Pediatric respiratory disease diagnosis system and method based on pattern recognition

Abstract

The invention provides a pediatric respiratory disease diagnosis system and method based on pattern recognition. The pediatric respiratory disease diagnosis method comprises the following steps: constructing a multi-modal pathological space by combining pathophysiological features of children in various medical institutions; performing breathing mode matching on the breathing time sequence data of the target patient and the subtype distribution of the breathing disease in the multi-mode pathological space to obtain pathological state constraints, and determining a pathological association relationship between the target patient and the known pediatric breathing disease according to the pathological state constraints and gray texture features of the lung medical image of the target patient; constructing a disease progress evolution graph of the pediatric respiratory disease; and carrying out map fusion on the disease progress evolution map and the pathology association relationship to obtain a multi-dimensional decision index of the respiratory disease of the target patient, and carrying out multi-modal identification on the respiratory disease of the target patient based on the multi-dimensional decision index to obtain a disease auxiliary identification report. By adopting the scheme of the invention, a comprehensive diagnosis system considering multi-center feature alignment and multi-modal information fusion can be constructed to identify pediatric respiratory diseases.

Description

Technical Field

[0001] The present application relates to the technical field of pattern recognition, and more specifically, to a pediatric respiratory disease diagnosis system and method based on pattern recognition. Background Art

[0002] The incidence of pediatric respiratory diseases ranks first among pediatric diseases. Common respiratory diseases such as acute upper and lower respiratory tract infections, bronchitis and asthma pose a serious threat to children's growth, development and life and health. Early diagnosis is crucial to reducing mortality and improving prognosis. With the development of wearable sensing technology, medical image analysis and machine learning, pattern recognition-based methods have begun to be applied to lung sound signal processing and image texture analysis, providing a new diagnostic approach for early screening of pediatric respiratory diseases.

[0003] Existing technologies mainly use machine learning or deep learning models based on feature engineering to classify and diagnose the time-frequency features of lung sound signals, or use gray-level co-occurrence matrices and convolutional neural networks to detect lesions in lung medical images. However, most of these methods focus on single-center data or a single modality (such as relying solely on lung sound signals or images) and are constrained by the redundancy or lack of single-modality information. At the same time, the heterogeneity of data formats and clinical practices between different medical institutions further exacerbates the difficulty of cross-center model deployment and limits the system's widespread application in actual pediatric scenarios. Therefore, how to build a comprehensive diagnostic system that takes into account multi-center feature alignment and multimodal information fusion to identify pediatric respiratory diseases has become a difficult problem facing the industry. Summary of the Invention

[0004] The present application provides a pediatric respiratory disease diagnosis system and method based on pattern recognition, which can construct a comprehensive diagnostic system that takes into account multi-center feature alignment and multimodal information fusion to identify pediatric respiratory diseases.

[0005] In a first aspect, the present application provides a method for assisting in identifying pediatric respiratory diseases, which is used in a pediatric respiratory disease diagnosis system to assist in identifying children's respiratory diseases. The method comprises: Acquire respiratory time series data consisting of lung sound signals of the target patient; A multimodal pathology space for pediatric respiratory diseases was constructed by combining the pathophysiological characteristics of children with respiratory diseases in various medical institutions through a comparative learning model. Performing respiratory pattern matching on the respiratory time series data with the subtype distribution of respiratory diseases in the multimodal pathology space, thereby obtaining a pathological state constraint that matches the current pathological state of the target patient, and determining a pathological correlation between the pathological state of the target patient and known pediatric respiratory diseases based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image; constructing a disease progression evolution map of pediatric respiratory diseases by using a co-occurrence probability matrix of disease subtypes in the multimodal pathology space; The cross-attention mechanism is used to fuse the disease progression evolution map and the pathological correlation relationship, and then the multidimensional decision indicators of the target patient's respiratory disease are extracted. Based on the multidimensional decision indicators, the target patient's respiratory disease is multimodally identified to obtain a disease-assisted identification report for the target patient.

[0006] In some embodiments, constructing a multimodal pathology space for pediatric respiratory diseases by combining the pathophysiological characteristics of children with respiratory diseases in various medical institutions through a comparative learning model specifically includes: For each medical institution in the distributed medical network, access pediatric respiratory disease data and lung medical imaging archives in the medical institution's local database; extracting lung sound spectrum features of respiratory diseases from the pediatric respiratory disease data; Performing feature extraction on all images in the lung medical image archive to obtain image texture features of the lung medical images; Performing dimensionality reduction and fusion on the lung sound spectrum features and the image texture features to obtain feature basis vectors of pediatric respiratory diseases in medical institutions, and further obtaining feature basis vectors of pediatric respiratory diseases in various medical institutions; All feature basis vectors are aligned using a contrastive learning model to generate a multimodal pathology space for pediatric respiratory diseases.

[0007] In some embodiments, performing feature alignment on all feature basis vectors using a contrastive learning model to generate a multimodal pathology space for pediatric respiratory diseases specifically includes: Perform multimodal data augmentation on each feature basis vector to generate enhanced positive sample pairs and heterogeneous negative sample pairs; Training a contrastive learning model by using the enhanced positive sample pairs and the heterogeneous negative sample pairs; Generating a multimodal pathology space for pediatric respiratory diseases using a trained contrastive learning model.

[0008] In some embodiments, performing respiratory pattern matching on the respiratory time series data and the subtype distribution of respiratory diseases in the multimodal pathology space to obtain a pathological state constraint that matches the current pathological state of the target patient specifically includes: Extracting features from the respiratory time series data to obtain respiratory features of the target patient; obtaining a subtype distribution of respiratory diseases in the multimodal pathology space; performing pattern matching on the respiratory characteristics and the subtype distribution using a dynamic time programming algorithm to obtain a matching confidence between the target patient's respiratory characteristics and each disease subtype; A pathological state constraint matching the current pathological state of the target patient is determined based on all matching confidences.

[0009] In some embodiments, determining the pathological correlation between the target patient's pathological state and known pediatric respiratory diseases based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image specifically includes: Obtain lung medical images of target patients; Extracting grayscale texture features of the target patient's lung medical image from the lung medical image; Performing multimodal feature fusion on the grayscale texture feature and the pathological state constraint to obtain a multimodal feature representation of the target patient's respiratory disease; A pathological correlation relationship between the target patient's pathological state and known pediatric respiratory diseases is established based on the multimodal feature representation.

