Esophageal manometry method, system and device based on big data

Abstract

The present invention provides an esophageal pressure measurement method, system and device based on big data, which relates to the field of medical image processing technology, including: obtaining historical detection data; performing three-dimensional reconstruction processing based on the historical detection data to obtain a three-dimensional model set; performing feature extraction based on the three-dimensional model set, integrating to obtain a structural feature set; performing simulation processing based on the structural feature set to obtain an esophageal pressure annotated data set; performing model construction processing based on the esophageal pressure annotated data set and the structural feature set, and obtaining an esophageal pressure prediction model by analyzing the mapping relationship between esophageal structural features and local pressure and training a neural network model; performing prediction processing on real-time endoscopic video based on the esophageal pressure prediction model to obtain a real-time predicted value of esophageal pressure. The present invention realizes non-invasive real-time prediction of esophageal pressure through direct analysis and three-dimensional reconstruction processing of endoscopic video data, combined with deep learning and mechanical simulation methods.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing technology, and in particular to an esophageal manometry method, system and device based on big data. Background Art

[0002] In the field of modern medicine, the diagnosis and monitoring of esophageal-related diseases are of vital importance for early detection and treatment. Current esophageal function detection technologies mainly include invasive esophageal manometry and non-invasive imaging detection methods. Traditional esophageal manometry uses a catheter inserted into the esophagus to measure pressure changes. Although this method provides more accurate esophageal function information, it often causes discomfort and pain to patients due to its invasiveness, affecting the patient's compliance and the stability of the diagnostic results. On the other hand, non-invasive imaging detection methods, such as ultrasound and magnetic resonance imaging (MRI), although they can alleviate the patient's discomfort to a certain extent, cannot effectively evaluate the dynamic pressure changes and structural characteristics of the esophagus because they can only provide a planar perspective and limited dynamic data.

[0003] The present application aims to provide an esophageal manometry method, system and device based on big data, which realizes real-time esophageal pressure prediction by combining endoscopic video with big data and deep learning algorithms, thereby eliminating the need for secondary detection. Summary of the invention

[0004] The purpose of the present invention is to provide an esophageal manometry method, system and device based on big data to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides an esophageal manometry method based on big data, comprising:

[0006] Acquiring historical test data, wherein the historical test data includes an endoscopic test video of at least one patient and a corresponding esophageal pressure value;

[0007] Performing three-dimensional reconstruction processing according to the historical detection data, extracting depth information from the video image, and performing frame alignment in combination with feature point registration to obtain a three-dimensional model set;

[0008] Performing feature extraction according to the three-dimensional model set, using a convolutional neural network to extract features from each three-dimensional model, extracting geometric information of the cardia, esophageal wall, and lower esophageal sphincter and converting them into structural feature vectors, and integrating to obtain a structural feature set;

[0009] Performing simulation processing according to the structural feature set, using finite element analysis and Navier-Stokes equation to simulate esophageal pressure distribution, calculating the effect of the opening diameter change of the cardia area on the esophageal pressure and evaluating the pressure distribution in the stenotic area, and obtaining an esophageal pressure annotated data set;

[0010] Performing model building processing according to the esophageal pressure annotated data set and the structural feature set, obtaining an esophageal pressure prediction model by analyzing the mapping relationship between esophageal structural features and local pressure and training a neural network model;

[0011] A real-time endoscopic video of the patient to be tested is obtained, and prediction processing is performed on the real-time endoscopic video based on the esophageal pressure prediction model to obtain a real-time prediction value of the esophageal pressure.

[0012] In a second aspect, the present application also provides an esophageal manometry system based on big data, including:

[0013] An acquisition module, used to acquire historical detection data, wherein the historical detection data includes an endoscopic detection video of at least one patient and a corresponding esophageal pressure value;

[0014] A reconstruction module is used to perform three-dimensional reconstruction processing according to the historical detection data, extract depth information from the video image, and perform frame alignment in combination with feature point registration to obtain a three-dimensional model set;

[0015] An extraction module, used to perform feature extraction based on the three-dimensional model set, perform feature extraction on each three-dimensional model using a convolutional neural network, extract geometric information of the cardia, esophageal wall and lower esophageal sphincter and convert it into a structural feature vector, and integrate to obtain a structural feature set;

[0016] A simulation module, used to perform simulation processing according to the structural feature set, use finite element analysis and Navier-Stokes equation to simulate the esophageal pressure distribution, calculate the influence of the opening diameter change of the cardia area on the esophageal pressure and evaluate the distribution of pressure in the stenosis area, so as to obtain an esophageal pressure annotated data set;

[0017] A construction module, used to perform model construction processing according to the esophageal pressure annotated data set and the structural feature set, and obtain an esophageal pressure prediction model by analyzing the mapping relationship between esophageal structural features and local pressure and training a neural network model;

[0018] The prediction module is used to obtain the real-time endoscopic video of the patient to be tested, and perform prediction processing on the real-time endoscopic video based on the esophageal pressure prediction model to obtain a real-time prediction value of the esophageal pressure.

[0019] In a third aspect, the present application also provides an esophageal manometry device based on big data, comprising:

[0020] Memory for storing computer programs;

[0021] A processor is used to implement the steps of the esophageal manometry method based on big data when executing the computer program.

[0022] The beneficial effects of the present invention are:

[0023] The present invention realizes non-invasive real-time prediction of esophageal pressure by directly analyzing and three-dimensionally reconstructing endoscopic video data, combined with deep learning and mechanical simulation methods. Compared with the traditional invasive catheter pressure measurement method, the present invention avoids patient discomfort and significantly improves the convenience and real-time performance of detection, eliminating the need for secondary detection. At the same time, by utilizing the historical pressure and structural feature data in big data and constructing a deep learning model, high-precision prediction of pressure in different areas of the esophagus can be achieved, thereby improving the accuracy of clinical diagnosis.

[0024] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0026] Figure 1 A schematic diagram of the process of the esophageal manometry method based on big data described in an embodiment of the present invention;

[0027] Figure 2 Schematic diagram of an esophageal manometry system based on big data described in an embodiment of the present invention;

[0028] Figure 3 Schematic diagram of the structure of the esophageal manometry device based on big data described in an embodiment of the present invention.

