

[0018] The training method, recognition method, and apparatus for segmenting gastrointestinal perforations provided in this disclosure can achieve the following technical effects: In this embodiment, due to the various causes of gastrointestinal perforation and the different manifestations that perforations in different tissue structures may exhibit, a classification segmentation method combining supervised and unsupervised segmentation is employed. This allows for the segmentation of different manifestations of gastrointestinal perforation, providing a reference for subsequent analysis of possible causes. A portion of the data is labeled, including all possible manifestations of gastrointestinal perforation, such as medical images of different perforation locations within the abdominal cavity, forming a training set for the semi-supervised model. This allows the use of a large amount of unlabeled gastrointestinal perforation data, ensuring model accuracy with minimal labeling workload. Furthermore, image segmentation methods can be used to pinpoint the specific distribution location and size of gastrointestinal perforation manifestations, providing a reference for subsequent analysis of the severity of gastrointestinal perforation.

[0019] The above general description and the description below are exemplary and illustrative only and are not intended to limit this application. Attached Figure Description

[0020] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations and drawings do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are shown as similar elements. The drawings are not to be scaled. And wherein: Figure 1 This is a schematic diagram of the implementation environment for the training method of the digestive tract perforation segmentation model provided in this embodiment of the disclosure; Figure 2 This is a schematic diagram of a training method for a digestive tract perforation segmentation model provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of another training method for a gastrointestinal perforation segmentation model provided in an embodiment of this disclosure; Figure 4 This is a schematic diagram of another training method for a gastrointestinal perforation segmentation model provided in an embodiment of this disclosure; Figure 5 This is a schematic diagram of a method for identifying gastrointestinal perforation provided in an embodiment of this disclosure; Figure 6 This is a schematic diagram of a device for identifying gastrointestinal perforation provided in an embodiment of this disclosure; Figure 7 This is a schematic diagram of the basic structure of a deep learning-based method for detecting gastrointestinal perforation provided in an embodiment of this disclosure; Figure 8 This is a schematic diagram of a training module for a training method for a digestive tract perforation segmentation model provided in an embodiment of this disclosure; Figure 9 This is a CT enhanced image of a patient with duodenal perforation provided in an embodiment of this disclosure. Detailed Implementation

[0021] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.

[0022] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0023] Unless otherwise stated, the term "multiple" means two or more.

[0024] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0025] The term "and / or" describes an association 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" can refer to an association or binding relationship. The correspondence between A and B means that there is an association or binding relationship between A and B.

[0027] The digestive tract is a complex system in the human body responsible for the digestion and absorption of food. It begins in the mouth, passes through the pharynx, esophagus, stomach, small intestine, and large intestine, and finally reaches the anus. It includes not only digestive organs but also digestive juices secreted by digestive glands, such as saliva, gastric juice, and intestinal juice. Within the digestive tract, food is broken down into smaller molecules, making it easier for the body to absorb nutrients. The small intestine is the primary site of absorption, while the large intestine is mainly responsible for absorbing water and electrolytes, forming and storing waste products. The entire digestive tract propels food forward through the peristaltic movements of muscles, completing the digestion and absorption process.

[0028] Gastrointestinal tract perforation is a serious and urgent medical condition that occurs when the wall of the digestive tract (including the esophagus, stomach, and small intestine) ruptures, allowing contents to leak into the abdominal cavity or other cavities. This rupture is a surgical emergency requiring immediate intervention to prevent contamination of the peritoneal cavity; otherwise, it can lead to serious complications and even death. Causes of gastrointestinal perforation include peptic ulcers and the growth of tumors in the digestive tract that damage surrounding tissues.