[0010] In some embodiments, constructing a disease progression evolution map of pediatric respiratory diseases using a co-occurrence probability matrix of disease subtypes in the multimodal pathology space specifically includes: Extracting the transfer frequencies between disease subtypes in the multimodal pathology space and generating a co-occurrence tensor of disease subtypes; Performing probability modeling on the co-occurrence tensor to obtain a co-occurrence probability matrix of disease subtypes in the multimodal pathology space; Dynamic heterogeneous graph mapping is performed based on the co-occurrence probability matrix to obtain a disease progression evolution map of pediatric respiratory diseases.

[0011] In some embodiments, the cross-attention mechanism is used to fuse the disease progression evolution graph and the pathological association relationship to extract the multidimensional decision indicators of the target patient's respiratory disease, specifically including: Obtaining node embedding representations in the disease progression evolution graph and the pathological association relationship respectively; Determining the graph relevance between the disease progression graph and the pathological association relationship based on all node embedding representations through a cross-attention mechanism; Based on the graph correlation, feature fusion is performed on the nodes in the disease progression evolution graph and the pathological association relationship to obtain an enhanced disease progression evolution graph of the target patient's respiratory disease; A multidimensional decision indicator for the respiratory disease of the target patient is generated based on the fused nodes in the enhanced disease progression evolution map.

[0012] In a second aspect, the present application provides a pediatric respiratory disease diagnosis system based on pattern recognition, including a disease auxiliary identification unit, wherein the disease auxiliary identification unit includes: An acquisition module, used to acquire respiratory time series data composed of lung sound signals of a target patient; A processing module is used to construct a multimodal pathology space for pediatric respiratory diseases by combining the pathophysiological characteristics of children with respiratory diseases in various medical institutions through a comparative learning model; The processing module is configured to perform respiratory pattern matching on the respiratory time series data and the subtype distribution of respiratory diseases in the multimodal pathological space, thereby obtaining a pathological state constraint that matches the current pathological state of the target patient, and determining a pathological correlation between the pathological state of the target patient and known pediatric respiratory diseases based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image; The processing module is configured to construct a disease progression evolution map of pediatric respiratory diseases using a co-occurrence probability matrix of disease subtypes in the multimodal pathology space; An execution module is used to use a cross-attention mechanism to fuse the disease progression evolution map and the pathological association relationship, thereby extracting multidimensional decision indicators of the target patient's respiratory disease, performing multimodal identification of the target patient's respiratory disease based on the multidimensional decision indicators, and obtaining a disease-assisted identification report for the target patient.

[0013] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a code, and the processor is configured to obtain the code and execute the above-mentioned auxiliary identification method for pediatric respiratory diseases.

[0014] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned auxiliary identification method for pediatric respiratory diseases.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In the pediatric respiratory disease diagnosis system and method based on pattern recognition provided by the present application, respiratory time series data consisting of lung sound signals of the target patient are first obtained; a multimodal pathological space of pediatric respiratory diseases is constructed by combining the pathophysiological characteristics of children suffering from respiratory diseases in various medical institutions through a comparative learning model; the respiratory time series data is matched with the subtype distribution of respiratory diseases in the multimodal pathological space for respiratory pattern matching, thereby obtaining a pathological state constraint that matches the current pathological state of the target patient, and the pathological correlation relationship between the pathological state of the target patient and known pediatric respiratory diseases is determined based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image; a disease progression evolution map of pediatric respiratory diseases is constructed through the co-occurrence probability matrix of disease subtypes in the multimodal pathological space; the disease progression evolution map and the pathological correlation relationship are fused using a cross-attention mechanism, thereby extracting a multidimensional decision indicator of the target patient's respiratory disease, and multimodally identifying the target patient's respiratory disease based on the multidimensional decision indicator to obtain a disease auxiliary identification report for the target patient.

[0016] It can be seen that the present application performs multimodal identification of respiratory diseases of target patients based on the multidimensional decision-making indicators to obtain an auxiliary disease identification report for the target patients; first, the multimodal pathological space is determined to obtain a multidimensional embedding space for feature alignment and fusion of pediatric respiratory disease data from different medical institutions. The multimodal pathological space is determined through horizontal federated learning and domain adaptation technology to vectorize and align the lung sound signals, image features and electronic medical records extracted locally by each medical institution to a shared high-dimensional space, fundamentally eliminating equipment differences and process heterogeneity problems, and helping to achieve consistent representation of cross-institutional data, thereby improving disease identification and analysis capabilities under multi-center collaboration; then, the pathological correlation relationship is determined to obtain a relationship map of the intrinsic correlation between the multimodal features characterizing the target patient's respiratory disease and the known pathological mechanisms of respiratory diseases. The confirmation of the pathological correlation relationship Determination helps to reveal the potential associations between different disease subtypes and imaging manifestations and respiratory patterns, as well as the relationships between different pathological mechanisms, thereby promoting a deeper understanding of the pathological mechanisms of pediatric respiratory diseases, providing personalized diagnostic identification for target patients, and improving the accuracy and reliability of diagnosis; finally, determining the multidimensional decision-making index can obtain a vector that characterizes the respiratory disease type and pathological evolution path of the target patient. The determination of the multidimensional decision-making index can provide accurate and interpretable input for the diagnostic model, and support fine-grained output of different diagnostic indicators (such as subtype probability and severity score), thereby improving the classification accuracy and interpretability of the diagnostic model, thereby significantly improving the accuracy of pediatric respiratory disease identification and the credibility of the decision; in summary, based on the above scheme, a comprehensive diagnostic system that takes into account multi-center feature alignment and multimodal information fusion can be constructed to identify pediatric respiratory diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is an exemplary flow chart of a method for assisting in identifying pediatric respiratory diseases according to some embodiments of the present application; Figure 2 is an operational flow chart for constructing a multimodal pathology space according to some embodiments of the present application; Figure 3 is an exemplary flow chart for determining pathological association relationships according to some embodiments of the present application; Figure 4 is a schematic structural diagram of a disease auxiliary identification unit according to some embodiments of the present application; Figure 5 This is a diagram of the internal structure of a computer device for implementing an auxiliary identification method for pediatric respiratory diseases according to some embodiments of the present application. DETAILED DESCRIPTION

[0018] In order to better understand the technical solution of the present application, the technical solution of the present application will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0019] refer to Figure 1 , which is an exemplary flow chart of a method for assisting in identifying pediatric respiratory diseases according to some embodiments of the present application. The method 100 for assisting in identifying pediatric respiratory diseases mainly includes the following steps: In step 101, respiratory time series data consisting of lung sound signals of a target patient is acquired.

