Intelligent medical data processing system and method based on artificial intelligence

Abstract

The present application relates to the field of medical data processing technology, and specifically discloses an intelligent medical data processing system and method based on artificial intelligence, which uses deep learning-based image processing technology to perform multi-level feature extraction on CT scan images of chronic disease patients, and mines out the texture features and structural features of the bronchiectasis area in the CT scan images, and performs fine-grained semantic interactive fusion of the texture features and structural features of the bronchiectasis state based on core correlation features to achieve a comprehensive understanding of the bronchiectasis state, and then intelligently identify the bronchiectasis type. In this way, the recognition accuracy of the bronchiectasis type can be effectively improved, thereby providing doctors with a more accurate diagnosis basis.

Description

Technical Field

[0001] The present application relates to the field of medical data processing technology, and more specifically, to an intelligent medical data processing system and method based on artificial intelligence. Background Art

[0002] With the development of medical technology, medical imaging has become increasingly important as one of the important tools for clinical diagnosis. Especially for patients with chronic diseases, regular imaging examinations are a key means to evaluate disease progression and adjust treatment plans. Among the many medical imaging technologies, computed tomography (CT) is widely used in the diagnosis and monitoring of various diseases because it can provide high-resolution images of in vivo tissue structures.

[0003] Patients with chronic diseases usually need to track their condition over a long period of time, especially lung diseases such as bronchiectasis. Bronchiectasis is a common chronic respiratory disease characterized by abnormal dilation of the bronchi, leading to respiratory dysfunction. In order to accurately assess the extent of bronchiectasis and its effects, doctors often rely on the detailed anatomical information provided by CT scan images. However, traditional CT image analysis methods mainly rely on the doctor's experience and judgment, which is not only time-consuming and labor-intensive, but also highly subjective and easily affected by individual differences.

[0004] In recent years, with the development of artificial intelligence technology, especially in the field of image processing, automated CT image analysis has become possible. For example, the invention patent with publication number CN116612891A discloses a data processing system for chronic disease patients, which analyzes the type of bronchiectasis, the length, position, bronchial-arterial ratio and bronchial wall thickening coefficient of abnormal bronchial segments in various types of bronchiectasis and other bronchiectasis pathological information of chronic disease patients based on the CT scan images of chronic disease patients, and then calculates the severity coefficient of bronchiectasis pathology of patients.

[0005] However, when identifying the type of bronchiectasis, the above scheme extracts the contours of bronchiectasis at various locations and compares them with the contours corresponding to the preset types of bronchiectasis (such as cystic bronchiectasis, columnar bronchiectasis, and varicose bronchiectasis), thereby screening out the types of bronchiectasis at various locations in the reexamination CT scan images. Although this simple comparison method can distinguish different types of bronchiectasis to a certain extent, it mainly focuses on the contours of bronchiectasis and lacks in-depth analysis of the overall structure and complex texture of bronchiectasis. It is easily affected by factors such as noise and individual differences, resulting in low recognition accuracy and difficulty in adapting to different forms of bronchiectasis.

[0006] Therefore, an optimized artificial intelligence-based intelligent medical data processing system and method is expected. Summary of the invention

[0007] In order to solve the above technical problems, this application is proposed. The embodiment of the present application provides an intelligent medical data processing system and method based on artificial intelligence, which uses deep learning-based image processing technology to perform multi-level feature extraction on CT scan images of chronic disease patients, and mines out the texture features and structural features of the bronchiectasis area in the CT scan image, and through the fine-grained semantic interactive fusion of the texture features and structural features of the bronchiectasis state based on core correlation features, to achieve a comprehensive understanding of the bronchiectasis state, and then intelligently identify the bronchiectasis type. In this way, the recognition accuracy of the bronchiectasis type can be effectively improved, thereby providing doctors with a more accurate diagnostic basis.

[0008] Accordingly, according to one aspect of the present application, there is provided an artificial intelligence-based smart medical data processing system, comprising:

[0009] The CT scan image acquisition module is used to set the duration of the monitoring period and extract the CT scan images of each follow-up of each bronchiectasis chronic disease patient received by the designated hospital during the monitoring period from the patient information database of the designated hospital;

[0010] A bronchiectasis type identification module is used to obtain the bronchiectasis type of each part of the CT scan images of each chronic disease patient at each review, based on the CT scan images of each chronic disease patient at each review;

[0011] The pathological severity estimation module is used to classify the bronchiectasis in the CT scan images of each chronic disease patient at each follow-up according to the same bronchiectasis type, and calculate the bronchiectasis pathological severity coefficient of each follow-up based on the length, position, bronchial-arterial ratio and bronchial wall thickening coefficient of the abnormal bronchial segment corresponding to each bronchiectasis in each type of bronchiectasis in the CT scan images of each follow-up.

[0012] In the above-mentioned artificial intelligence-based smart medical data processing system, the bronchiectasis type identification module includes: a target area detection unit, which is used to perform bronchial target area detection on the CT scan image to obtain a bronchial target area image; an expansion state feature extraction unit, which is used to perform multi-scale feature extraction on the bronchial target area image to obtain a bronchiectasis state texture feature map and a bronchiectasis state structural feature map; a fine-grained joint encoding unit, which is used to perform global semantic interactive encoding based on core joint features on the bronchiectasis state texture feature map and the bronchiectasis state structural feature map to obtain a bronchiectasis state texture-structure fine-grained joint encoding feature map; a bronchiectasis type determination unit, which is used to identify the bronchiectasis type of bronchiectasis based on the bronchiectasis state texture-structure fine-grained joint encoding feature map.

[0013] According to another aspect of the present application, there is provided a method for processing intelligent medical data based on artificial intelligence, comprising:

[0014] Performing bronchial target area detection on the CT scan image to obtain a bronchial target area image;

[0015] Performing multi-scale feature extraction on the bronchial target area image to obtain a bronchial dilatation state texture feature map and a bronchial dilatation state structural feature map;

[0016] Performing global semantic interactive encoding based on core joint features on the bronchiectasis state texture feature map and the bronchiectasis state structural feature map to obtain a bronchiectasis state texture-structure fine-grained joint encoding feature map;

[0017] Based on the bronchiectasis state texture-structure fine-grained joint encoding feature map, the bronchiectasis type of the bronchiectasis is identified.