[0029] Markings in the figure: 1, acquisition module; 2, reconstruction module; 21, first calculation unit; 22, first detection unit; 23, first matching unit; 24, first reconstruction unit; 3, extraction module; 31, first segmentation unit; 32, first conversion unit; 33, first extraction unit; 34, first dimensionality reduction unit; 4, simulation module; 41, first processing unit; 42, first simulation unit; 421, first division unit; 422, second processing unit; 423, second simulation unit; 424, first analysis unit; 43. Second calculation unit; 44. Second extraction unit; 5. Construction module; 51. Third processing unit; 52. Third extraction unit; 53. First construction unit; 54. First optimization unit; 6. Prediction module; 61. Fourth extraction unit; 62. First integration unit; 63. First prediction unit; 64. Fourth processing unit; 800. Reinforcement equipment for corrugated plate reinforced tunnel structure; 801. Processor; 802. Memory; 803. Multimedia component; 804. I / O interface; 805. Communication component. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance. Embodiment 1:

[0032] This embodiment provides an esophageal manometry method based on big data.

[0033] See also Figure 1 , the figure shows that the method includes step S100, step S200, step S300, step S400, step S500 and step S600.

[0034] Step S100, obtaining historical test data, where the historical test data includes an endoscopic test video of at least one patient and a corresponding esophageal pressure value;

[0035] It is understandable that this step establishes a "big data" foundation that widely covers the individual differences of different patients by collecting a large number of endoscopic detection videos and corresponding esophageal pressure values ​​of real patients. These endoscopic videos can non-invasively record a variety of dynamic structural features of the esophagus, especially capturing its morphological changes in areas such as the cardia, esophageal wall and lower esophageal sphincter, providing real and dynamic visual data for subsequent feature extraction and model training. At the same time, the esophageal pressure value at each moment is synchronously matched with the video to form a high-dimensional, time-consistent data set, which provides the model with a dynamic correspondence between structure and pressure. In practical applications, through the accumulation of such massive data, the system has established a rich set of esophageal pressure samples. The coverage of big data enables the esophageal model to fully learn the diverse structural and pressure characteristics in the deep learning algorithm, enhancing the adaptability and generalization ability of the model.

[0036] Step S200, performing 3D reconstruction processing according to historical detection data, extracting depth information from the video image, and performing frame alignment in combination with feature point registration to obtain a 3D model set;

[0037] It should be noted that this step can obtain a three-dimensional perspective of the esophagus by converting the two-dimensional endoscopic video into a three-dimensional model set, which clearly reflects the layers and details of the esophageal structure. This three-dimensional model not only improves the integrity of spatial information, but also lays the foundation for subsequent feature extraction, pressure simulation and other steps.

[0038] Step S300, extracting features from the three-dimensional model set, using a convolutional neural network to extract features from each three-dimensional model, extracting geometric information of the cardia, esophageal wall, and lower esophageal sphincter and converting them into structural feature vectors, and integrating to obtain a structural feature set;

[0039] It can be understood that the generation of structural feature vectors condenses the geometric information of the cardia, esophageal wall, and lower esophageal sphincter into numerical data, making it suitable for subsequent deep learning model training and providing high-precision structural input for subsequent pressure prediction.

[0040] Step S400, performing simulation processing according to the structural feature set, using finite element analysis and Navier-Stokes equation to simulate esophageal pressure distribution, calculating the effect of the opening diameter change of the cardia area on the esophageal pressure and evaluating the pressure distribution in the stenotic area, and obtaining an esophageal pressure annotated data set;

[0041] It should be noted that the esophageal pressure annotation dataset can accurately reflect the pressure distribution of the esophagus under different states, providing the model with high-resolution real data annotations, helping the model to learn the complex relationship between structural features and pressure changes.

[0042] Step S500: constructing a model based on the esophageal pressure annotated data set and the structural feature set, analyzing the mapping relationship between the esophageal structural features and the local pressure and training the neural network model to obtain an esophageal pressure prediction model;

[0043] It should be noted that during the training process, the esophageal pressure annotation dataset provides a variety of samples of esophageal structural features and corresponding pressures, ensuring that the model can learn the complex relationship between the opening and closing of the cardia, the thickness and elasticity of the esophageal wall, and the dynamic response of the lower esophageal sphincter and the actual pressure. Furthermore, cross-validation and early stopping techniques are introduced in model training to improve the generalization ability of the model and prevent overfitting.

[0044] Step S600: obtaining a real-time endoscopic video of the patient to be tested, and performing prediction processing on the real-time endoscopic video based on the esophageal pressure prediction model to obtain a real-time prediction value of the esophageal pressure.

[0045] It is understandable that the prediction process combines the nonlinear mapping capabilities of the neural network, allowing the model to accurately reflect the changes in the opening and closing of the cardia and lower esophageal sphincter, and the impact of the elastic state of the esophageal wall on local pressure. By using a long short-term memory network (LSTM) or a similar time-dependent model, the temporal dynamic characteristics of the esophageal structure can be further captured to ensure the temporal consistency and accuracy of the pressure prediction.

[0046] Further, step S200 includes step S210 to step S240.

[0047] Step S210: according to the image sequence of the endoscope video in the historical detection data, the depth information in the video image is extracted using structured light analysis, and a fringe map is generated by performing difference calculation on the grayscale images of adjacent frames to obtain a depth map sequence;

[0048] It should be noted that this step projects stripe light into the endoscopic image, and uses the way the stripes are reflected and returned on the surface of the esophagus to capture depth data. In this process, the stripe deformation and displacement are detected by calculating the difference between the grayscale images of adjacent frames, and a stripe mapping image is generated. This method is suitable for the complex anatomical environment inside the esophagus. The inner surface of the esophagus is not flat and has a variety of shapes, especially in the cardia, lower esophageal sphincter and other areas where there are significant structural changes. Structured light analysis can accurately capture the local bending and stretching of stripes in such a complex inner cavity structure, and generate detailed depth information through stripe mapping.

[0049] Step S220, performing feature point detection processing according to the depth map sequence, performing SIFT feature point detection on the cardia, esophageal wall and lower esophageal sphincter in the esophageal detection area and generating descriptors to obtain a feature point set;

[0050] It can be understood that this step applies the SIFT (Scale Invariant Feature Transform) algorithm to detect feature points in key areas of the depth map sequence. The SIFT algorithm can detect and extract stable feature points at different scales, and is particularly suitable for detecting morphological details in areas such as the cardia and lower esophageal sphincter. For each feature point, SIFT generates a feature descriptor containing information such as position, direction, and scale, so that the feature point has a high matching accuracy in subsequent registration and three-dimensional reconstruction. The application of the SIFT algorithm can adapt to the complexity of these areas and capture the detailed changes in esophageal wall thickness, cardia contraction, and lower esophageal sphincter expansion through multi-scale analysis.