[0020] It should be noted that, in the present application, the respiratory timing data is a data set that records the lung sound signals of the target patient, and the respiratory timing data can dynamically reflect the changes in the target patient's breathing pattern. In specific implementation, the respiratory timing data composed of the target patient's lung sound signals can be obtained in the following manner, namely: an electronic stethoscope (such as Thinklabs One or Littmann 3200) can be used to collect the lung sound signals of the target patient within a preset collection time period (such as half an hour), and the set of all collected lung sound signals is used as the respiratory timing data. In specific implementation, the collected lung sound signals may contain noise, and the collected lung sound signals can be denoised by wavelet transform to extract the respiratory timing data, which will not be repeated here.

[0021] In step 102, a multimodal pathological space of pediatric respiratory diseases is constructed by combining the pathophysiological characteristics of children suffering from respiratory diseases in various medical institutions through a comparative learning model.

[0022] In some embodiments, reference Figure 2This figure is an operational flow chart for constructing a multimodal pathology space according to some embodiments of the present application. In this application, a multimodal pathology space for pediatric respiratory diseases is constructed by combining a comparative learning model with the pathophysiological characteristics of children suffering from respiratory diseases in various medical institutions. The following steps can be used: For each medical institution in the distributed medical network, access pediatric respiratory disease data and lung medical imaging archives in the medical institution's local database; extracting pathophysiological characteristics of children suffering from respiratory diseases, namely, lung sound spectrum characteristics, from the pediatric respiratory disease data; Extracting pathophysiological features of all images in the lung medical image archive to obtain image texture features of the lung medical images; Performing dimensionality reduction and fusion on the lung sound spectrum features and the image texture features to obtain feature basis vectors of pediatric respiratory diseases in medical institutions, and further obtaining feature basis vectors of pediatric respiratory diseases in various medical institutions; All feature basis vectors are aligned using a contrastive learning model to generate a multimodal pathology space for pediatric respiratory diseases.

[0023] It should be noted that in this application, the distributed medical network is a cross-domain data and service collaborative architecture built by interconnecting multi-source medical nodes such as hospitals, clinics and laboratories through distributed computing and network technologies in the form of microservices, message queues and secure channels to break information silos and achieve seamless integration between heterogeneous systems; this distributed medical network can realize the secure sharing and synchronous storage of medical data between medical institutions, support the collaborative training and online updating of multi-center diagnostic models, and provide real-time remote monitoring, online consultation and emergency dispatch services, which greatly improves the diagnosis and treatment efficiency and resource utilization, and provides a solid network foundation for the multi-center, multi-modal fusion diagnosis of this solution.

[0024] In specific implementation, access to pediatric respiratory disease data and lung medical imaging archives in the local database of a medical institution can be achieved in the following manner, namely: the local database of the medical institution can be accessed through an existing application programming interface (such as the application programming interface of the Rapid Medical Interoperability Resources standard) to obtain structured data and lung medical imaging archives of pediatric respiratory disease patients; wherein, the pediatric respiratory disease data refers to a standardized format data set in the medical institution that records information related to pediatric respiratory diseases, including basic information of the child (such as age and gender), preliminary diagnosis results and lung sound signals. The pediatric respiratory disease data provides systematic data support for the identification and diagnosis of pediatric respiratory diseases; the lung medical imaging archive refers to a collection of lung images of pediatric respiratory diseases stored in the medical institution. The lung medical imaging archive can be used to assist in the diagnosis and assessment of the severity and scope of respiratory diseases.

[0025] It should be noted that, in the present application, the pathophysiological characteristics of children suffering from respiratory diseases include the lung sound spectrum characteristics during breathing and the image texture characteristics of lung medical images; wherein, the lung sound spectrum characteristics are key acoustic features reflecting the pathological state of the respiratory system of children in medical institutions, and the lung sound spectrum characteristics can be used to construct a disease recognition model, thereby improving the accuracy of diagnosis and early recognition capabilities of pediatric respiratory diseases; the image texture features are multidimensional vectors in lung medical images that reflect the lung lesion characteristics of pediatric respiratory diseases, and the image texture features can provide strong support for the diagnosis and treatment of pediatric respiratory diseases.

[0026] In a specific implementation, the extraction of the lung sound spectrum characteristics of respiratory diseases in the pediatric respiratory disease data can be achieved in the following manner, namely: all lung sound signals in the pediatric respiratory disease data can be denoised by existing audio signal processing technologies (such as bandpass filters and short-time Fourier transforms) and the time domain signals can be converted into frequency domain signals to generate a time-frequency spectrum diagram of the lung sounds of pediatric respiratory diseases. Then, a convolutional neural network (such as ResNet18) can be used to extract features from the time-frequency spectrum diagram to obtain a feature representation of the lung sounds. The feature representation is then fused with the corresponding basic information of the child and the preliminary diagnosis results in the pediatric respiratory disease data into a multidimensional feature vector. Finally, the existing dimensionality reduction technology (such as principal component analysis technology) is used to reduce the dimensionality of the multidimensional feature vector and retain the main information to obtain the lung sound spectrum characteristics of the respiratory disease.

[0027] In a specific implementation, feature extraction is performed on all images in the lung medical image archive to obtain the image texture features of the lung medical image. This can be achieved in the following manner: for each image in the lung medical image archive, the image is first segmented into lung regions using an existing segmentation tool (such as ITK-SNAP or Mimics) and a three-dimensional volume of interest is generated. Then, the texture features of the volume of interest are calculated using a gray-level co-occurrence matrix to obtain the texture features of the image. The texture features of each image in the lung medical image archive can be obtained through the above steps. Finally, a multidimensional feature vector composed of the texture features of all images is used as the image texture features of the lung medical image. The gray-level co-occurrence matrix can extract a variety of texture features (such as energy, contrast, correlation, entropy, and roughness) in the lung medical image, thereby comprehensively describing the texture information of the lung medical image.

[0028] It should be noted that, in the present application, the characteristic basis vector is a high-dimensional vector representation of the multimodal physiological data of pediatric respiratory diseases in medical institutions. The characteristic basis vector can comprehensively and compactly reflect the physiological state of children when they are sick. In specific implementation, the lung sound spectrum characteristics and the image texture characteristics are subjected to dimensionality reduction fusion to obtain the characteristic basis vectors of pediatric respiratory diseases in medical institutions. This can be achieved in the following way, namely: first, a linear dimensionality reduction method (such as principal component analysis) can be used to reduce the dimensionality of the lung sound spectrum characteristics to obtain the key characteristic components of the lung sound spectrum characteristics. Then, a nonlinear dimensionality reduction method (such as a kernel function based on the Riemann measure) can be used. The principal component analysis method (PCA) is used to map the texture features of lung medical images to a positive definite symmetric matrix space to form a Riemann manifold and perform dimensionality reduction processing to obtain the key feature components of the texture features of lung medical images. Finally, the existing deep learning network model (such as the encoding-decoding structure model) is used to perform multi-scale feature fusion on the key feature components of the lung sound spectrum features and the lung medical image texture features, and the fused feature vector is used as the feature basis vector of pediatric respiratory diseases in medical institutions; among them, the use of deep learning network models to extract and fuse features of different scales can alleviate the feature inconsistency problem caused by direct fusion and improve the fusion effect.