[0029] The diagnosis of gastrointestinal perforation largely depends on imaging examinations. The challenges of segmenting gastrointestinal perforations lie in three main aspects: First, the perforation site occupies a very small proportion of all images, and the diameter of the hole is very small, making direct and accurate identification of the perforation highly unlikely. Second, the manifestations of perforation can vary depending on the organ's location. For example, gastrointestinal perforation may present as free gas in the abdominal cavity, ascites, intestinal wall thickening, increased mesenteric density, or mid-segment intestinal wall continuity. Therefore, labeling every image is time-consuming and laborious, limiting the number of labeled samples for training data. Third, the indistinct features of the lesion area further complicate gastrointestinal perforation identification. While some technologies aim to identify free gas in the abdominal cavity as a sign of gastrointestinal perforation, these are based on image classification and do not provide the specific location and size of the free gas or perforation, nor do they consider other features of gastrointestinal perforation, offering limited clinical diagnostic value. Finally, the small amount of gas is not readily apparent in images, leading to a high false negative rate in classification models. like Figure 9 As shown, the small amount of extracavitary air on the anterior surface of the liver peritoneum, in the middle of the abdomen, and between the peritoneal folds is easily overlooked.

[0030] In related technologies, considering that the perforation site is very small and difficult to find, most methods determine the presence of gastrointestinal perforation by judging whether there is free gas in medical images. For example... Figure 7 The diagram illustrates the basic algorithm structure of a deep learning-based method for detecting gastrointestinal perforation. Images are input into a deep classification network for training and testing. The network model is constructed through stacking operations such as convolution, pooling, and fully connected layers. It is then trained using a manually labeled dataset. The model outputs classification results, ultimately identifying gastrointestinal perforations. During training, all training data is input into the network model. The network model calculates a prediction result (0 or 1), indicating the presence or absence of free gas. The prediction result and its corresponding label are simultaneously substituted into the loss function to calculate the error value. This error value is backpropagated to each layer of the network model, updating the corresponding parameters to reduce the error. During testing, test data is input into the trained network model to obtain the network's prediction results. Based on the classification results, the system identifies whether the image contains free gas.

[0031] The embodiments disclosed herein aim to provide an accurate technique for separating gastrointestinal perforations. Figure 1 This is a schematic diagram illustrating the implementation environment of the training method for a gastrointestinal perforation segmentation model provided in this embodiment of the disclosure. Figure 1 As shown, the implementation environment of the above method may include processor 600 and database 604.

[0032] Database 604 stores a large number of medical images of the digestive tract, and a training dataset can be constructed through image annotation. Processor 600 can train a digestive tract perforation segmentation model based on the training dataset provided by database 604, thereby enabling the analysis of digestive tract perforations.

[0033] Combination Figure 2 As shown, this disclosure provides a training method for a gastrointestinal perforation segmentation model, including: S201, the processor obtains the predicted value output by the network model based on the gastrointestinal perforation marker data and the corresponding labels.

[0034] S202, the processor obtains the pseudo-optimized labels corresponding to the unlabeled data of gastrointestinal perforation based on the unlabeled data of gastrointestinal perforation; wherein, the different pseudo-optimized labels change in the same direction.

[0035] S203, the processor trains the network model based on the predicted values ​​and pseudo-optimized labels to obtain a digestive tract perforation segmentation model.

[0036] In the training method for a gastrointestinal perforation segmentation model provided in this disclosure, due to the various causes of gastrointestinal perforation and the different manifestations that perforations in different tissue structures may exhibit, a classification segmentation method combining supervised and unsupervised segmentation is employed. This allows for the segmentation of different manifestations of gastrointestinal perforation, providing a reference for subsequent analysis of possible causes of gastrointestinal perforation. A portion of the data is labeled, including all possible manifestations of gastrointestinal perforation, including medical images of different perforation locations within the abdominal cavity, forming a training set for the semi-supervised model. This allows the use of a large amount of unlabeled gastrointestinal perforation data, ensuring model accuracy with minimal labeling workload. Furthermore, image segmentation methods can be used to locate the specific distribution and size of gastrointestinal perforation manifestations, providing a reference for subsequent analysis of the severity of gastrointestinal perforation based on the perforation situation.

[0037] Combination Figure 8 As shown in this embodiment, the processor uses input module 80 to input labeled and unlabeled gastrointestinal perforation data into the network. The model is trained using network training module 81, which includes a supervised segmentation module 811, an unsupervised segmentation module 812, and a consistency constraint module 813. Finally, model generation module 82 selects the model with the best segmentation performance from the multiple trained models.