[0018] Compared with the prior art, the artificial intelligence-based smart medical data processing system and method provided by the present application uses deep learning-based image processing technology to perform multi-level feature extraction on CT scan images of chronic disease patients, dig out the texture features and structural features of the bronchiectasis area in the CT scan images, and perform fine-grained semantic interactive fusion of the texture features and structural features of the bronchiectasis state based on core correlation features to achieve a comprehensive understanding of the bronchiectasis state, and then intelligently identify the bronchiectasis type. In this way, the recognition accuracy of the bronchiectasis type can be effectively improved, thereby providing doctors with a more accurate diagnosis basis. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 4 is a block diagram of an artificial intelligence-based smart medical data processing system according to an embodiment of the present application.

[0021] Figure 2 Schematic diagram of data flow of an artificial intelligence-based smart medical data processing system according to an embodiment of the present application.

[0022] Figure 3 It is a block diagram of an expansion state feature extraction unit in an artificial intelligence-based smart medical data processing system according to an embodiment of the present application.

[0023] Figure 4 It is a block diagram of a fine-grained joint encoding unit in an artificial intelligence-based smart medical data processing system according to an embodiment of the present application.

[0024] Figure 5 This is a block diagram of a fine-grained common feature extraction subunit in an artificial intelligence-based smart medical data processing system according to an embodiment of the present application.

[0025] Figure 6 Flow chart of an artificial intelligence-based smart medical data processing method according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0027] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0028] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0029] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0030] As mentioned in the above background technology, patent CN116612891A proposes an intelligent medical data processing system based on artificial intelligence, which includes: a CT scan image acquisition module, which is used to set the duration of the monitoring period, and extract the CT scan images of each follow-up of each chronic disease patient with bronchiectasis who is admitted to the designated hospital during the monitoring period from the patient information database of the designated hospital; a bronchiectasis type identification module, which is used to obtain the bronchiectasis type of each bronchiectasis in the CT scan images of each follow-up of each chronic disease patient based on the CT scan images of each follow-up; a pathological severity estimation module, which is used to classify the bronchiectasis in the CT scan images of each chronic disease patient in each follow-up according to the same bronchiectasis type, and calculate the bronchiectasis pathological severity coefficient of each follow-up based on the length, position, bronchial-arterial ratio and bronchial wall thickening coefficient of the abnormal bronchial segment corresponding to each bronchiectasis in each type of bronchiectasis in the CT scan images of each follow-up.

[0031] In the process of identifying the type of bronchiectasis, the above-mentioned intelligent medical data processing system based on artificial intelligence captures the contour of the bronchiectasis area and matches it with the predefined contours of various types of bronchiectasis, so as to identify the types of bronchiectasis in different areas in the reviewed CT scan images. Although this simple matching method based on contours can distinguish different types of bronchiectasis to a certain extent, it mainly relies on contour features without in-depth analysis of the overall structure and complex texture features of bronchiectasis. Therefore, this method is easily affected by image noise and individual differences, resulting in insufficient recognition accuracy and difficulty in adapting to the diverse forms of bronchiectasis. In response to this technical problem, the present application proposes an optimized intelligent medical data processing system based on artificial intelligence, which uses deep learning-based image processing technology to perform multi-level feature extraction on CT scan images of chronic disease patients for review, and mines the texture features and structural features of the bronchiectasis area in the CT scan images. The texture features and structural features of the bronchiectasis state are interactively fused based on core correlation features to achieve a comprehensive understanding of the bronchiectasis state, and then intelligently identify the bronchiectasis type. In this way, the accuracy of identifying bronchiectasis types can be effectively improved, thereby providing doctors with more accurate diagnostic basis.

[0032] In the intelligent medical data processing system based on artificial intelligence, the main task of the CT scan image acquisition module is to set the duration of the monitoring period and extract the CT scan images of each follow-up of patients with chronic bronchiectasis who were seen in the designated hospital during the monitoring period from the patient information database of the designated hospital.

[0033] First, in order to ensure the accuracy and completeness of data collection, it is necessary to conduct preliminary research and docking on the patient information database of the target hospital, which includes understanding the hospital's existing information system architecture, data storage format, and key information such as data access permissions. Through cooperation with the hospital's information department, the development team can develop a data access plan that meets the actual situation of the hospital. For example, for hospitals that use the HL7 (Health Level Seven International) standard interface, relevant services can be directly called through the API (application programming interface) to automatically capture patient information; for hospitals that use private databases or non-standardized data formats, it may be necessary to customize the development of data conversion tools to convert raw data into a unified standard format for subsequent processing.

[0034] After the data access plan is determined, the next step is to define the monitoring cycle and its related parameters. The choice of monitoring cycle needs to be determined according to the characteristics of bronchiectasis chronic disease and clinical needs. Usually, doctors will recommend appropriate review frequencies based on the patient's specific conditions, such as severity of the disease, treatment response and other factors, such as every 3 or 6 months. Therefore, when designing the CT scan image acquisition module, a flexible time range setting function should be provided to allow users to freely select the start and end dates of the monitoring cycle according to actual needs, while supporting batch import of patient lists to improve work efficiency.

[0035] After the monitoring cycle is set, the system will automatically query all eligible patient records within the specified time period and screen out individuals with chronic bronchiectasis. This process involves parsing and matching patient diagnostic information, which requires high data processing capabilities and accuracy. To this end, natural language processing technology can be used to intelligently analyze text descriptions in electronic medical records to identify cases containing the keyword "bronchiectasis"; at the same time, combined with ICD codes (International Classification of Diseases codes), the correctness of the diagnostic results can be further confirmed to ensure that the patient group finally selected is highly relevant and representative.

[0036] Subsequently, for each selected patient, the system will call the API interface in the hospital's Picture Archiving and Communication System (PACS) to obtain all CT scan image data of the patient during the monitoring period. As an indispensable information management system for the imaging department of modern hospitals, the PACS system can not only store a large amount of medical imaging data, but also provide a wealth of query and retrieval functions. Through the call of the API interface, the CT scan image acquisition module can directly access the original DICOM files on the PACS server. These files contain detailed image information and metadata, such as scan time, equipment model, imaging parameters, etc., laying a solid foundation for subsequent image analysis.