[0051] Step S230, performing feature point matching processing according to the feature point set, aligning feature points of adjacent frames through FLANN matching and rigid registration based on the least squares method, and obtaining aligned feature point pairs;

[0052] It should be noted that this step first uses the FLANN (Fast Nearest Neighbor Search) algorithm to match the feature point set. The FLANN algorithm can quickly search and match similar feature points in adjacent frames, so that the feature points can find the best pair in a large amount of data, thereby accelerating the matching process. Next, a rigid registration method based on the least squares method is applied to finely align the spatial positions of the matched feature point pairs. Rigid registration ensures that the feature points in adjacent frames maintain spatial consistency in the best alignment by minimizing the error between feature points.

[0053] Step S240: Perform three-dimensional reconstruction processing based on the aligned feature point pairs, generate a three-dimensional structure of the detection area through Delaunay triangulation and interpolation, and integrate to obtain a three-dimensional model set.

[0054] Specifically, this step first applies the Delaunay triangulation algorithm to spatially divide the aligned feature points to generate a three-dimensional mesh that conforms to the internal structure of the esophagus. Delaunay triangulation can construct an optimal triangular network on a given feature point to ensure that the angle of each triangle is maximized, thereby generating a stable and uniform mesh structure, which is particularly important when capturing surface changes in the esophagus (such as the cardia and lower esophageal sphincter area). Then, the blank areas in the mesh are filled and refined by the interpolation method, so that the generated three-dimensional model can smoothly reflect the geometric features and morphological changes inside the esophagus. The interpolation process completes the depth information of the uncovered area to improve the detail quality and accuracy of the three-dimensional reconstruction.

[0055] Further, step S300 includes step S310 to step S340.

[0056] Step S310, performing structural segmentation processing according to the three-dimensional model set, determining the relative positions and boundary differences of the cardia, esophageal wall and lower esophageal sphincter by combining three-dimensional curvature analysis, and obtaining a structural segmentation map of the detection area;

[0057] It is understandable that the curvature and morphology of the structures in the esophagus vary significantly, especially in the cardia and lower esophageal sphincter, where the dynamic changes and boundary curvature characteristics are extremely complex. Through the three-dimensional curvature analysis method, the boundaries of these parts can be effectively distinguished, the relative positions of the cardia and lower esophageal sphincter can be determined, and clear regional segmentation lines can be generated. The final structural segmentation map can accurately present the spatial layout of the various layers of the internal structure of the esophagus, laying a clear regional division foundation for subsequent geometric feature extraction and pressure simulation.

[0058] Step S320, extracting geometric features and gridding according to the structure segmentation map, by converting the cardia, esophageal wall and lower esophageal sphincter into regular grids, extracting the dynamic range of the opening diameter in the cardia area, extracting the local gradient of the thickness in the esophageal wall area, and extracting the curvature and texture structure of the surface in the lower esophageal sphincter area, to obtain a geometric feature matrix;

[0059] Specifically, the segmented cardia, esophageal wall and lower esophageal sphincter regions are first converted into regular grids. This gridding step can provide a unified benchmark, which facilitates the subsequent accurate extraction and processing of features. The regular grid evenly divides the geometric information of each region, ensuring that the morphological changes of different parts can be accurately captured in subsequent calculations.

[0060] In the cardia region, the dynamic range of the opening diameter is extracted by analyzing the changes in the positions of the grid nodes. This feature directly reflects the opening and closing movement of the cardia and is an important indicator for evaluating the esophageal swallowing function and changes in cardia pressure. In the esophageal wall region, the local gradient of thickness is extracted to reveal the elastic change characteristics of the esophageal wall. Local gradient information is suitable for the detection of esophageal wall thickness changes under dynamic conditions, which helps to understand the response characteristics of this area under pressure. For the lower esophageal sphincter region, the curvature and texture structure of the surface are extracted to capture the morphological characteristics of the sphincter and the arrangement of muscle fibers. The curvature and texture of the lower esophageal sphincter can reveal the contraction and expansion patterns of the sphincter, thereby providing detailed data on its functional status.

[0061] Step S330, performing deep feature extraction processing according to the geometric feature matrix, and obtaining a structural feature vector by using a convolutional neural network to learn the contraction and expansion patterns of the cardia and lower esophageal sphincter, the thickness gradient and elasticity distribution of the esophageal wall;

[0062] In this step, a convolutional neural network (CNN) is used to learn the key dynamic characteristics of the cardia, lower esophageal sphincter, and esophageal wall from the geometric feature matrix. First, CNN extracts features of the contraction and expansion patterns of the cardia and lower esophageal sphincter through layer-by-layer convolution and pooling operations. These patterns reflect the dynamic changes of the opening and closing of the cardia and the activity of the sphincter, and are important indicators for describing the functional activity of the esophagus. In addition, CNN further analyzes the thickness gradient and elastic distribution of the esophageal wall. By extracting the thickness gradient, the network can identify the thickness changes of the esophageal wall in different regions, thereby capturing the elastic characteristics of the local region. The elastic distribution of the esophageal wall is a key factor in responding to pressure changes and can provide information on the adaptability of the structure to pressure.

[0063] Step S340: Perform dimensionality reduction processing on the structural feature vector, and obtain a set of structural features by screening features that are sensitive to pressure changes.

[0064] In this process, dimensionality reduction algorithms such as principal component analysis (PCA) or recursive feature elimination (RFE) are used to reduce high-dimensional feature vectors to the most representative feature subsets. Dimensionality reduction can effectively reduce the redundancy of data, making the model more efficient in subsequent calculations, while retaining the features most closely related to pressure changes. When screening features, priority is given to the contraction and expansion characteristics of the cardia and lower esophageal sphincter, the esophageal wall thickness gradient, and other parts that are most sensitive to pressure changes. This dimensionality reduction process can make the model more focused on the core dynamic features related to esophageal pressure changes, thereby improving the accuracy and response speed of the model in pressure prediction.