[0038] Optionally, based on the unlabeled data of gastrointestinal perforation, pseudo-optimized labels are obtained, including: obtaining pseudo-initial labels output by the network model; obtaining voxels in the pseudo-initial labels; and assigning weights to the voxels in the pseudo-initial labels based on the confidence level of the pseudo-initial labels to obtain pseudo-optimized labels.

[0039] Combination Figure 3 As shown, this disclosure provides another training method for a gastrointestinal perforation segmentation model, including: S301, the processor obtains the predicted value output by the network model based on the gastrointestinal perforation marker data and the corresponding labels.

[0040] S302, the processor obtains the pseudo-initial labels output by the network model.

[0041] S303, the processor acquires voxels from the pseudo-initial tags.

[0042] S304, the processor assigns weights to the voxels in the pseudo-initial label based on the confidence level of the pseudo-initial label, and obtains the pseudo-optimized label.

[0043] S305, the processor trains the network model based on the predicted values ​​and pseudo-optimized labels to obtain a digestive tract perforation segmentation model.

[0044] Pseudo-optimized labels refer to using the model's own predictions as labels when no real labels are available. It leverages information from unlabeled data, generates pseudo-labels by predicting data using the model, and then uses these pseudo-labels along with the real labels to further train the model, thereby improving its generalization ability and performance.

[0045] When generating pseudo-labels, the category with the highest predicted probability by the model is typically selected as the pseudo-label. Different strategies may be employed to ensure the reliability of the pseudo-labels, such as probability-based thresholding or ranking-based methods. Furthermore, a common problem with pseudo-labeling methods is the potential introduction of noise and errors; therefore, careful selection of pseudo-labels is necessary, and uncertainty estimation may be incorporated to improve their quality.

[0046] In practical applications, pseudo-label optimization can significantly improve the predictive performance of models on unlabeled data, especially when labeled data is scarce. For example, in image classification tasks, pseudo-label learning can help models learn richer feature representations, thereby improving classification accuracy. Furthermore, pseudo-label methods are also widely used in other fields such as text classification and recommender systems.

[0047] In this embodiment, the input to the network model is unlabeled data of gastrointestinal perforation. The network output is a pseudo-initial label. However, since the pseudo-initial label may contain potential noise, an uncertainty estimation is used to correct the pseudo-supervision process. The uncertainty estimation introduces adaptive voxel weights into the 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 based on the prediction confidence of voxels (the corresponding pixels in an image in three-dimensional space). Voxels with higher confidence have higher weights, while voxels with lower confidence have lower weights. By reducing the weight of voxels with low confidence, the negative impact caused by misprediction or noise can be reduced. This reduces the impact of label noise and improves the robustness of segmentation.

[0048] Optionally, based on the unlabeled data of gastrointestinal perforation, pseudo-optimized labels are obtained corresponding to the unlabeled data of gastrointestinal perforation, including: obtaining pseudo-initial labels output by the network model; calculating the similarity of multiple pseudo-initial labels according to the Cosine distance loss function; and applying regularization constraints to the multiple pseudo-initial labels based on the similarity to obtain pseudo-optimized labels.

[0049] Combination Figure 4 As shown, this disclosure provides another training method for a gastrointestinal perforation segmentation model, including: S401, the processor obtains the predicted value output by the network model based on the gastrointestinal perforation marker data and the corresponding labels.

[0050] S402, the processor obtains the pseudo-initial labels output by the network model.

[0051] S403, the processor calculates the similarity of multiple pseudo-initial labels based on the Cosine distance loss function.

[0052] S404, the processor applies regularization constraints to multiple pseudo-initial labels based on similarity to obtain pseudo-optimized labels.

[0053] S405, the processor trains the network model based on the predicted values ​​and pseudo-optimized labels to obtain a digestive tract perforation segmentation model.

[0054] In this embodiment, the Cosine distance loss function is used to minimize the dissimilarity between pseudo-initial labels, ensuring consistency in prediction results obtained from the same original image using different enhancement methods. Cosine distance loss is a loss function that measures the angular difference between two vectors, commonly used in machine learning and deep learning, especially in scenarios where vector direction rather than magnitude needs to be considered. The calculation of the Cosine loss function is based on the cosine similarity between vectors, with values ​​ranging from -1 to 1, where 1 indicates that the vectors are identical, 0 indicates that the vectors are orthogonal, and -1 indicates that the vectors are completely opposite.