[0037] It is worth noting that during the data transmission process, relevant privacy protection laws and regulations must be strictly observed, and security measures such as encrypted transmission must be taken to prevent the leakage of sensitive information. In addition, considering that there may be differences in network environments between different hospitals, the system must also have good compatibility and fault tolerance to ensure stable operation under various complex conditions.

[0038] Finally, all collected CT scan images will be centrally stored in a cloud database or local server for researchers to conduct in-depth analysis. With the help of artificial intelligence algorithms, such as deep learning models, the characteristic manifestations of bronchiectasis can be automatically identified from a large amount of imaging data, helping doctors to more accurately judge the trend of disease changes, evaluate treatment effects, and even predict the risk of complications that may occur in the future. At the same time, through the comprehensive analysis of multi-center, large sample size data, it is also helpful to discover new pathological mechanisms and promote the updating and improvement of clinical diagnosis and treatment guidelines.

[0039] In summary, setting the duration of the monitoring cycle and extracting the CT scan images of each follow-up of each patient with chronic bronchiectasis who was admitted to the designated hospital during the monitoring cycle from the patient information database of the designated hospital is an important part of the intelligent medical data processing system based on artificial intelligence. This process requires not only scientific monitoring cycle settings, but also efficient data retrieval and sorting technology. Through this series of steps, the system can extract the CT scan images of each follow-up of each patient with chronic bronchiectasis who was admitted to the designated hospital during the monitoring cycle from the patient information database of the designated hospital, providing a basis for subsequent data processing and analysis.

[0040] It is worth mentioning that in this application, all actions to obtain information are carried out in compliance with the relevant data protection laws and policies of the country where the application is located, and with the authorization of the owner of the corresponding device.

[0041] Figure 1 4 is a block diagram of an artificial intelligence-based smart medical data processing system according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of an artificial intelligence-based smart medical data processing system according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the artificial intelligence-based smart medical data processing system 100 includes: a target area detection unit 110, which is used to perform bronchial target area detection on the CT scan image to obtain a bronchial target area image; an expansion state feature extraction unit 120, which is used to perform multi-scale feature extraction on the bronchial target area image to obtain a bronchial expansion state texture feature map and a bronchial expansion state structural feature map; a fine-grained joint encoding unit 130, which is used to perform global semantic interactive encoding based on core joint features on the bronchial expansion state texture feature map and the bronchial expansion state structural feature map to obtain a bronchial expansion state texture-structure fine-grained joint encoding feature map; a bronchiectasis type determination unit 140, which is used to identify the bronchiectasis type of the bronchiectasis based on the bronchial expansion state texture-structure fine-grained joint encoding feature map.

[0042] In the above-mentioned intelligent medical data processing system based on artificial intelligence, the target area detection unit 110 is used to perform bronchial target area detection on the CT scan image to obtain a bronchial target area image. In a specific example of the present application, the target area detection unit 110 further includes: inputting the CT scan image into a bronchial target area detection module based on the YOLO model to obtain the bronchial target area image. It should be understood that the present application takes into account that in large-scale clinical applications, directly processing the entire CT scan image will not only consume a large amount of computing resources, but also because the CT scan image usually contains a large amount of non-target area information, such as other lung tissues, bones, etc., these irrelevant information may interfere with the accuracy of bronchial dilatation state feature extraction, thereby affecting the recognition accuracy. Therefore, the present application uses a bronchial target area detection module based on the YOLO model to pre-process the CT scan image, and uses the efficient target detection capability of the YOLO model to accurately identify the bronchial structure in the CT scan image, quickly locate and extract the bronchial target area image, thereby effectively eliminating the interference of other non-target areas and improving the efficiency and accuracy of subsequent processing steps.

[0043] In the above-mentioned intelligent medical data processing system based on artificial intelligence, the expansion state feature extraction unit 120 is used to perform multi-scale feature extraction on the bronchial target area image to obtain a bronchial expansion state texture feature map and a bronchial expansion state structural feature map. Figure 3 FIG. 1 is a block diagram of an expansion state feature extraction unit in an artificial intelligence-based smart medical data processing system according to an embodiment of the present application. Figure 3 As shown, the expansion state feature extraction unit 120 includes: a texture feature extraction subunit 121, which is used to input the bronchial target area image into a texture feature extractor based on a first atrous convolutional neural network model to obtain the bronchial expansion state texture feature map; a structural feature extraction subunit 122, which is used to input the bronchial target area image into a structural feature extractor based on a second atrous convolutional neural network model to obtain the bronchial expansion state structural feature map.

[0044] Specifically, the texture feature extraction subunit 121 is used to input the bronchial target area image into a texture feature extractor based on the first atrous convolutional neural network model to obtain the bronchial dilatation state texture feature map. It should be understood that during bronchiectasis, the bronchial wall thickens due to inflammation and infection, and the surrounding lymphoid tissue also proliferates, thereby causing increased blood circulation in the lungs, thereby increasing the lung texture. In addition, different types of bronchial dilatation (such as columnar dilatation, cystic dilatation, and irregular dilatation) have different texture manifestations. For example, columnar dilatation may appear as a uniform thickening of the bronchi, while cystic dilatation may appear as multiple cystic translucent areas. Therefore, the present application uses a texture feature extractor based on the first atrous convolutional neural network model to capture the complex texture features of the bronchial target area image. Those skilled in the art should know that the atrous convolutional neural network model has excellent performance in image feature extraction. It can control the size of the receptive field by adjusting the atrous rate, thereby expanding the receptive field without losing image resolution, more carefully analyzing the texture changes in the bronchial dilation area, and extracting the texture feature map of the bronchial dilation state.

[0045] Specifically, the structural feature extraction subunit 122 is used to input the bronchial target area image into a structural feature extractor based on the second hole convolutional neural network model to obtain the structural feature map of the bronchial dilatation state. It should be understood that the present application takes into account that the overall expansion structure of the bronchus is also of great significance for the characterization of its branch expansion type. For example, the columnar dilated bronchus is uniformly tubular, but suddenly becomes thinner at one place, which usually indicates that the wall damage is relatively mild, and the inflammation may be mainly confined to a certain area of ​​the bronchus; the cystic dilated bronchial cavity is cystic, and the bronchial end is also an unrecognizable cystic structure, which usually indicates that the wall damage is more serious, and the inflammation has widely affected the bronchial wall and its surrounding tissues; the irregular dilated bronchial cavity is irregular or beaded, which may indicate that the wall damage and inflammatory changes are significantly different in different regions, resulting in irregular bronchial morphology. Therefore, in order to improve the recognition accuracy of branch expansion type, the present application further introduces a structural feature extractor based on the second hole convolutional neural network model to extract the overall structural features of the bronchi in the bronchial target area image.