[0029] It should be noted that in this application, the fusion of image texture features and lung sound spectrum features enhances the multi-dimensional perception ability of pediatric respiratory diseases, which can improve the accuracy of disease identification and the sensitivity of early diagnosis. The construction of feature basis vectors enables pediatric respiratory disease data from different medical institutions to be aligned in a unified feature space, enhancing the comparability and fusion of cross-institutional data, and providing a solid data foundation for subsequent pathological status analysis, subtype identification and multi-dimensional decision-making.

[0030] Preferably, in some embodiments, the following steps may be used to perform feature alignment on all feature basis vectors using a contrastive learning model to generate a multimodal pathology space for pediatric respiratory diseases: Perform multimodal data augmentation on each feature basis vector to generate enhanced positive sample pairs and heterogeneous negative sample pairs; Training a contrastive learning model by using the enhanced positive sample pairs and the heterogeneous negative sample pairs; Generating a multimodal pathology space for pediatric respiratory diseases using a trained contrastive learning model.

[0031] In specific implementation, multimodal data enhancement is performed on each feature basis vector to generate enhanced positive sample pairs and heterogeneous negative sample pairs. This can be achieved in the following way: first, for each feature basis vector, a data enhancement method (such as random cropping, rotation, and color jittering, etc.) can be used to perform data enhancement on the feature basis vector and generate enhanced positive sample pairs. The positive sample pairs of each feature basis vector can be obtained through the above steps, and then the set of all positive sample pairs is used as the enhanced positive sample pairs. Then, the vector composed of the feature basis vectors of all medical institutions is used as the heterogeneous negative sample pairs; wherein, the enhanced positive sample pairs are sample pairs composed of multiple semantically consistent samples generated by data enhancement of the original samples. The enhanced positive sample pairs can be used to train the contrastive learning model so that the contrastive learning model can Learning to maintain consistent feature representations under different transformations effectively improves the generalization and robustness of the contrastive learning model. Especially in multi-center medical data, enhancing positive sample pairs helps the contrastive learning model to achieve feature alignment between different institutions and build a unified multimodal pathology space, thereby improving the recognition accuracy and reliability of pediatric respiratory diseases; the heterogeneous negative sample pairs are sample pairs composed of original samples from different institutions and semantically different in contrastive learning. The heterogeneous negative sample pairs can be used to train the contrastive learning model to distinguish feature representations of different categories, which can effectively enhance the discrimination ability of the contrastive learning model. Especially in multi-center medical data, by distinguishing the feature differences between different institutions, the generalization and robustness of the model can be improved, thereby building a unified multimodal pathology space.

[0032] In specific implementation, the contrastive learning model can be trained by the enhanced positive sample pairs and the heterogeneous negative sample pairs in the following manner, namely: a contrastive learning model (such as SimCLR, MoCo or CaCo) can be initialized, and the enhanced positive sample pairs and the heterogeneous negative sample pairs are input into the contrastive learning model as training samples of the contrastive learning model, and the contrastive learning model is trained using a contrastive loss function (such as NT-Xent) to obtain a trained contrastive learning model.

[0033] It should be noted that in the contrastive learning model, enhanced positive sample pairs are used to optimize the model so that the feature basis vectors of the same pathology category are as close as possible in the embedding space, while heterogeneous negative sample pairs are used to widen the distance between the features of different pathology categories, thereby improving the model's discrimination ability. In this way, the contrastive learning model can learn richer and more robust feature representations, which helps to achieve feature alignment in multi-center medical data and build a unified multimodal pathology space. The use of contrastive loss functions (such as NT-Xent) for model training can make the feature distance of positive sample pairs closer and the feature distance of negative sample pairs farther. In this way, the contrastive learning model can learn the common features between different medical institutions and the feature differences between data from different medical institutions.

[0034] It should be noted that, in this application, the multimodal pathology space is a multidimensional embedding space for feature alignment and fusion of pediatric respiratory disease data from different medical institutions. The multimodal pathology space extracts the common and specific features of pediatric respiratory disease data from various medical institutions through contrastive learning and other technologies, including: the subtype distribution of respiratory diseases and the co-occurrence probability matrix of disease subtypes, achieving consistent representation of cross-institutional data, thereby improving disease identification and analysis capabilities under multi-center collaboration; in specific implementation, the generation of a multimodal pathology space for pediatric respiratory diseases using a trained contrastive learning model can be achieved in the following way, namely: first, the target data can be obtained in the local database. The pediatric respiratory disease data of the target patient's medical institution (such as lung medical imaging archives and pediatric respiratory disease data) are then input into the trained contrastive learning model. The contrastive learning model is run to extract the positional feature vectors of the data of each medical institution in a unified embedding space, and all the positional feature vectors are clustered to identify different pathological feature patterns and disease subtypes. All pathological feature patterns and disease subtypes are then mapped back to the pediatric respiratory disease data of the target patient's medical institution to construct a unified feature space. Finally, the feature space in the output of the contrastive learning model is used as the multimodal pathological space of pediatric respiratory diseases.

[0035] In step 103, the respiratory time series data is matched with the subtype distribution of respiratory diseases in the multimodal pathological space for respiratory pattern matching, thereby obtaining a pathological state constraint that matches the current pathological state of the target patient. The pathological correlation between the pathological state of the target patient and known pediatric respiratory diseases is determined based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image.

[0036] In some embodiments, performing respiratory pattern matching on the respiratory time series data and the subtype distribution of respiratory diseases in the multimodal pathology space to obtain a pathological state constraint that matches the current pathological state of the target patient can be achieved by the following steps: Extracting features from the respiratory time series data to obtain respiratory features of the target patient; obtaining a subtype distribution of respiratory diseases in the multimodal pathology space; performing pattern matching on the respiratory characteristics and the subtype distribution using a dynamic time programming algorithm to obtain a matching confidence between the target patient's respiratory characteristics and each disease subtype; A pathological state constraint matching the current pathological state of the target patient is determined based on all matching confidences.