[0065] Further, step S400 includes step S410 to step S440.

[0066] Step S410, initialization processing is performed according to the set of structural features, the boundary conditions of the pressure simulation are set using the discrete boundary method, the pressure and velocity boundaries of the cardia area are set by introducing a dynamic opening and closing control function according to the change characteristics of the cardia opening diameter, the elastic coefficient of the esophageal wall is set according to the thickness characteristics of the esophageal wall, and the gradient change of the pressure boundary value is defined in the lower esophageal sphincter to simulate the airflow resistance, and the simulation initial condition matrix is ​​obtained;

[0067] Specifically, the discrete boundary method is used to set the boundary conditions of pressure simulation. The discrete boundary method can accurately define the behavior of pressure changes in complex boundaries. In the cardia region, based on the change characteristics of its opening diameter, a dynamic opening and closing control function is introduced to reflect its opening and closing during swallowing or contraction. Through this control function, the pressure and velocity boundaries of the cardia region are dynamically adjusted so that the model can flexibly adapt to the real-time motion state of the cardia. For the esophageal wall region, the elastic coefficient of the esophageal wall is set using the thickness feature to simulate its response characteristics under pressure. The thickness and elasticity of the esophageal wall directly affect the conduction effect of pressure, so the accurate setting of the elastic coefficient can improve the simulation accuracy of the model to the real pressure response of the esophageal wall. For the lower esophageal sphincter, by defining the gradient change of the pressure boundary value, the resistance of the sphincter to the airflow is simulated, thereby reflecting the pressure changes of the sphincter under different states. The simulation initial condition matrix generated by this initialization process provides accurate boundary and mechanical conditions for pressure simulation, ensuring that the model can truly reproduce the airflow resistance state and pressure distribution in the complex esophageal dynamic environment.

[0068] Step S420, constructing a model according to the simulation initial condition matrix, simulating the dynamic influence of airflow and cardia opening and closing on the surrounding pressure by the lattice Boltzmann method, and obtaining an esophageal pressure simulation model;

[0069] It can be understood that the lattice Boltzmann method is a numerical calculation method for fluid mechanics simulation, which is particularly suitable for dealing with complex boundary conditions and irregular geometric structures, such as the cardia and lower esophageal sphincter in the esophagus. In this step, the simulation initial condition matrix is ​​first input into the lattice Boltzmann model to initialize the velocity distribution and density state of the lattice nodes. The lattice Boltzmann model simulates the propagation of airflow and the transmission of pressure in the esophagus by performing collision and migration processes on each lattice node. The collision process adjusts the velocity distribution function through the BGK (Bhatnagar-Gross-Krook) approximation so that the node reaches local equilibrium; while the migration process transfers the velocity distribution function to the adjacent lattice nodes to complete the global propagation of pressure in the esophagus. In the cardia area, the lattice Boltzmann method is combined with a dynamic opening and closing control function to simulate the effect of the cardia on the surrounding pressure during swallowing or expansion. Since the opening and closing of the cardia will cause rapid changes in local pressure, this method can update the pressure distribution in the cardia area in real time, reflecting the pressure changes in the esophagus and the pressure fluctuations and flow characteristics generated during the opening and closing process.

[0070] Step S430, performing pressure field solution processing according to the esophageal pressure simulation model, decomposing the pressure field by applying fast Fourier transform, extracting high-frequency pressure changes and low-frequency trend components in the cardia area, and generating pressure field distribution data;

[0071] It can be understood that this step applies the fast Fourier transform (FFT) to perform frequency decomposition on the pressure field data in the pressure simulation model. As an efficient frequency domain analysis method, FFT can decompose complex pressure time series into components of different frequencies, thereby identifying the main mode of pressure change. Specifically, in the cardia area, FFT extracts high-frequency and low-frequency pressure components to analyze the pressure changes in this area under different states. The high-frequency component reflects the instantaneous pressure fluctuations generated by the rapid opening and closing of the cardia. This feature is particularly important for capturing the pressure response during swallowing or rapid contraction; while the low-frequency component represents the average pressure trend of the cardia on a longer time scale, such as the pressure distribution changes during chronic contraction or relaxation of the cardia. This decomposition enables the model to focus on both the instantaneous dynamic changes of the cardia and its stability trend.

[0072] Step S440: perform feature labeling processing according to the pressure field distribution data, analyze the pressure change trend of the cardia and the lower esophageal sphincter through wavelet decomposition, extract dynamic change features and convert them into an esophageal pressure labeling data set.

[0073] It should be noted that wavelet decomposition can decompose the pressure signal into different frequency and time components, so as to capture the pressure changes of the cardia and lower esophageal sphincter at different time scales. Compared with fast Fourier transform, wavelet decomposition has higher time resolution and can accurately locate instantaneous pressure fluctuations and continuous pressure trends. In the cardia area, wavelet decomposition captures the short-term characteristics of rapid pressure fluctuations during its opening and closing process, and extracts the overall trend of pressure changes over a longer period of time. In the lower esophageal sphincter area, wavelet decomposition can reveal the pressure response pattern of contraction and expansion of this part, so that the model can accurately describe the pressure behavior of the sphincter area under different states. Finally, these extracted dynamic change features are converted into an esophageal pressure annotation dataset, and the multi-level pressure features of each key area are digitized.

[0074] Further, step S420 includes step S421 to step S424.

[0075] Step S421, performing grid initialization processing according to the simulation initial condition matrix, by spatially gridding the cardia, esophageal wall and lower esophageal sphincter region and setting boundary nodes, to obtain a discrete lattice Boltzmann simulation grid;

[0076] It should be noted that spatial meshing can not only effectively capture the geometric characteristics of each region in the esophagus, but also provide the discrete spatial basis required for the calculation of the lattice Boltzmann method (LBM). LBM is particularly suitable for airflow simulation with complex geometry, and the discreteness of the mesh enables it to flexibly adapt to the irregular anatomical structure in the esophagus. During the meshing process, boundary nodes are set specifically for the cardia, esophageal wall, and lower esophageal sphincter. The dynamic opening and closing of the cardia boundary nodes, the elastic response of the esophageal wall, and the pressure resistance effect of the sphincter are all precisely controlled through these boundary nodes. These nodes can dynamically reflect the state changes of each region during the simulation process, such as the contraction of the cardia during swallowing, or the response of the sphincter under different pressure conditions.