[0046] In particular, since the first atrous convolutional neural network model is mainly responsible for extracting the texture features of the bronchial dilation area, focusing on local subtle changes, such as texture uniformity and directionality, etc., while the second atrous convolutional neural network model is mainly responsible for extracting the structural features of the bronchial dilation area, focusing on the overall morphology and layout, such as the shape, size, and branching pattern of the bronchi, etc. Therefore, in the technical solution of the present application, the network depth and expansion rate of the first atrous convolutional neural network are relatively small to adapt to the fine analysis of texture features; while the network depth and expansion rate of the second atrous convolutional neural network are relatively large to adapt to the macroscopic analysis of structural features.

[0047] In the above-mentioned intelligent medical data processing system based on artificial intelligence, the fine-grained joint encoding unit 130 is used to perform global semantic interaction encoding on the bronchial dilatation state texture feature map and the bronchial dilatation state structural feature map based on core joint features to obtain a bronchial dilatation state texture-structure fine-grained joint encoding feature map. It should be understood that, considering that the bronchial dilatation state texture feature map and the bronchial dilatation state structural feature map respectively represent different aspects of bronchiectasis, the two have significant complementarity. Therefore, the present application further fuses the bronchial dilatation state texture feature map and the bronchial dilatation state structural feature map. In particular, considering that traditional fusion methods such as feature splicing and pixel-by-pixel addition are usually unable to effectively utilize the intrinsic connection between texture features and structural features, there are problems such as low fusion efficiency and information loss. In this regard, the present application proposes a global semantic interaction encoding method based on core joint features, which realizes fine-grained semantic interaction fusion by mining the intrinsic connection between the two, thereby enhancing the comprehensive ability of feature expression. Among them, Figure 4 FIG. 1 is a block diagram of a fine-grained joint encoding unit in an artificial intelligence-based smart medical data processing system according to an embodiment of the present application. Figure 4 As shown, the fine-grained joint encoding unit 130 includes: a fine-grained common feature extraction subunit 131, which is used to perform fine-grained common feature extraction on the bronchial dilation state texture feature map and the bronchial dilation state structure feature map to obtain a set of bronchial dilation state texture-structure joint fine-grained core feature vectors; a global semantic interaction subunit 132, which is used to perform global interactive encoding based on semantic clustering centers on the set of bronchial dilation state texture-structure joint fine-grained core feature vectors to obtain the bronchial dilation state texture-structure fine-grained joint encoding feature map.

[0048] Figure 5 FIG. 1 is a block diagram of a fine-grained common feature extraction subunit in an artificial intelligence-based smart medical data processing system according to an embodiment of the present application. Figure 5As shown, the fine-grained common feature extraction subunit 131 includes: a feature fine-grained decoupling secondary subunit 1311, which is used to perform feature fine-grained decoupling on the bronchial dilation state texture feature map and the bronchial dilation state structural feature map to obtain a set of local fine-grained texture feature vectors of the bronchial dilation state and a set of local fine-grained structural feature vectors of the bronchial dilation state; a common feature extraction secondary subunit 1312, which is used to input each group of corresponding local fine-grained texture feature vectors of the bronchial dilation state and local fine-grained structural feature vectors of the bronchial dilation state in the set of local fine-grained texture feature vectors of the bronchial dilation state and the set of local fine-grained structural feature vectors of the bronchial dilation state into a fine-grained common feature extraction network to obtain a set of joint fine-grained core feature vectors of the bronchial dilation state texture-structure.

[0049] More specifically, the feature fine-grained decoupling secondary subunit 1311 is used to perform feature fine-grained decoupling on the bronchiectasis state texture feature map and the bronchiectasis state structural feature map to obtain a set of bronchiectasis state local fine-grained texture feature vectors and a set of bronchiectasis state local fine-grained structural feature vectors, which are expressed as follows:

[0050] ;

[0051] ;

[0052] in, A texture feature map representing the bronchiectasis state, A structural characteristic diagram representing the bronchiectasis state, Indicates feature decoupling, , , and The first, second, and third local fine-grained texture feature vectors in the set represent the bronchial dilation state. and bronchiectasis state local fine-grained texture feature vector, is the number of feature vectors in the set of local fine-grained texture feature vectors of the bronchial dilation state, , , and The first, second, and third local fine-grained structural feature vectors of the bronchiectasis state are represented respectively. and The local fine-grained structural feature vector of bronchiectasis state.

[0053] That is, in order to refine the feature granularity of feature interaction encoding, the present application performs fine-grained decoupling processing on the bronchial dilation state texture feature map and the bronchial dilation state structural feature map, decomposing the texture features and structural features into more subtle feature components, so as to better identify and distinguish the local detail information in the feature map, thereby generating a set of local fine-grained texture feature vectors of the bronchial dilation state and a set of local fine-grained structural feature vectors of the bronchial dilation state.

[0054] More specifically, the common feature extraction secondary subunit 1312 is further used to: respectively calculate the position point subtraction, position point multiplication and position point addition between the local fine-grained texture feature vector of the bronchial dilation state and the local fine-grained structural feature vector of the bronchial dilation state to obtain the first dilation state texture-structure joint interaction representation vector, the second dilation state texture-structure joint interaction representation vector and the third dilation state texture-structure joint interaction representation vector; cascade the first dilation state texture-structure joint interaction representation vector, the second dilation state texture-structure joint interaction representation vector and the third dilation state texture-structure joint interaction representation vector and input them into the neural network layer based on the tanh function to obtain the bronchial dilation state texture-structure joint fine-grained core feature vector, which is expressed as follows:

[0055] ;

[0056] in, represents dot product, Indicates point addition, Indicates point reduction, Indicates cascade, and denote the weight matrix and bias term respectively, represents the hyperbolic tangent function, The first of the set of texture-structure joint fine-grained core feature vectors representing the bronchiectasis state A fine-grained core feature vector of texture-structure joint for bronchiectasis status.