[0037] In specific implementation, feature extraction is performed on the respiratory time series data to obtain the respiratory characteristics of the target patient, which can be achieved in the following manner: first, the respiratory time series data can be denoised and the time domain signal can be converted into a frequency domain signal through existing audio signal processing technologies (such as bandpass filters and short-time Fourier transforms) to generate a time-frequency spectrum of the target patient's lung sound signal. Then, a convolutional neural network (such as ResNet18) can be used to extract features from the time-frequency spectrum and construct a multidimensional respiratory feature vector, thereby obtaining the respiratory characteristics of the target patient; wherein, the respiratory characteristics are audio features that characterize the respiratory function status and respiratory regularity of the target patient, and the respiratory characteristics can realize the monitoring and evaluation of the target patient's respiratory condition.

[0038] It should be noted that in this application, subtype distribution refers to the probability distribution of different subtypes under the same disease category in pediatric respiratory diseases. This subtype distribution helps to reveal the heterogeneity of the same disease in different children and provide a basis for personalized diagnosis and treatment. By identifying the disease subtype that best matches the respiratory characteristics of the target patient, more precise pathological state constraints can be formulated for it, thereby improving the accuracy of diagnosis and the effectiveness of treatment.

[0039] In specific implementation, a dynamic time planning algorithm is used to perform pattern matching on the respiratory characteristics and the subtype distribution, and the matching confidence between the target patient's respiratory characteristics and each disease subtype is obtained in the following manner, namely: for each disease subtype in the multimodal pathology space, the optimal path distance between the target patient's respiratory characteristics and the disease subtype can be calculated by a dynamic time warping algorithm, and then the optimal path distance is mapped to between 0 and 1 by an existing normalization formula (such as the matching confidence equals the inverse of the sum of 1 and the optimal path distance) to obtain the matching confidence between the respiratory characteristics and the disease subtype. The matching confidence between the target patient's respiratory characteristics and each disease subtype can be obtained through the above steps; wherein, the matching confidence refers to the degree of similarity between the target patient's respiratory characteristics and the disease subtype characteristics. The matching confidence can help identify the disease subtype that best matches the target patient's characteristics, provide a basis for personalized pathological state constraints, and assist in accurate diagnosis.

[0040] It should be noted that, in the present application, the pathological state constraint refers to a set of disease subtypes that match the current pathological state of the patient. The pathological state constraint can identify the subtype that best matches the current pathological state of the target patient and provide accurate diagnostic information for the clinic. In specific implementation, the pathological state constraint that matches the current pathological state of the target patient is determined based on all matching confidences. This can be achieved in the following way, namely: for each disease subtype in the multimodal pathological space, if the matching confidence of the disease subtype exceeds a preset similarity threshold (such as 0.8), the subtype disease subtype is marked as a matching disease subtype related to the pathological state of the target patient. Through the above steps, all matching disease subtypes in the multimodal pathological space can be obtained, and then the set of all matching disease subtypes is used as the pathological state constraint that matches the current pathological state of the target patient.

[0041] In some embodiments, reference Figure 3 This figure is an exemplary flow chart for determining a pathological association relationship according to some embodiments of the present application. In the present application, the pathological association relationship between the target patient's pathological state and known pediatric respiratory diseases can be determined based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image. The following steps can be used: In step 1031 , a lung medical image of a target patient is acquired; In step 1032, grayscale texture features of the lung medical image of the target patient are extracted from the lung medical image; In step 1033, the grayscale texture features and the pathological state constraints are subjected to multimodal feature fusion to obtain a multimodal feature representation of the target patient's respiratory disease; In step 1034 , a pathological association relationship between the target patient's pathological state and known pediatric respiratory diseases is established based on the multimodal feature representation.

[0042] In specific implementation, obtaining the target patient's lung medical image can be achieved in the following manner, namely: accessing the local database of the target patient's medical institution and obtaining the target patient's lung medical image in the lung medical image archive; wherein, the lung medical image is an image that characterizes the target patient's lung structure and functional status, and the lung medical image can be used to assist in diagnosis, monitor disease progression and evaluate treatment effects. In this solution, the lung medical image provides detailed structural information of the target patient's lung, and combined with grayscale texture features, it can reveal the potential pathological changes of the target patient's respiratory disease.

[0043] In a specific implementation, the grayscale texture features of the target patient's lung medical image can be extracted from the lung medical image in the following manner: first, the lung region of the lung medical image is segmented using an existing segmentation tool (such as ITK-SNAP or Mimics) to generate a three-dimensional volume of interest; then, the texture features of the volume of interest are calculated using a grayscale co-occurrence matrix to obtain the grayscale texture features of the lung medical image; wherein the grayscale texture features are multidimensional feature vectors extracted from the spatial distribution of the target patient's lung medical image, reflecting the characteristics of the target patient's respiratory disease lesions. The grayscale texture features can reveal the microstructural characteristics of the lesion area and provide key data support for subsequent pathological mechanism mapping.

[0044] In specific implementation, the grayscale texture features and the pathological state constraints are subjected to multimodal feature fusion to obtain a multimodal feature representation of the target patient's respiratory disease. This can be achieved in the following manner, namely: the grayscale texture features and the pathological state features can be fused using an existing feature fusion method (such as a feature weighted fusion method) to obtain a unified multimodal feature representation; wherein, the multimodal feature representation refers to a feature representation that integrates different modal information in the feature space for analysis and modeling. In this scheme, constructing multimodal features can comprehensively capture the structural and functional characteristics of respiratory diseases and improve the ability to recognize complex pathological mechanisms; the feature weighted fusion method can assign different weights to features from different sources to reflect their importance in the final decision. The weights can be set through expert experience or automatically learned through training data. This fusion method can highlight features that contribute more to identification and diagnosis and suppress the influence of noise or redundant information.

[0045] It should be noted that, in the present application, pathological correlation refers to a relationship map that characterizes the intrinsic correlation between the multimodal features of the target patient's respiratory disease and the known pathological mechanisms of respiratory diseases. The relationship between the nodes in the pathological correlation can accurately identify the correspondence between the target patient's respiratory disease and the specific respiratory pathological mechanism, and reveal the mutual relationship between different pathological mechanisms, promote an in-depth understanding of the pathological mechanism of respiratory diseases, thereby providing personalized diagnosis and identification for the target patient and improving the accuracy and reliability of diagnosis; in specific implementation, the pathological correlation between the target patient's pathological state and the known pediatric respiratory disease based on the multimodal feature representation can be achieved in the following way, namely: existing Some graph neural network models (such as graph convolutional network models) perform graph structure modeling, graph convolution, and learning mapping relationships between nodes based on this multimodal feature representation, and generate pathological correlation relationships between the pathological status of the target patient and known pediatric respiratory diseases; among them, the graph structure construction can use the multimodal features of the target patient as the feature vectors of the graph nodes to construct an undirected graph, and the graph convolution operation can aggregate the information of neighbor nodes in the undirected graph to update the representation of the node. At the same time, the graph neural network model also introduces a graph attention network to assign different attention weights to each edge to more accurately capture the relationship between nodes. The trained graph neural network model can learn the mapping relationship between nodes to generate pathological correlation relationships.