[0077] Step S422, performing velocity distribution function initialization processing according to the lattice Boltzmann simulation grid, and obtaining an initial velocity distribution matrix by setting the initial airflow velocity and density distribution function to reflect the initial esophageal pressure state of the cardia and esophageal wall;

[0078] It can be understood that by initializing the airflow velocity distribution, the pressure distribution of the cardia, esophageal wall and lower esophageal sphincter under static or initial conditions can be reflected. This step ensures that at the beginning of the simulation, the airflow and pressure states in each region can accurately reflect the actual physiological conditions. In the specific processing, the BGK (Bhatnagar-Gross-Krook) approximation model is used to adjust the velocity distribution function of each node to achieve the initial equilibrium state, especially in the cardia area, the initial velocity and density that adapt to its opening and closing dynamics are set to ensure that the initial pressure characteristics at the cardia can be reflected. For the esophageal wall area, the elastic response is simulated by density distribution, which provides a basis for pressure conduction and airflow in the simulation.

[0079] Specifically, in the Lattice Boltzmann Method (LBM), the velocity distribution function Indicates that each grid node is along different directions The distribution of gas particles in the initial state is expressed by the following formula:

[0080] ;

[0081] in, represents the initial velocity distribution function; represents the equilibrium attitude distribution function; Indicates direction The weight coefficient is related to the number of directions of the grid; Indicates density; Indicates along the direction The discrete velocity vector of represents the initial velocity vector of the node; Indicates the speed of sound.

[0082] In order to reflect the dynamic state of the cardia opening and closing, esophageal wall elasticity and sphincter, the initial density and speed settings of each area are different. Assume that the initial density of the cardia is The esophageal wall is The lower esophageal sphincter is , the initial density and velocity of each region are defined by the following formula:

[0083] ;

[0084] ;

[0085] in, It represents the static reference density, reflecting the basic pressure level of the esophagus; , and are the density adjustment coefficients of the cardia, esophageal wall, and sphincter, respectively, simulating the initial elasticity and density differences of each region; and Represents the angle variable that controls the cardia opening and closing state and sphincter pressure; and are the initial velocities of the cardia and sphincter, reflecting the opening and closing of the cardia and the tension of the sphincter at the beginning of swallowing, respectively; represents the initial velocity vector of the cardia region; represents the initial velocity vector of the esophageal wall region; Represents the initial velocity vector of the lower esophageal sphincter region.

[0086] Step S423, performing airflow evolution processing according to the initial velocity distribution matrix, simulating the dynamic evolution of airflow in the cardia and lower esophageal sphincter region through the fluid collision and migration process in the lattice Boltzmann method, and obtaining velocity field-pressure field distribution data;

[0087] During this process, LBM distributes the airflow on discrete grid nodes and realizes the natural evolution of the airflow through collision and migration mechanisms. The collision process adjusts the velocity distribution of each node according to the BGK (Bhatnagar-Gross-Krook) approximation model so that it reaches equilibrium in the local area and reflects the pressure difference between nodes. The migration process transmits the airflow information on the node to the adjacent nodes to realize the flow of airflow in the entire grid. In the cardia area, the airflow is affected by the dynamic opening and closing of the cardia. The collision and migration process of LBM can accurately simulate the effect of the opening and closing of the cardia on the airflow velocity and pressure, and capture the pressure fluctuations and flow rate changes in this area. The simulation of the lower esophageal sphincter reflects the regulatory effect of its contraction and expansion on the airflow resistance, and the pressure changes and flow rate field distribution in the sphincter area are displayed through the migration process.

[0088] Specifically, the collision process is based on the BGK model by adjusting the velocity distribution function Distribute it to equilibrium Get closer. The collision process formula is as follows:

[0089] ;

[0090] in, Indicates at location Along the direction The velocity distribution function of represents the equilibrium distribution function of the node; is the relaxation time, which is used to control the rate at which the collision process reaches equilibrium; Indicates the current time point; Represents a time increment.

[0091] Step S424: Analyze and process the velocity field-pressure field distribution data, analyze the high-frequency and low-frequency components in the pressure distribution according to the dynamic response characteristics of the cardiac contraction, esophageal wall elasticity and lower esophageal sphincter, identify the pressure change pattern, and construct an esophageal pressure simulation model.

[0092] It is understandable that the high-frequency components mainly reflect the instantaneous pressure fluctuations generated during the rapid contraction and opening and closing of the cardia, which is particularly critical for capturing the pressure response during swallowing. The low-frequency components reveal the slow changes in the elasticity of the esophageal wall under continuous pressure, as well as the long-term pressure trend and stability of the lower esophageal sphincter. In the cardia region, how the contraction and opening and closing movements affect the distribution of the local pressure field is analyzed to extract its dynamic characteristics. For the esophageal wall, its thickness and elastic characteristics are evaluated in response to different pressure conditions, and its morphological changes under high and low frequency pressures are identified. The analysis of the lower esophageal sphincter focuses on its regulatory effect on airflow and observes how it affects the pressure field during expansion and contraction.

[0093] Further, step S500 includes step S510 to step S540.

[0094] Step S510: performing data alignment processing according to the esophageal pressure annotated data set and the structural feature set, aligning the pressure data and the structural features in the time axis and space through time synchronization and space interpolation methods to obtain a spatiotemporal feature matrix;

[0095] Specifically, a time synchronization method is applied to align the pressure data with the set of structural features on the time axis to ensure that they correspond to each other at the same time point. This step is crucial to capturing the dynamic changes of the cardia, esophageal wall, and lower esophageal sphincter, because the structural features at each moment need to correspond to accurate pressure data to reflect the real-time physical state. Interpolation methods are used in space for spatial alignment to accurately map the discrete positions of the pressure data to the points or regions in the structural features. Using cubic splines or linear interpolation, the gaps between spatially distributed data can be filled to form continuous spatial features, so that the pressure state of each spatial node accurately matches its geometric position.