[0057] That is, a neural network structure is used to construct a fine-grained common feature extraction network to identify and extract common feature patterns between two feature sets, refine the intrinsic connection between the two, reduce the impact of inter-domain differences on subsequent feature context interaction analysis, and generate a set of texture-structure joint fine-grained core feature vectors of bronchial dilation status.

[0058] Specifically, the global semantic interaction subunit 132 is further used to: input the set of the bronchial dilation state texture-structure joint fine-grained core feature vectors into the clustering network to obtain the bronchial dilation state texture-structure joint feature semantic clustering center vector. In a specific example of the present application, the set of the bronchial dilation state texture-structure joint fine-grained core feature vectors is input into the clustering network to obtain the bronchial dilation state texture-structure joint feature semantic clustering center vector, further including: calculating the position mean vector of the set of the bronchial dilation state texture-structure joint fine-grained core feature vectors to obtain the bronchial dilation state texture-structure joint feature semantic clustering center vector, which is expressed as:

[0059] ;

[0060] in, The semantic clustering center vector of the texture-structure joint feature representing the bronchiectasis state, The number of fine-grained core feature vectors representing the texture-structure joint of bronchiectasis status.

[0061] That is, through the clustering network, the global aggregate representation of the joint features of the texture and structure of the bronchial dilation state is realized, and the semantic clustering center vector of the texture-structure joint features of the bronchial dilation state is generated, and it is used as the global context information to guide the subsequent global context interaction of the joint features, so as to enhance the consistency and semantic richness of the feature representation.

[0062] In a specific example of the present application, the global semantic interaction subunit 132 is further used to: construct a query vector and a value vector based on each bronchial dilation state texture-structure joint fine-grained core feature vector in the set of the bronchial dilation state texture-structure joint fine-grained core feature vectors, and construct a key vector based on the bronchial dilation state texture-structure joint feature semantic clustering center vector, and input the query vector, the value vector and the key vector into the global semantic interaction encoder based on the converter structure to obtain a set of bronchial dilation state texture-structure context joint fine-grained core feature vectors, which is expressed by the formula:

[0063] ;

[0064] ;

[0065] ;

[0066] ;

[0067] ;

[0068] in, , and denote the query embedding matrix, key embedding matrix and value embedding matrix respectively, and Respectively represent The query vector and value vector corresponding to the fine-grained core feature vector of the bronchiectasis state texture-structure joint, represents the key vector, represents the length of the key vector, represents the matrix multiplication operation, represents the transpose of a vector, represents the normalized exponential function, A collection of fine-grained core feature vectors representing the joint texture-structural context of bronchiectasis state, , , and The first, second, and third eigenvectors in the set of the bronchial dilation state texture-structure context joint fine-grained core feature vectors are represented respectively. and A fine-grained core feature vector of texture-structure context joint for bronchiectasis status.

[0069] That is, a converter structure is used to perform global context interaction encoding of the joint features of the bronchial dilation state texture structure, and through the self-attention mechanism, information exchange on a global scale is achieved, which helps to adjust its own representation according to the global context information and obtain a richer feature representation.

[0070] In a specific example of the present application, the global semantic interaction subunit 132 is further used to: perform feature shape reshaping on the set of the bronchial dilation state texture-structure context joint fine-grained core feature vectors to obtain the bronchial dilation state texture-structure fine-grained joint encoding feature map, which is expressed by the formula:

[0071] ;

[0072] in, represents the feature shape reshaping, A texture-structure fine-grained joint encoding feature map representing the bronchiectasis state.

[0073] That is, a feature reshape operation is performed on the generated bronchiectasis state texture-structure context joint fine-grained core feature vector set to restore the original feature structure and generate a bronchiectasis state texture-structure fine-grained joint encoding feature map. In this way, the deep fusion of texture features and structural features can be achieved at the fine-grained level, thereby improving the recognition accuracy of bronchiectasis state.

[0074] In the above-mentioned intelligent medical data processing system based on artificial intelligence, the bronchiectasis type determination unit 140 is used to identify the bronchiectasis type of bronchiectasis based on the bronchiectasis state texture-structure fine-grained joint coding feature map. In a specific example of the present application, the bronchiectasis type determination unit 140 is further used to: input the bronchiectasis state texture-structure fine-grained joint coding feature map into a dilation type identifier based on a classifier to obtain the bronchiectasis type of bronchiectasis. It should be understood that after the above processing, the bronchiectasis state texture-structure fine-grained joint coding feature map contains the deep fusion information of the texture features and structural features of the bronchial dilation area, which can effectively reflect the bronchiectasis type. After the classifier receives the bronchiectasis state texture-structure fine-grained joint coding feature map, it is further processed using the internal neural network structure to extract higher-level abstract features, and the probability distribution is calculated through the softmax function of the output layer to obtain the probability value of the bronchiectasis state texture-structure fine-grained joint coding feature map belonging to each bronchiectasis type. Furthermore, the most likely bronchiectasis type of the patient can be determined based on the probability value and output as the identification result.

[0075] In a preferred example of the present application, the bronchiectasis state texture-structure fine-grained joint encoding feature map is input into a classifier-based dilation type identifier to obtain the bronchiectasis type of the bronchiectasis, including:

[0076] The absolute value sum of all eigenvalues ​​of the bronchiectasis state texture-structure fine-grained joint encoding feature map and the square root of the square sum are calculated to obtain the first bronchiectasis state texture-structure fine-grained joint encoding spatial structure value and the second bronchiectasis state texture-structure fine-grained joint encoding spatial structure value, that is:

[0077] ;

[0078] ;

[0079] in, The first one represents the texture-structure fine-grained joint encoding feature map of the bronchiectasis state eigenvalues, Indicates the first bronchiectasis state texture-structure fine-grained joint encoding spatial structure value, Indicates the second bronchiectasis state texture-structure fine-grained joint encoding spatial structure value;

[0080] Determine the total number of eigenvalues ​​of all eigenvalues ​​of the bronchiectasis state texture-structure fine-grained joint encoding feature map ;