[0046] In step 104, a disease progression evolution map of pediatric respiratory diseases is constructed using the co-occurrence probability matrix of disease subtypes in the multimodal pathology space.

[0047] In some embodiments, constructing a disease progression evolution map of pediatric respiratory diseases using a co-occurrence probability matrix of disease subtypes in the multimodal pathology space can be achieved by the following steps: Extracting the transfer frequencies between disease subtypes in the multimodal pathology space and generating a co-occurrence tensor of disease subtypes; Performing probability modeling on the co-occurrence tensor to obtain a co-occurrence probability matrix of disease subtypes in the multimodal pathology space; Dynamic heterogeneous graph mapping is performed based on the co-occurrence probability matrix to obtain a disease progression evolution map of pediatric respiratory diseases.

[0048] In specific implementation, extracting the transfer frequency between each disease subtype in the multimodal pathology space and generating a co-occurrence tensor of the disease subtypes can be achieved in the following manner, namely: the disease subtype of each patient can be extracted from the multimodal pathology space, and a three-dimensional co-occurrence tensor can be constructed based on the disease subtypes of all patients; wherein the co-occurrence tensor is a multidimensional data structure that records the frequency of occurrence of different disease subtypes in the same patient. The co-occurrence tensor can reveal the co-occurrence relationship between disease subtypes, and provide a basic data structure for subsequent probabilistic modeling and construction of disease progression evolution maps, thereby supporting accurate pathological state constraint inference and evolutionary path analysis.

[0049] In specific implementation, the co-occurrence tensor is probabilistically modeled to obtain the co-occurrence probability matrix of the disease subtypes in the multimodal pathology space, which can be achieved in the following manner: first, the co-occurrence tensor can be normalized using an existing normalization method (such as row normalization or column normalization), and then, the normalized co-occurrence tensor is modeled using a tensor decomposition technique (such as CP decomposition or Tucker decomposition) to obtain the co-occurrence probability matrix of the disease subtypes in the multimodal pathology space; wherein, the co-occurrence probability matrix is a symmetric matrix representing the strength of association between different disease subtypes, and the co-occurrence probability matrix can quantify the co-occurrence relationship between different disease subtypes, providing basic data support for the subsequent construction of a disease progression evolution map; normalizing the co-occurrence tensor can convert the co-occurrence frequency into a probability value, which represents the strength of association between different disease subtypes; modeling the co-occurrence tensor through tensor decomposition technology can reduce the dimensionality of high-dimensional data and extract potential patterns and relationships, thereby better understanding the co-occurrence pattern between disease subtypes.

[0050] It should be noted that in this application, the disease progression evolution map is a weighted directed graph that reveals the potential evolutionary relationship between pediatric respiratory disease subtypes. The disease progression evolution map can identify the transformation paths and comorbidity patterns between disease subtypes, and provide accurate disease evolution predictions for clinicians. In specific implementation, dynamic heterogeneous graph mapping is performed based on the co-occurrence probability matrix to obtain the disease progression evolution map of pediatric respiratory diseases. This can be achieved in the following way, namely: first, the co-occurrence probability matrix is converted into a weighted directed graph through existing modeling methods (such as Markov chain modeling methods), in which each node represents a disease subtype, and the weight of the edge represents the co-occurrence probability between the two subtypes. Then, a graph embedding algorithm (such as Node2Vec) is used to generate a low-dimensional vector representation of each node in the weighted directed graph. Finally, the weighted directed graph with the low-dimensional vector representation of the node is used as the disease progression evolution map of pediatric respiratory diseases; wherein the low-dimensional vector of the node can represent the potential evolutionary relationship between the disease subtypes.

[0051] In step 105, the cross-attention mechanism is used to fuse the disease progression evolution map and the pathological correlation relationship, thereby extracting multidimensional decision indicators of the target patient's respiratory disease, and multimodal identification of the target patient's respiratory disease is performed based on the multidimensional decision indicators to obtain a disease auxiliary identification report for the target patient.

[0052] In some embodiments, the cross-attention mechanism is used to fuse the disease progression evolution graph and the pathological association relationship to extract the multidimensional decision indicator of the target patient's respiratory disease. The following steps can be used: Obtaining node embedding representations in the disease progression evolution graph and the pathological association relationship respectively; Determining the graph relevance between the disease progression graph and the pathological association relationship based on all node embedding representations through a cross-attention mechanism; Based on the graph correlation, feature fusion is performed on the nodes in the disease progression evolution graph and the pathological association relationship to obtain an enhanced disease progression evolution graph of the target patient's respiratory disease; A multidimensional decision indicator for the respiratory disease of the target patient is generated based on the fused nodes in the enhanced disease progression evolution map.

[0053] It should be noted that in this application, the node embedding representation is a vector that represents the structural information and attribute characteristics of the nodes in the graph; the node embedding representation of the disease progression evolution graph is the low-dimensional vector of the node, which can represent the potential evolutionary relationship between disease subtypes; the node embedding representation of the pathological association relationship is the mapping relationship between nodes.

[0054] In specific implementation, the graph correlation between the disease progression evolution graph and the pathological association relationship is determined based on all node embedding representations through a cross-attention mechanism, which can be achieved in the following manner: a cross-attention mechanism (such as scaled dot-product) can be used to calculate the attention weights between nodes by taking the node embeddings of the pathological association relationship as queries and the node embeddings of the disease progression evolution graph as keys and values, and the calculated attention weights are used as the correlations between the corresponding nodes, and then a multidimensional vector composed of all correlations is used as the graph correlation between the disease progression evolution graph and the pathological association relationship; wherein the graph correlation is a multidimensional vector that measures the correlation between each node in the disease progression evolution graph and the pathological association relationship, and the graph correlation can be used to quantify the credibility of the complementary information of the two graphs, thereby significantly improving the accuracy of multimodal graph fusion and the reliability of subsequent diagnostic decisions.