[0096] Step S520: performing feature extraction processing based on spatiotemporal feature evidence, using a one-dimensional convolutional neural network to perform convolution operations on multiple time windows, respectively extracting the cardia opening and closing frequency, the contraction pattern of the lower esophageal sphincter, and the effects of the esophageal wall elastic gradient and thickness changes on the esophageal pressure, and performing vector stacking to obtain a multi-scale feature vector;

[0097] It is understandable that the features such as the frequency of cardia opening and closing, the contraction pattern of the lower esophageal sphincter, and the thickness gradient of the esophageal wall extracted by the one-dimensional convolutional neural network (1D-CNN) have unique dynamic patterns on their respective time scales. After extracting these short-term and long-term features respectively through convolution operations of short, medium, and long time windows, vector stacking unifies these features of different time scales into a high-dimensional vector to form a multi-scale feature vector. The stacking operation ensures that the features of each time scale occupy a specific dimension in the final feature vector, allowing the model to access and process these multi-level information simultaneously. In this way, the instantaneous opening and closing of the cardia, the periodic contraction and expansion of the sphincter, and the elastic adaptability of the esophageal wall are effectively integrated together to form a feature vector that comprehensively reflects the effect of the esophagus on pressure at different time scales.

[0098] Step S530, performing nonlinear mapping construction processing according to the multi-scale feature vector, modeling the contraction frequency of the cardia, the expansion characteristics of the lower esophageal sphincter and the nonlinear effect of the esophageal wall elasticity change on the local pressure through a bidirectional long short-term memory network, and obtaining a preliminary structure-pressure mapping model;

[0099] In this step, the bidirectional long short-term memory network (Bi-LSTM) learns the contraction frequency of the cardia from the multi-scale feature vector, and can accurately model the periodic impact of the opening and closing of the cardia on the surrounding pressure. In addition, Bi-LSTM also captures the expansion and contraction characteristics of the lower esophageal sphincter, and generates a dynamic response pattern of the sphincter to airflow regulation, so that the model can understand the changes in the resistance of the sphincter under different states. For the elastic changes of the esophageal wall, Bi-LSTM can identify the response behavior of its thickness gradient under pressure fluctuations, and further learn the elastic adaptability of the esophageal wall in time. The bidirectional modeling of these features enables Bi-LSTM to capture the impact of the current features on pressure, and to more comprehensively understand the dynamic pattern through the information of the previous and next time steps.

[0100] Step S540: Optimize the preliminary structure-pressure mapping model, train and tune the mapping model through a multi-layer deep neural network, and obtain a final esophageal pressure prediction model.

[0101] Specifically, a multi-layer deep neural network (DNN) is used to train the preliminary mapping model, and the model parameters are adjusted through the back-propagation algorithm and the gradient descent method to minimize the error between the predicted pressure value and the actual labeled pressure. The multi-layer structure of DNN enables it to learn and represent highly complex nonlinear relationships. During the training process, the model performance is optimized by adjusting the model's hyperparameters such as the number of layers, the number of nodes, and the activation function. Regularization techniques such as L2 regularization or dropout are used to avoid overfitting of the model and ensure its generalization ability on different data sets. Cross-validation during the tuning process is used to evaluate the performance of the model on different sample sets to further enhance the robustness of the model.

[0102] Further, step S600 includes step S610 to step S640.

[0103] Step S610: extract image features based on the real-time endoscope video, use a convolutional neural network to process the video frames frame by frame, extract the geometric features of the cardia, lower esophageal sphincter and esophageal wall, and obtain a real-time feature matrix;

[0104] It should be noted that each frame of endoscopic image contains key structural information inside the esophagus. The edge, morphology and local geometric features in these images are extracted layer by layer using a convolutional neural network (CNN), capturing the degree of cardia opening and closing, the thickness gradient and elastic properties of the esophageal wall, and the contraction state of the lower esophageal sphincter. During the feature extraction process, CNN identifies important feature points in the image through convolution and pooling operations, ensuring that stable and clear geometric information is extracted in the complex dynamic environment of the esophagus.

[0105] Step S620: Perform time series feature integration processing according to the real-time feature matrix, use a long short-term memory network to convert the real-time feature matrix into a sequence feature with time correlation, and obtain a time series structure feature;

[0106] It should be noted that the long short-term memory network (LSTM) is a recursive neural network that is good at capturing long-term dependencies in time series data. It controls the flow of information through gating mechanisms (input gate, forget gate, and output gate), which can retain historical information about the changes in esophageal geometric features over time and suppress irrelevant or outdated information. When processing the real-time feature matrix, LSTM gradually receives the geometric feature input of each frame, integrating the dynamic changes of the opening and closing of the cardia, the contraction pattern of the lower esophageal sphincter, and the elastic changes of the esophageal wall. This temporal integration can identify the evolution pattern of features over time, such as the coordinated contraction of the cardia and sphincter during swallowing, or the slow adaptation process of the esophageal wall under different pressures.

[0107] Step S630: using the time series structure features as the input value of the esophageal pressure prediction model, and performing time point prediction by mapping the relationship between the cardia, lower esophageal sphincter and esophageal wall features and local pressure through a multi-layer perception mechanism and an activation function nonlinear mapping, to obtain a pressure prediction value at each time point;

[0108] Specifically, the multi-layer perception mechanism in the model uses a fully connected layer to process the time series features layer by layer, and gradually refines the nonlinear association between the structural features of each region and the pressure change. The activation function (such as ReLU or sigmoid) performs nonlinear mapping on the input data in each layer, so that the model can capture the complex effects of the opening and closing of the cardia, changes in esophageal wall thickness, and contraction of the lower esophageal sphincter on pressure. After the feature input at each time point is mapped through multiple layers, the model outputs the pressure prediction value at the corresponding moment. This time-point prediction method can not only reflect the pressure dynamics inside the esophagus in real time, but also ensure the synchronization of the prediction with the real-time feature input, so that the model can quickly adjust the prediction value according to the changes in the input cardia, esophageal wall, and sphincter.

[0109] Step S640: Smoothing the predicted pressure value at each time point, taking a weighted average of the predicted value at each time point, and calculating a real-time predicted value of esophageal pressure.