[0081] For each eigenvalue of the bronchiectasis state texture-structure fine-grained joint encoding feature map, the first bronchiectasis state texture-structure fine-grained joint encoding spatial structure value minus the product of the eigenvalue and the total number of eigenvalues ​​is calculated to obtain the first bronchiectasis state texture-structure fine-grained joint encoding long-range dependency value ,in, Indicates the first bronchiectasis state texture-structure fine-grained joint encoding spatial structure value, The first one represents the texture-structure fine-grained joint encoding feature map of the bronchiectasis state eigenvalues, the total number of eigenvalues ​​representing all eigenvalues ​​of the bronchial dilation state texture-structure fine-grained joint encoding feature map, Indicates the first bronchiectasis state texture-structure fine-grained joint encoding long-range dependency value;

[0082] Calculate the second bronchiectasis state texture-structure fine-grained joint encoding long-range dependency value obtained by multiplying the square root of the total number of eigenvalues ​​by the product of the eigenvalues ​​minus the second bronchiectasis state texture-structure fine-grained joint encoding spatial structure value ,in, Indicates the second bronchiectasis state texture-structure fine-grained joint encoding spatial structure value, The first one represents the texture-structure fine-grained joint encoding feature map of the bronchiectasis state eigenvalues, the total number of eigenvalues ​​representing all eigenvalues ​​of the bronchial dilation state texture-structure fine-grained joint encoding feature map, Indicates the second bronchiectasis state texture-structure fine-grained joint encoding long-range dependency value;

[0083] The index value calculated by taking the first bronchiectasis state texture-structure fine-grained joint coding long-range dependency value as the exponent of the natural constant and the inverse of the second bronchiectasis state texture-structure fine-grained joint coding long-range dependency value are weighted summed to obtain the optimized eigenvalue corresponding to each eigenvalue ,in, Indicates the first bronchiectasis state texture-structure fine-grained joint encoding long-range dependency value, Indicates the second bronchiectasis state texture-structure fine-grained joint encoding long-range dependency value, represents a natural constant, and represents the weighted hyperparameter, Indicates the optimized eigenvalue corresponding to each eigenvalue;

[0084] Combining the optimized feature values ​​into an optimized bronchial dilation state texture-structure fine-grained joint encoding feature map;

[0085] The optimized bronchial dilatation state texture-structure fine-grained joint encoding feature map is input into the classifier-based dilatation type identifier to obtain the dilatation type of the bronchial dilatation.

[0086] Here, in the case where the bronchial dilation state texture feature map and the bronchial dilation state structural feature map respectively represent the texture image semantic coding features and the structural image semantic coding features of the bronchial target area image, when performing global semantic interaction coding based on cross-domain fine-grained core joint features, the cross-domain fine-grained core joint feature representation of the bronchial dilation state texture-structure fine-grained joint coding feature map will also cause distribution field selective state space squeezing due to differences in different modal feature depths and receptive fields, which will make the bronchial dilation state texture-structure fine-grained joint coding feature map have feature interaction distribution space structure differences, affecting the convergence consistency of the classifier, thereby affecting the accuracy of the bronchiectasis type obtained by the classifier-based dilation type identifier.

[0087] That is, in view of the possible spatial structure loss in the high-dimensional space of the feature set of the bronchial dilation state texture-structure fine-grained joint encoding feature map, which causes the weight of the classifier to implicitly infer the spatial structure information based on the features, resulting in inconsistent convergence. A long-distance feature dependency relationship is established based on the overall feature scale of the bronchial dilation state texture-structure fine-grained joint encoding feature map relative to the spatial structure representation of the bronchial dilation state texture-structure fine-grained joint encoding feature map, so as to establish the feature local connectivity of the bronchial dilation state texture-structure fine-grained joint encoding feature map, and to capture the spatial ambiguous information of the object feature value through the unstructured feature value point prediction of the bronchial dilation state texture-structure fine-grained joint encoding feature map, thereby improving the spatial inductive bias perception ability of the feature set of the bronchial dilation state texture-structure fine-grained joint encoding feature map, improving the convergence consistency of the classifier, and improving the accuracy of the bronchiectasis type obtained by the bronchial dilation type identifier based on the classifier of the bronchial dilation state texture-structure fine-grained joint encoding feature map.

[0088] In the intelligent medical data processing system based on artificial intelligence, the pathological severity estimation module receives the bronchiectasis type information output by the bronchiectasis type recognition module, such as cystic bronchiectasis, columnar bronchiectasis or varicose bronchiectasis, and identifies it after multi-level feature extraction and fine-grained semantic interactive fusion of CT scan images through a deep learning model. The pathological severity estimation module will classify the bronchiectasis in each CT scan image of each review according to the same bronchiectasis type, and perform detailed feature extraction on each bronchiectasis in each type of bronchiectasis to obtain the length, position, bronchial-arterial ratio and bronchial wall thickening coefficient of the abnormal bronchial segment.

[0089] Measuring the length of abnormal bronchial segments is one of the important indicators for assessing the severity of bronchiectasis. Specifically, the module uses edge detection and path planning based methods to achieve this goal. First, the Canny edge detection algorithm is used to process the CT scan image to extract the boundary of the bronchiectasis area. The Canny edge detection algorithm can accurately detect the edge through Gaussian filtering, gradient calculation, non-maximum suppression and double threshold detection. Then, the path planning algorithm is used to find the shortest path on the extracted boundary line. The Dijkstra algorithm is applicable to graphs without negative weight edges and can find the shortest path from the starting point to the end point. The A algorithm introduces a heuristic function based on the Dijkstra algorithm, which can find the optimal path faster. Finally, the length of the abnormal bronchial segment is obtained by calculating the total length of the shortest path. This length value will be used as an important parameter for assessing the severity of bronchiectasis.

[0090] The location information of abnormal bronchial segments is also an important parameter for assessing the severity of bronchiectasis. The location information can help doctors understand the specific distribution of bronchiectasis in the lungs, so as to develop a more accurate treatment plan. Specifically, the module uses a coordinate system to quantify the location of the bronchiectasis area in the CT scan image. First, a two-dimensional or three-dimensional coordinate system is established in the CT scan image, and each pixel or voxel in the image is mapped to a point in the coordinate system. Then, the location of the bronchiectasis area in the coordinate system is annotated manually or automatically. Manual annotation is usually done by experienced doctors, while automatic annotation can be achieved through a deep learning model. Finally, the location information of the annotated bronchiectasis area is quantified. For example, the center coordinates and boundary coordinates of each bronchiectasis area can be recorded. This location information will serve as an important parameter for assessing the severity of bronchiectasis.