[0055] The Cosine distance loss function can be calculated using the following formula:

[0056] Here, cosine distance loss is the cosine distance loss, and a and b are two vectors.

[0057] The cosine distance loss function is unaffected by vector length and focuses only on directional differences. In neural networks, cosine distance loss can be used in regularization layers to encourage the model output vector to be directionally consistent with the expected output vector, rather than simply reducing the Euclidean distance. For example, in the consistency constraints of a gastrointestinal perforation segmentation model, cosine distance loss might be used to ensure that predictions obtained through different data augmentation methods are directionally consistent, thereby improving the model's stability and robustness.

[0058] Optionally, unlabeled data of gastrointestinal perforation can be obtained by performing various data augmentation transformations on labeled data of gastrointestinal perforation to obtain multiple unlabeled data of gastrointestinal perforation.

[0059] In this embodiment of the disclosure, two different pixel-intensity-based data augmentation methods are applied to the input gastrointestinal perforation labeled data to obtain unlabeled gastrointestinal perforation data. The network input consists of the labeled and unlabeled gastrointestinal perforation data. The data augmentation methods include random intensity scaling, random intensity shifting, and Gaussian noise.

[0060] Optionally, the network model is trained based on the predicted values ​​and pseudo-optimized labels to obtain a gastrointestinal perforation segmentation model, including: combining the predicted values ​​with multiple pseudo-optimized labels to obtain multiple combined labels; using the multiple combined labels to optimize the network model with pseudo-supervised loss to obtain multiple training models; and selecting the training model with the best segmentation effect among the multiple training models as the gastrointestinal perforation segmentation model.

[0061] In this embodiment, pseudo-supervised loss optimization is a semi-supervised learning method that uses unlabeled data (pseudo-labels) to train the model while incorporating a small amount of labeled data. It is particularly suitable for situations where labeled data is scarce but unlabeled data is abundant.

[0062] The current model is used to predict unlabeled data related to gastrointestinal perforation, generating pseudo-labels. These pseudo-labels are then combined with the model's predictions to design a loss function to optimize the model. This function includes both supervised and pseudo-supervised losses. The uncertainty of the generated pseudo-labels is estimated to identify potentially noisy or erroneous labels. Different weights are assigned to each sample in the loss function based on the uncertainty estimate. From the multiple trained models, the one with the best segmentation performance is selected as the gastrointestinal perforation segmentation model.

[0063] Optionally, multiple training models are obtained by optimizing the pseudo-supervised loss of the network model using multiple combined labels, including: calculating the combined loss of the predicted values ​​and pseudo labels in multiple combined labels according to the loss function; and updating the parameters of multiple network models according to the combined loss of the predicted values ​​and pseudo labels in multiple data sets to obtain multiple training models.

[0064] In this embodiment, supervised loss and pseudo-supervised loss are combined to form a comprehensive loss function. This 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 trained model to learn from unlabeled data while improving the robustness of the trained model by combining predictions from multiple data sets. Through adaptive weight allocation, the trained model can focus more on regions that are easier to predict correctly, thereby improving overall segmentation accuracy and model performance.

[0065] Optionally, the method for obtaining multiple training models further includes: calculating the loss of the predicted values ​​in multiple combined labels according to the loss function; and updating the parameters of the network model according to the loss of the predicted values ​​to obtain multiple training models.

[0066] During training, an iterative optimization strategy is employed, repeatedly calculating loss and updating parameters until the model performance on the validation set no longer improves or reaches the predetermined number of iterations. Implementing this approach requires building a flexible training framework to adapt to different combinations of characteristics. First, the data is preprocessed and augmented to improve the model's adaptability to new situations. Then, a multi-task learning architecture is designed, allowing the model to learn from both labeled and unlabeled data simultaneously. In each iteration, the model uses cross-entropy loss and a custom consistency loss to optimize parameters, ensuring semantic consistency in predictions across different combinations. This allows the model to self-correct during training, gradually improving its understanding of complex data distributions. Finally, through ensemble learning or model selection strategies, the optimal solution is extracted from multiple trained models to achieve the best performance.