[0110] It should be noted that the weighted average method generates a smooth output by assigning weights to the predicted values ​​at each time point, performing weighted calculations on the predicted values ​​at a series of time points according to the weights. The weight distribution is usually set according to the length of the time window and the degree of smoothing required to ensure that the smoothed results can retain important dynamic features while avoiding errors caused by short-term anomalies. Through smoothing, unstable short-term fluctuations in the pressure values ​​output by the model can be avoided, thereby more realistically reflecting the pressure change trend inside the esophagus. Embodiment 2:

[0111] like Figure 2 As shown, this embodiment provides an esophageal manometry system based on big data, and the system includes:

[0112] An acquisition module 1 is used to acquire historical test data, where the historical test data includes an endoscopic test video of at least one patient and a corresponding esophageal pressure value;

[0113] Reconstruction module 2 is used to perform 3D reconstruction processing based on historical detection data, extract depth information from video images, and perform frame alignment in combination with feature point registration to obtain a 3D model set;

[0114] Extraction module 3, used for extracting features according to the three-dimensional model set, using convolutional neural network to extract features from each three-dimensional model, extracting geometric information of the cardia, esophageal wall and lower esophageal sphincter and converting them into structural feature vectors, and integrating to obtain a structural feature set;

[0115] Simulation module 4 is used to perform simulation processing according to the structural feature set, use finite element analysis and Navier-Stokes equation to simulate the esophageal pressure distribution, calculate the influence of the opening diameter change of the cardia area on the esophageal pressure and evaluate the distribution of pressure in the stenosis area, and obtain the esophageal pressure annotation data set;

[0116] Construction module 5, used to perform model construction processing according to the esophageal pressure annotated data set and the structural feature set, and obtain the esophageal pressure prediction model by analyzing the mapping relationship between the esophageal structural features and the local pressure and training the neural network model;

[0117] The prediction module 6 is used to obtain the real-time endoscopic video of the patient to be tested, and perform prediction processing on the real-time endoscopic video based on the esophageal pressure prediction model to obtain a real-time prediction value of the esophageal pressure.

[0118] In a specific embodiment disclosed in the present application, the reconstruction module 2 includes:

[0119] A first calculation unit 21 is used to extract depth information in the video image by using structured light analysis according to the image sequence of the endoscope video in the historical detection data, generate a fringe map by performing difference calculation on the grayscale images of adjacent frames, and obtain a depth map sequence;

[0120] A first detection unit 22 is used to perform feature point detection processing according to the depth map sequence, and obtain a feature point set by performing SIFT feature point detection and generating descriptors on the cardia, esophageal wall and lower esophageal sphincter in the esophageal detection area;

[0121] A first matching unit 23 is used to perform feature point matching processing according to the feature point set, align feature points of adjacent frames through FLANN matching and rigid registration based on the least square method, and obtain aligned feature point pairs;

[0122] The first reconstruction unit 24 is used to perform 3D reconstruction processing according to the aligned feature point pairs, generate a 3D structure of the detection area through Delaunay triangulation and interpolation, and integrate to obtain a 3D model set.

[0123] In a specific implementation disclosed in the present application, the extraction module 3 includes:

[0124] The first segmentation unit 31 is used to perform structural segmentation processing according to the three-dimensional model set, determine the relative positions and boundary differences of the cardia, esophageal wall and lower esophageal sphincter by combining three-dimensional curvature analysis, and obtain a structural segmentation map of the detection area;

[0125] The first conversion unit 32 is used to perform geometric feature extraction and grid processing according to the structure segmentation map, by converting the cardia, esophageal wall and lower esophageal sphincter into regular grids, extracting the dynamic change range of the opening diameter in the cardia area, extracting the local gradient of the thickness in the esophageal wall area, and extracting the curvature and texture structure of the surface in the lower esophageal sphincter area, to obtain a geometric feature matrix;

[0126] A first extraction unit 33 is used to perform deep feature extraction processing according to the geometric feature matrix, and obtain a structural feature vector by learning the contraction and expansion mode of the cardia and lower esophageal sphincter, the thickness gradient and elasticity distribution of the esophageal wall using a convolutional neural network;

[0127] The first dimension reduction unit 34 is used to perform dimension reduction processing according to the structural feature vector, and obtain a structural feature set by screening features that are sensitive to pressure changes.

[0128] In a specific implementation disclosed in the present application, the simulation module 4 includes:

[0129] The first processing unit 41 is used to perform initialization processing according to the set of structural features, set the boundary conditions of pressure simulation by using the discrete boundary method, introduce the dynamic opening and closing control function according to the change characteristics of the cardia opening diameter to set the pressure and velocity boundaries of the cardia area, set the elastic coefficient of the esophageal wall according to the thickness characteristics of the esophageal wall, and define the gradient change of the pressure boundary value in the lower esophageal sphincter to simulate the airflow resistance, so as to obtain the simulation initial condition matrix;

[0130] The first simulation unit 42 is used to perform model construction processing according to the simulation initial condition matrix, and simulate the dynamic influence of airflow and cardia opening and closing on the esophageal peri-esophageal pressure by the lattice Boltzmann method to obtain an esophageal pressure simulation model;

[0131] The first calculation unit 43 is used to perform pressure field solution processing according to the esophageal pressure simulation model, decompose the pressure field by applying fast Fourier transform, extract the high-frequency pressure change and low-frequency trend components of the cardia area, and generate pressure field distribution data;

[0132] The second extraction unit 44 is used to perform feature labeling processing according to the pressure field distribution data, analyze the pressure change trend of the cardia and the lower esophageal sphincter through wavelet decomposition, extract dynamic change features and convert them into esophageal pressure labeling data sets.

[0133] In a specific implementation disclosed in the present application, the first simulation unit 42 includes:

[0134] The first division unit 421 is used to perform grid initialization processing according to the simulation initial condition matrix, and obtain a discrete lattice Boltzmann simulation grid by spatially dividing the cardia, esophageal wall and lower esophageal sphincter region and setting boundary nodes;

[0135] The second processing unit 422 is used to perform velocity distribution function initialization processing according to the lattice Boltzmann simulation grid, and obtain an initial velocity distribution matrix by setting the initial airflow velocity and density distribution function to reflect the initial esophageal pressure state of the cardia and esophageal wall;

[0136] The second simulation unit 423 is used to perform airflow evolution processing according to the initial velocity distribution matrix, simulate the dynamic evolution of airflow in the cardia and lower esophageal sphincter region through the fluid collision and migration process in the lattice Boltzmann method, and obtain velocity field-pressure field distribution data;

[0137] The first analysis unit 424 is used to analyze and process the velocity field-pressure field distribution data, analyze the high-frequency and low-frequency components in the pressure distribution according to the dynamic response characteristics of the cardiac contraction, esophageal wall elasticity and lower esophageal sphincter, identify the pressure change pattern, and construct an esophageal pressure simulation model.