[0091] The bronchial-arterial ratio is another important parameter for assessing the severity of bronchiectasis. The bronchial-arterial ratio reflects the ratio of the diameters of the bronchus and the adjacent arteries and can be used to assess the extent of bronchiectasis. First, the bronchi and adjacent arteries are segmented from the CT scan images using a vessel segmentation algorithm (such as a level set-based method or a region growing-based method). These algorithms can accurately extract the contours and diameters of the vessels. Then, the bronchial-arterial ratio is calculated by measuring the diameters of the bronchi and adjacent arteries. Diameter measurement can be achieved by calculating the diameter of the minimum circumscribed circle of the vessel contour. Finally, the bronchial diameter is divided by the diameter of the adjacent artery to obtain the bronchial-arterial ratio. This ratio value will serve as an important parameter for assessing the severity of bronchiectasis.

[0092] The bronchial wall thickening coefficient is another important parameter for assessing the severity of bronchiectasis. The bronchial wall thickening coefficient reflects the degree of thickening of the bronchial wall and can be used to assess the severity of bronchiectasis. First, the bronchial wall is segmented from the CT scan image using a wall segmentation algorithm (such as a level set-based method or a region growing-based method). These algorithms can accurately extract the contour and thickness of the bronchial wall. Then, the bronchial wall thickening coefficient is calculated by measuring the thickness of the bronchial wall. Thickness measurement can be achieved by calculating the maximum distance of the bronchial wall contour. Finally, the bronchial wall thickness is divided by the thickness of the normal bronchial wall to obtain the bronchial wall thickening coefficient. This thickening coefficient value will serve as an important parameter for assessing the severity of bronchiectasis.

[0093] After completing the above feature extraction, the pathological severity estimation module will enter the next step, that is, calculating the bronchiectasis pathological severity coefficient based on the extracted features. It should be understood that it is a prior art to calculate the bronchiectasis pathological severity coefficient of each review based on the bronchiectasis type, the length and position of the abnormal bronchial segment, the bronchial-arterial ratio and the bronchial wall thickening coefficient. It can use the principle disclosed in Chinese patent CN116612891A to calculate the bronchiectasis pathological severity coefficient. Of course, it can also use other principles to calculate the bronchiectasis pathological severity coefficient, which is not limited to this application.

[0094] In order to facilitate doctors' understanding and use, the pathology severity estimation module can also display the calculation results in a visual way. For example, the area of ​​bronchiectasis can be marked in the CT scan image, and different levels of severity can be represented by different colors or icons. In addition, the module can also generate a detailed report, listing the specific values ​​and calculation results of each feature, to help doctors fully understand the patient's condition.

[0095] Through the above steps, the calculation of the bronchiectasis pathological severity coefficient of each patient's follow-up examination can be effectively realized, which not only improves the accuracy and efficiency of diagnosis, but also provides valuable decision-making support for doctors, helps to formulate more scientific and reasonable treatment plans, and improves patients' treatment effects and quality of life.

[0096] In summary, the artificial intelligence-based smart medical data processing system according to the embodiment of the present application is explained, which uses deep learning-based image processing technology to perform multi-level feature extraction on the CT scan images of chronic disease patients, and mines the texture features and structural features of the bronchiectasis area in the CT scan images, and performs fine-grained semantic interactive fusion of the texture features and structural features of the bronchiectasis state based on core correlation features to achieve a comprehensive understanding of the bronchiectasis state, and then intelligently identify the bronchiectasis type. In this way, the recognition accuracy of the bronchiectasis type can be effectively improved, thereby providing doctors with a more accurate diagnosis basis.

[0097] Furthermore, an intelligent medical data processing method based on artificial intelligence is also provided.

[0098] Figure 6 Flow chart of the method for processing intelligent medical data based on artificial intelligence according to an embodiment of the present application. Figure 6 As shown, the artificial intelligence-based smart medical data processing method includes the following steps: S1, performing bronchial target area detection on a CT scan image to obtain a bronchial target area image; S2, performing multi-scale feature extraction on the bronchial target area image to obtain a bronchial dilatation state texture feature map and a bronchial dilatation state structural feature map; S3, performing global semantic interactive encoding based on core joint features on the bronchial dilatation state texture feature map and the bronchial dilatation state structural feature map to obtain a bronchial dilatation state texture-structure fine-grained joint encoding feature map; S4, identifying the bronchiectasis type of the bronchiectasis based on the bronchial dilatation state texture-structure fine-grained joint encoding feature map.

[0099] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned intelligent medical data processing method based on artificial intelligence have been referred to above. Figures 1 to 5 The present invention has been introduced in detail in the description of the artificial intelligence-based smart medical data processing system, and therefore, its repeated description will be omitted.

[0100] The basic principle of the present invention is described above in conjunction with specific embodiments. However, it should be pointed out that the advantages, strengths, effects, etc. mentioned in the present invention are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. must be possessed by each embodiment of the present invention. In addition, the specific details of the above embodiments are only for the purpose of illustration and facilitation of understanding, rather than limitation, and the above details do not limit the present invention to being implemented by adopting the above specific details.