[0055] In specific implementation, feature fusion is performed on the nodes in the disease progression evolution map and the pathological association relationship based on the graph correlation to obtain an enhanced disease progression evolution map of the target patient's respiratory disease. This can be achieved in the following manner, namely: the node features of the pathological association relationship can be weightedly summed row by row through the attention weights between the nodes in the graph correlation to generate a fusion feature representation, and the fusion features are spliced with the node features of the disease progression evolution map to generate a mapping feature matrix that corresponds one-to-one to the nodes of the disease progression evolution map, and then the mapping feature matrix is used as the enhanced disease progression evolution map of the target patient's respiratory disease; wherein, the enhanced disease progression evolution map is a multimodal graph representation generated by weighted fusion of the nodes of the disease progression evolution map and the pathological association relationship. The enhanced disease progression evolution map can simultaneously capture the evolution of the pathological state and the image mapping features, thereby significantly improving the accuracy and credibility of the respiratory disease diagnosis decision vector.

[0056] It should be noted that in this application, the multidimensional decision index is a vector that characterizes the respiratory disease type and pathological evolution path of the target patient. The multidimensional decision index can provide accurate and interpretable input for the diagnostic model, thereby significantly improving the accuracy of respiratory disease identification and the credibility of the decision. In specific implementation, the generation of the multidimensional decision index of the target patient's respiratory disease based on the fused nodes in the enhanced disease progression evolution graph can be achieved in the following manner: first, the fused node embedding representation can be compressed into a graph-level representation of a fixed dimension through global average pooling (such as the average or maximum aggregation of node features) in the enhanced disease progression evolution graph. Then, the graph-level representation of all nodes is input into a multilayer perceptron, and the multilayer perceptron is used to generate a multidimensional decision index of the target patient's respiratory disease. The multilayer perceptron is composed of multiple fully connected layers and nonlinear activation functions (such as ReLU), which can map the graph-level representation into a multidimensional decision index of a fixed dimension. Each dimension in the multidimensional decision index corresponds to a different pathological subtype probability, and the pathological subtype probability describes the probability of the target patient suffering from a specific pathological subtype of respiratory disease (such as lung adenocarcinoma with different degrees of infiltration).

[0057] In some embodiments, multimodal identification of the respiratory disease of the target patient based on the multidimensional decision indicator and obtaining an auxiliary disease identification report of the target patient can be achieved by the following steps: determining a respiratory disease sequence for a target patient based on all pathological subtype probabilities in the multidimensional decision indicator; The respiratory disease sequence is input into a preset rule engine to generate a disease auxiliary identification report for the target patient.

[0058] In specific implementation, determining the respiratory disease sequence of the target patient based on the pathological subtype probabilities of various respiratory diseases in the multidimensional decision indicator can be achieved in the following manner, namely: for each pathological subtype probability in the multidimensional decision indicator, the pathological subtype probability is compared with a preset probability threshold (such as 0.9); if the value of the pathological subtype probability is greater than the probability threshold, the respiratory disease corresponding to the pathological subtype probability is added to the disease sequence of the target patient. The above steps can be used to obtain the disease sequence of the target patient, and then all respiratory diseases in the disease sequence are sorted in descending order according to the size of the pathological subtype probability to obtain the respiratory disease sequence of the target patient; wherein, the respiratory disease sequence is a priority disease list containing the predicted probability of the target patient's respiratory disease. The respiratory disease sequence provides a clear input focus for the subsequent rule engine, ensuring that the system gives priority to high-risk diseases, thereby achieving personalized, efficient and accurate diagnostic recommendations.

[0059] It should be noted that in this application, the disease-assisted identification report is a diagnostic conclusion for the target patient's respiratory disease. The disease-assisted identification report can be automatically triggered and accurately transmitted across systems, speeding up the diagnosis and treatment process, reducing medical errors, and improving the consistency and traceability of diagnosis and treatment. In specific implementation, the respiratory disease sequence is input into the preset rule engine, and the generation of the disease-assisted identification report can be achieved in the following way, namely: the respiratory disease sequence can be input into the existing preset rule engine (such as DRL rules or DMN decision tables) to generate a disease-assisted identification report, wherein the preset rule engine uses Drools native syntax or decision tables to match facts based on pediatric respiratory disease diagnosis and treatment guidelines to generate diagnostic recommendations containing diagnostic recommendation content, priority and evidence level.

[0060] In addition, another aspect of the present application, in some embodiments, the present application provides a pediatric respiratory disease diagnosis system based on pattern recognition, the system includes a disease auxiliary recognition unit, reference Figure 4 , which is a schematic diagram of the structure of a disease auxiliary identification unit according to some embodiments of the present application. The disease auxiliary identification unit 400 includes: an acquisition module 401, a processing module 402 and an execution module 403, which are described as follows: Acquisition module 401, in this application, acquisition module 401 is mainly used to acquire respiratory time series data composed of lung sound signals of the target patient; Processing module 402, in this application, is mainly used to construct a multimodal pathology space of pediatric respiratory diseases by combining the pathophysiological characteristics of children suffering from respiratory diseases in various medical institutions through a comparative learning model; It should be noted that the processing module 402 in the present application is further configured to perform respiratory pattern matching on the respiratory time series data and the subtype distribution of respiratory diseases in the multimodal pathological space, thereby obtaining a pathological state constraint that matches the current pathological state of the target patient, and determining a pathological correlation between the pathological state of the target patient and known pediatric respiratory diseases based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image; In addition, it should be noted that the processing module 402 in the present application is also used to construct a disease progression evolution map of pediatric respiratory diseases through the co-occurrence probability matrix of disease subtypes in the multimodal pathology space; Execution module 403. In this application, execution module 403 is mainly used to use the cross-attention mechanism to perform graph fusion on the disease progression evolution graph and the pathological correlation relationship, and then extract the multidimensional decision indicators of the target patient's respiratory disease, and perform multimodal identification of the target patient's respiratory disease based on the multidimensional decision indicators to obtain a disease auxiliary identification report for the target patient.

[0061] Each module in the aforementioned pattern recognition-based pediatric respiratory disease diagnosis system can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0062] In addition, in one embodiment, the present application provides a computer device, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory and a network interface connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data of an auxiliary identification method for pediatric respiratory organ diseases. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, an auxiliary identification method for pediatric respiratory organ diseases is implemented.

[0063] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0064] In one embodiment, a computer device is also provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the above-mentioned embodiment of the auxiliary identification method for pediatric respiratory diseases are implemented.

[0065] In one embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the steps in the above-mentioned embodiment of the auxiliary identification method for pediatric respiratory diseases.

[0066] In one embodiment, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the aforementioned embodiment of the method for assisting in identifying pediatric respiratory diseases.