[0067] Combination Figure 5 As shown in the embodiments of this disclosure, a method for identifying gastrointestinal perforation is provided, comprising: S501, the processor acquires the 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, and obtains the prediction result of digestive tract perforation.

[0069] The S503 processor annotates the predicted results in the identified medical digestive tract images.

[0070] The S504 processor generates diagnostic reports based on labeled medical images of the digestive tract.

[0071] In this embodiment, the gastrointestinal perforation segmentation model obtained during the training phase is used to test images to obtain prediction results for each gastrointestinal perforation feature. The predicted multi-class segmentation results are visualized, and the segmentation results are displayed on the original image. This clearly shows the location distribution and size of the gastrointestinal perforation features. Different categories of gastrointestinal perforation features are displayed in different colors, providing doctors with more direct reference information for diagnosis in conjunction with clinical symptoms.

[0072] The diagnostic report includes whether gastrointestinal perforation exists, the cause of the perforation, the number of perforation sites, and the volume (area) of each site. The initial assessment of the perforation cause is based on the segmentation results. If the segmentation results show free gas or fluid accumulation, gastrointestinal perforation is considered. If the segmentation results also indicate a disease category such as tumor, the cause of the perforation is further considered to be a disease such as tumor. If the segmentation results also indicate a foreign body ingestion category, the cause of the perforation is considered to be foreign body ingestion. Furthermore, multiple causes of gastrointestinal perforation may be present simultaneously based on the segmentation results.

[0073] Combination Figure 6 As shown, this embodiment of the disclosure provides a device 60 for identifying gastrointestinal perforation, including a processor 600 and a memory 601. Optionally, the device 60 may further include a communication interface 602 and a bus 603. The processor 600, communication interface 602, and memory 601 can communicate with each other via the bus 603. The communication interface 602 can be used for information transmission. The processor 600 can call logical instructions in the memory 601 to execute the training method for the gastrointestinal perforation segmentation model of the above embodiment, or the above-described method for identifying gastrointestinal perforation.

[0074] Furthermore, the logic instructions in the aforementioned memory 601 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0075] The memory 601, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this disclosure. The processor 600 executes functional applications and data processing by running the program instructions / modules stored in the memory 601, that is, implementing the training method for the gastrointestinal perforation segmentation model in the above embodiments, or the above-described method for identifying gastrointestinal perforations.

[0076] The memory 601 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 601 may include high-speed random access memory and may also include non-volatile memory.

[0077] This disclosure provides a medical device, including a medical device body and the aforementioned identification device 60 for gastrointestinal perforation. The identification device 60 for gastrointestinal perforation is mounted on the medical device body. The mounting relationship described herein is not limited to placement within the medical device body, but also includes mounting connections with other components of the medical device, including but not limited to physical connections, electrical connections, or signal transmission connections. Those skilled in the art will understand that the identification device 60 for gastrointestinal perforation can be adapted to feasible medical device bodies to achieve other feasible embodiments.

[0078] The technical solutions of this disclosure can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes one or more instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in this disclosure. The aforementioned storage medium can be a non-transitory storage medium, such as a USB flash drive, external hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, etc., and other media capable of storing program code.

[0079] The foregoing description and accompanying drawings fully illustrate embodiments of this disclosure to enable those skilled in the art to practice them. Other embodiments may include structural, logical, electrical, procedural, and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Parts and features of some embodiments may be included in or replace parts and features of other embodiments. Moreover, the terminology used in this application is for describing embodiments only and is not intended to limit the claims. As used in the description of embodiments and claims, the singular forms “a,” “an,” and “the” are intended to equally include the plural forms unless the context clearly indicates otherwise. Similarly, the term “and / or” as used in this application means including one or more of the associated listed items and all possible combinations thereof. Additionally, when used in this application, the term "comprise" and its variations "comprises" and / or "comprising" refer to the presence of stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof. Without further limitations, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element. In this document, each embodiment may focus on the differences from other embodiments, and similar or identical parts between embodiments can be referred to mutually. For methods, products, etc., disclosed in the embodiments, if they correspond to the method section disclosed in the embodiments, the relevant parts can be referred to the description of the method section.