Claims

1. An intelligent medical data processing system based on artificial intelligence, comprising: A CT scan image acquisition module is used to set the duration of the monitoring cycle, and extract the CT scan images of each follow-up of each chronic bronchiectasis patient admitted to the designated hospital during the monitoring cycle from the patient information database of the designated hospital; a bronchiectasis type identification module is used to obtain the bronchiectasis type of each bronchiectasis in each follow-up CT scan image of each chronic disease patient according to the CT scan images of each follow-up; a pathological severity estimation module is used to classify the bronchiectasis in each follow-up CT scan image of each chronic disease patient according to the same bronchiectasis type, and calculate the bronchiectasis pathological severity coefficient of each follow-up based on the length, position, bronchial-arterial ratio and bronchial wall thickening coefficient of the abnormal bronchial segment corresponding to each bronchiectasis in each type of bronchiectasis in the CT scan images of each follow-up, characterized in that the bronchiectasis type identification module includes: a target region detection unit, configured to perform bronchial target region detection on the CT scan image to obtain a bronchial target region image; An expansion state feature extraction unit, used for performing multi-scale feature extraction on the bronchial target area image to obtain a bronchial expansion state texture feature map and a bronchial expansion state structural feature map; A fine-grained joint encoding unit, used for performing global semantic interactive encoding on the bronchiectasis state texture feature map and the bronchiectasis state structural feature map based on core joint features to obtain a bronchiectasis state texture-structure fine-grained joint encoding feature map; a bronchiectasis type determination unit, configured to identify the bronchiectasis type based on the bronchiectasis state texture-structure fine-grained joint coding feature map; Wherein, the fine-grained joint coding unit includes: A fine-grained common feature extraction subunit, used for performing fine-grained common feature extraction on the bronchiectasis state texture feature map and the bronchiectasis state structural feature map to obtain a set of bronchiectasis state texture-structure joint fine-grained core feature vectors; The global semantic interaction subunit is used to perform global interactive encoding based on semantic clustering centers on the set of the bronchial dilation state texture-structure joint fine-grained core feature vectors to obtain the bronchial dilation state texture-structure fine-grained joint encoding feature map.

2. The intelligent medical data processing system based on artificial intelligence according to claim 1 is characterized in that: The target area detection unit is used to: The CT scan image is input into a bronchial target area detection module based on the YOLO model to obtain the bronchial target area image.

3. The intelligent medical data processing system based on artificial intelligence according to claim 2 is characterized in that: The expansion state feature extraction unit comprises: A texture feature extraction subunit, configured to input the bronchial target area image into a texture feature extractor based on a first atrous convolutional neural network model to obtain the bronchial dilation state texture feature map; The structural feature extraction subunit is used to input the bronchial target area image into a structural feature extractor based on a second hole convolutional neural network model to obtain the structural feature map of the bronchial dilation state.

4. The intelligent medical data processing system based on artificial intelligence according to claim 3 is characterized in that: The fine-grained common feature extraction subunit includes: A feature fine-grained decoupling secondary subunit, used to perform feature fine-grained decoupling on the bronchiectasis state texture feature map and the bronchiectasis state structural feature map to obtain a set of bronchiectasis state local fine-grained texture feature vectors and a set of bronchiectasis state local fine-grained structural feature vectors; The common feature extraction secondary subunit is used to input each corresponding group of local fine-grained texture feature vectors of the bronchiectasis state and the local fine-grained structural feature vectors of the bronchiectasis state in the set of local fine-grained texture feature vectors of the bronchiectasis state and the set of local fine-grained structural feature vectors of the bronchiectasis state into the fine-grained common feature extraction network to obtain the set of the bronchiectasis state texture-structure joint fine-grained core feature vectors.

5. The intelligent medical data processing system based on artificial intelligence according to claim 4 is characterized in that: The common feature extraction secondary subunit is used for: Respectively calculating the position point subtraction, position point multiplication and position point addition between the local fine-grained texture feature vector of the bronchial dilation state and the local fine-grained structural feature vector of the bronchial dilation state to obtain a first dilation state texture-structure joint interaction representation vector, a second dilation state texture-structure joint interaction representation vector and a third dilation state texture-structure joint interaction representation vector; The first expansion state texture-structure joint interaction representation vector, the second expansion state texture-structure joint interaction representation vector and the third expansion state texture-structure joint interaction representation vector are cascaded and fused and input into a neural network layer based on a tanh function to obtain the bronchial dilation state texture-structure joint fine-grained core feature vector.

6. The intelligent medical data processing system based on artificial intelligence according to claim 5 is characterized in that: The global semantic interaction subunit is used to: Inputting the set of the bronchiectasis state texture-structure joint fine-grained core feature vectors into a clustering network to obtain a bronchiectasis state texture-structure joint feature semantic clustering center vector; Constructing a query vector and a value vector based on each bronchial dilation state texture-structure joint fine-grained core feature vector in the set of the bronchial dilation state texture-structure joint fine-grained core feature vectors, and constructing a key vector based on the bronchial dilation state texture-structure joint feature semantic clustering center vector, and inputting the query vector, the value vector and the key vector into a global semantic interaction encoder based on a converter structure to obtain a set of bronchial dilation state texture-structure context joint fine-grained core feature vectors; The set of the bronchial dilation state texture-structure context joint fine-grained core feature vectors is feature reshaped to obtain the bronchial dilation state texture-structure fine-grained joint encoding feature map.

7. The intelligent medical data processing system based on artificial intelligence according to claim 6 is characterized in that: The branch expansion type determination unit is used to: The bronchial dilatation state texture-structure fine-grained joint encoding feature map is input into a classifier-based dilatation type identifier to obtain the bronchial dilatation type of the bronchial dilatation.

8. An intelligent medical data processing method based on artificial intelligence, characterized in that: include: Performing bronchial target area detection on the CT scan image to obtain a bronchial target area image; Performing multi-scale feature extraction on the bronchial target area image to obtain a bronchial dilatation state texture feature map and a bronchial dilatation state structural feature map; Performing global semantic interactive encoding based on core joint features on the bronchiectasis state texture feature map and the bronchiectasis state structural feature map to obtain a bronchiectasis state texture-structure fine-grained joint encoding feature map; Based on the bronchiectasis state texture-structure fine-grained joint coding feature map, identifying the bronchiectasis type of the bronchiectasis; The bronchiectasis state texture feature map and the bronchiectasis state structural feature map are subjected to global semantic interactive encoding based on core joint features to obtain a bronchiectasis state texture-structure fine-grained joint encoding feature map, including: Performing fine-grained common feature extraction on the bronchiectasis state texture feature map and the bronchiectasis state structural feature map to obtain a set of bronchiectasis state texture-structure joint fine-grained core feature vectors; The set of the bronchial dilation state texture-structure joint fine-grained core feature vectors is subjected to global interactive encoding based on semantic clustering centers to obtain the bronchial dilation state texture-structure fine-grained joint encoding feature map.

Images