[0067] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0068] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0069] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art could make various modifications and improvements without departing from the spirit of the present application, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be determined by the appended claims.

Claims

1. A method for assisting in identifying pediatric respiratory diseases, used in a pediatric respiratory disease diagnosis system to assist in identifying children's respiratory diseases, characterized in that: The method comprises the following steps: Collect respiratory time series data consisting of lung sound signals of the target patient; A multimodal pathology space for pediatric respiratory diseases was constructed by combining the pathophysiological characteristics of children with respiratory diseases in various medical institutions through a comparative learning model. Performing respiratory pattern matching on the respiratory time series data with the subtype distribution of respiratory diseases in the multimodal pathology space, thereby obtaining a pathological state constraint that matches the current pathological state of the target patient, and determining a pathological correlation between the pathological state of the target patient and known pediatric respiratory diseases based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image; constructing a disease progression evolution map of pediatric respiratory diseases by using a co-occurrence probability matrix of disease subtypes in the multimodal pathology space; The cross-attention mechanism is used to fuse the disease progression evolution map and the pathological correlation relationship, and then the multidimensional decision indicators of the target patient's respiratory disease are extracted. Based on the multidimensional decision indicators, the target patient's respiratory disease is multimodally identified to obtain a disease-assisted identification report for the target patient.

2. The method according to claim 1, wherein By combining the comparative learning model with the pathophysiological characteristics of children with respiratory diseases in various medical institutions, a multimodal pathological space for pediatric respiratory diseases was constructed, specifically including: For each medical institution in the distributed medical network, access pediatric respiratory disease data and lung medical imaging archives in the medical institution's local database; extracting lung sound spectrum features of respiratory diseases from the pediatric respiratory disease data; Performing feature extraction on all images in the lung medical image archive to obtain image texture features of the lung medical images; Performing dimensionality reduction and fusion on the lung sound spectrum features and the image texture features to obtain feature basis vectors of pediatric respiratory diseases in medical institutions, and further obtaining feature basis vectors of pediatric respiratory diseases in various medical institutions; All feature basis vectors are feature aligned using a contrastive learning model to generate a multimodal pathology space for pediatric respiratory diseases.

3. The method according to claim 2, wherein The contrastive learning model is used to align all feature basis vectors to generate a multimodal pathology space for pediatric respiratory diseases. Specifically, the following are performed: Perform multimodal data augmentation on each feature basis vector to generate enhanced positive sample pairs and heterogeneous negative sample pairs; Training a contrastive learning model by using the enhanced positive sample pairs and the heterogeneous negative sample pairs; Generating a multimodal pathology space for pediatric respiratory diseases using a trained contrastive learning model.

4. The method according to claim 1, wherein Performing respiratory pattern matching on the respiratory time series data and the subtype distribution of respiratory diseases in the multimodal pathology space, thereby obtaining a pathological state constraint that matches the current pathological state of the target patient specifically includes: Extracting features from the respiratory time series data to obtain respiratory features of the target patient; obtaining a subtype distribution of respiratory diseases in the multimodal pathology space; performing pattern matching on the respiratory characteristics and the subtype distribution using a dynamic time programming algorithm to obtain a matching confidence between the target patient's respiratory characteristics and each disease subtype; A pathological state constraint matching the current pathological state of the target patient is determined based on all matching confidences.

5. The method according to claim 1, wherein Determining the pathological correlation between the target patient's pathological state and known pediatric respiratory diseases based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image specifically includes: Obtain lung medical images of target patients; Extracting grayscale texture features of the target patient's lung medical image from the lung medical image; Performing multimodal feature fusion on the grayscale texture feature and the pathological state constraint to obtain a multimodal feature representation of the target patient's respiratory disease; A pathological correlation relationship between the target patient's pathological state and known pediatric respiratory diseases is established based on the multimodal feature representation.

6. The method according to claim 1, wherein Constructing a disease progression evolution map of pediatric respiratory diseases using the co-occurrence probability matrix of disease subtypes in the multimodal pathology space specifically includes: Extracting the transfer frequencies between disease subtypes in the multimodal pathology space and generating a co-occurrence tensor of disease subtypes; Performing probability modeling on the co-occurrence tensor to obtain a co-occurrence probability matrix of disease subtypes in the multimodal pathology space; Dynamic heterogeneous graph mapping is performed based on the co-occurrence probability matrix to obtain a disease progression evolution map of pediatric respiratory diseases.

7. The method according to claim 1, wherein The cross-attention mechanism is used to fuse the disease progression evolution map and the pathological association relationship, thereby extracting multidimensional decision indicators of the target patient's respiratory disease, including: Obtaining node embedding representations in the disease progression evolution graph and the pathological association relationship respectively; Determining the graph relevance between the disease progression graph and the pathological association relationship based on all node embedding representations through a cross-attention mechanism; Based on the graph correlation, feature fusion is performed on the nodes in the disease progression evolution graph and the pathological association relationship to obtain an enhanced disease progression evolution graph of the target patient's respiratory disease; A multidimensional decision indicator for the respiratory disease of the target patient is generated based on the fused nodes in the enhanced disease progression evolution map.

8. A pediatric respiratory disease diagnosis system based on pattern recognition, the system includes a disease auxiliary identification unit, characterized in that: The disease auxiliary identification unit includes: An acquisition module, used to acquire respiratory time series data composed of lung sound signals of a target patient; A processing module is used to construct a multimodal pathology space for pediatric respiratory diseases by combining the pathophysiological characteristics of children with respiratory diseases in various medical institutions through a comparative learning model; The processing module is configured to perform respiratory pattern matching on the respiratory time series data and the subtype distribution of respiratory diseases in the multimodal pathological space, thereby obtaining a pathological state constraint that matches the current pathological state of the target patient, and determining a pathological correlation between the pathological state of the target patient and known pediatric respiratory diseases based on the pathological state constraint and the grayscale texture features of the target patient's lung medical image; The processing module is configured to construct a disease progression evolution map of pediatric respiratory diseases using a co-occurrence probability matrix of disease subtypes in the multimodal pathology space; An execution module is used to use a cross-attention mechanism to fuse the disease progression evolution map and the pathological association relationship, thereby extracting multidimensional decision indicators of the target patient's respiratory disease, performing multimodal identification of the target patient's respiratory disease based on the multidimensional decision indicators, and obtaining a disease-assisted identification report for the target patient.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the auxiliary identification method for pediatric respiratory diseases according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the auxiliary identification method for pediatric respiratory diseases according to any one of claims 1 to 7 are implemented.

Images