[0080] Those skilled in the art will recognize that the units and algorithm steps of the various examples 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 implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this disclosure. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0081] The methods and products disclosed in the embodiments herein (including but not limited to devices and equipment) can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units may be merely a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to implement this embodiment according to actual needs. In addition, the functional units in the embodiments of this disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0082] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

Claims

1. A training method for a gastrointestinal perforation segmentation model, characterized in that, include: Based on the gastrointestinal perforation marker data and the corresponding labels, the predicted values ​​output by the network model are obtained. Based on unlabeled data of gastrointestinal perforation, pseudo-optimized labels are obtained corresponding to the unlabeled data of gastrointestinal perforation; wherein, the change direction of different pseudo-optimized labels is consistent; the process of obtaining pseudo-optimized labels corresponding to unlabeled data of gastrointestinal perforation includes: obtaining pseudo-initial labels output by the network model; calculating the similarity of multiple pseudo-initial labels according to the Cosine distance loss function; and applying regularization constraints to multiple pseudo-initial labels based on the similarity to obtain pseudo-optimized labels. The network model is trained based on the predicted values ​​and pseudo-optimized labels to obtain a digestive tract perforation segmentation model.

2. The training method according to claim 1, characterized in that, Based on the unlabeled data of gastrointestinal perforation, obtain the pseudo-optimized labels corresponding to the unlabeled data of gastrointestinal perforation, including: Obtain the pseudo-initial labels output by the network model; Retrieve voxels from pseudo-initial tags; Based on the confidence level of the pseudo-initial labels, weights are assigned to the voxels in the pseudo-initial labels to obtain pseudo-optimized labels.

3. The training method according to claim 1, characterized in that, Unlabeled data on gastrointestinal perforations were obtained as follows: Multiple data augmentation transformations were applied to the labeled data of gastrointestinal perforation to obtain multiple unlabeled data of gastrointestinal perforation.

4. The training method according to claim 1, characterized in that, Based on the predicted values ​​and pseudo-optimized labels, the network model is trained to obtain a gastrointestinal perforation segmentation model, including: The predicted values ​​are combined with multiple pseudo-optimized labels to obtain multiple combined labels; Multiple training models were obtained by optimizing the pseudo-supervised loss of the network model using multiple combined labels. The training model with the best segmentation effect among multiple training models was selected as the gastrointestinal perforation segmentation model.

5. The training method according to claim 4, characterized in that, Multiple training models were obtained by optimizing the pseudo-supervised loss of the network model using multiple combined labels, including: Based on the loss function, calculate the combined loss of the predicted values ​​and pseudo-labels in multiple combined labels respectively; Based on the combined loss of predicted values ​​and pseudo-labels from multiple datasets, the parameters of multiple network models are updated to obtain multiple trained models.

6. The training method according to claim 4, characterized in that, Methods for obtaining multiple trained models also include: Based on the loss function, calculate the loss of the predicted values ​​in multiple combined labels respectively; Based on the loss of the predicted values, the parameters of the network model are updated respectively, resulting in multiple trained models.

7. A method for identifying gastrointestinal perforation, characterized in that, include: Acquire images of the medical digestive tract to be identified; The medical digestive tract image to be identified is input into the digestive tract perforation segmentation model as described in any one of claims 1 to 6 to obtain the prediction result of digestive tract perforation. Annotate the predicted results in the identified medical digestive tract images; A diagnostic report is generated based on labeled medical images of the digestive tract.

8. A device for identifying gastrointestinal perforation, comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute, through the computer program, the training method for a gastrointestinal perforation segmentation model as described in any one of claims 1 to 6, or the identification method for gastrointestinal perforation as described in claim 7.

9. A medical device, characterized in that, include: The medical device itself; and, The identification device for gastrointestinal perforation as described in claim 8 is installed on the main body of the medical device.