[0138] In a specific embodiment disclosed in the present application, the building block 5 includes:

[0139] The third processing unit 51 is used to perform data alignment processing according to the esophageal pressure annotated data set and the structural feature set, and align the pressure data and the structural features in the time axis and space through time synchronization and space interpolation methods to obtain a spatiotemporal feature matrix;

[0140] The third extraction unit 52 is used to perform feature extraction processing according to the spatiotemporal feature evidence, and uses a one-dimensional convolutional neural network to perform convolution operations on multiple time windows to respectively extract the cardia opening and closing frequency, the contraction mode of the lower esophageal sphincter, and the influence of the esophageal wall elastic gradient and thickness change on the esophageal pressure, and perform vector stacking to obtain a multi-scale feature vector;

[0141] A first construction unit 53 is used to perform nonlinear mapping construction processing according to the multi-scale feature vector, and to model the contraction frequency of the cardia, the expansion characteristics of the lower esophageal sphincter, and the nonlinear effect of the elasticity change of the esophageal wall on the local pressure through a bidirectional long short-term memory network to obtain a preliminary structure-pressure mapping model;

[0142] The first optimization unit 54 is used to perform optimization processing according to the preliminary structure-pressure mapping model, train and tune the mapping model through a multi-layer deep neural network, and obtain a final esophageal pressure prediction model.

[0143] In a specific implementation disclosed in the present application, the prediction module 6 includes:

[0144] The fourth extraction unit 61 is used to extract image features according to the real-time endoscope video, process the video frames frame by frame using a convolutional neural network, extract the geometric features of the cardia, the lower esophageal sphincter and the esophageal wall, and obtain a real-time feature matrix;

[0145] The first integration unit 62 is used to perform time series feature integration processing according to the real-time feature matrix, and use a long short-term memory network to convert the real-time feature matrix into a sequence feature with time correlation to obtain a time series structure feature;

[0146] The first prediction unit 63 is used to use the time series structure feature as the input value of the esophageal pressure prediction model, and to predict the relationship between the cardia, lower esophageal sphincter and esophageal wall features and the local pressure through a multi-layer perception mechanism and an activation function nonlinear mapping, so as to obtain a pressure prediction value at each time point;

[0147] The fourth processing unit 64 is used to perform smoothing processing according to the pressure prediction value at each time point, perform weighted averaging on the prediction value at each time point, and calculate the real-time prediction value of esophageal pressure. Embodiment 3:

[0148] Corresponding to the above method embodiment, this embodiment also provides an esophageal manometry device based on big data. The esophageal manometry device based on big data described below and the method for determining the creep state of a rock mass described above can refer to each other.

[0149] Figure 3 is a block diagram of an esophageal manometry device 800 based on big data according to an exemplary embodiment. Figure 3 As shown, the esophageal manometry device 800 based on big data may include: a processor 801 and a memory 802. The esophageal manometry device 800 based on big data may also include one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0150] The processor 801 is used to control the overall operation of the esophageal manometry device 800 based on big data to complete all or part of the steps in the above-mentioned method for determining the creep state of the rock mass. The memory 802 is used to store various types of data to support the operation of the esophageal manometry device 800 based on big data, and these data may include, for example, instructions for any application or method operated on the esophageal manometry device 800 based on big data, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be keyboards, mice, buttons, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the big data-based esophageal manometry device 800 and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: Wi-Fi module, Bluetooth module, NFC module.

[0151] In an exemplary embodiment, the esophageal manometry device 800 based on big data can be implemented by one or more application specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processor, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components to perform the above-mentioned esophageal manometry method based on big data.

[0152] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the rock mass creep state determination method described above are implemented. For example, the computer-readable storage medium may be the memory 802 including the program instructions described above, and the program instructions may be executed by the processor 801 of the esophageal manometry device 800 based on big data to complete the rock mass creep state determination method described above.

[0153] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for constructing an esophageal pressure prediction model based on big data, characterized in that: include: Acquiring historical test data, wherein the historical test data includes an endoscopic test video of at least one patient and a corresponding esophageal pressure value; Performing three-dimensional reconstruction processing according to the historical detection data, extracting depth information from the video image, and performing frame alignment in combination with feature point registration to obtain a three-dimensional model set; Performing feature extraction according to the three-dimensional model set, using a convolutional neural network to extract features from each three-dimensional model, extracting geometric information of the cardia, esophageal wall, and lower esophageal sphincter and converting them into structural feature vectors, and integrating to obtain a structural feature set; Performing simulation processing according to the structural feature set, using finite element analysis and Navier-Stokes equation to simulate esophageal pressure distribution, calculating the effect of the opening diameter change of the cardia area on the esophageal pressure and evaluating the pressure distribution in the stenotic area, and obtaining an esophageal pressure annotated data set; A model is constructed based on the esophageal pressure annotated data set and the structural feature set, and an esophageal pressure prediction model is obtained by analyzing the mapping relationship between esophageal structural features and local pressure and training a neural network model.

2. The method for constructing an esophageal pressure prediction model based on big data according to claim 1, characterized in that: A three-dimensional reconstruction process is performed according to the historical detection data, by extracting depth information from the video image and performing frame alignment in combination with feature point registration to obtain a three-dimensional model set, including: According to the image sequence of the endoscope video in the historical detection data, using structured light analysis to extract depth information in the video image, generating a fringe map by performing difference calculation on the grayscale images of adjacent frames, and obtaining a depth map sequence; Performing feature point detection processing according to the depth map sequence, performing SIFT feature point detection on the cardia, esophageal wall and lower esophageal sphincter in the esophageal detection area and generating descriptors to obtain a feature point set; Performing feature point matching processing according to the feature point set, aligning feature points of adjacent frames through FLANN matching and rigid registration based on the least squares method, and obtaining aligned feature point pairs; A three-dimensional reconstruction process is performed based on the aligned feature point pairs, and a three-dimensional structure of the detection area is generated through Delaunay triangulation and interpolation, and a three-dimensional model set is obtained by integration.

3. The method for constructing an esophageal pressure prediction model based on big data according to claim 1, characterized in that: According to the three-dimensional model set, feature extraction is performed on each three-dimensional model using a convolutional neural network, and geometric information of the cardia, esophageal wall and mucosal layer is extracted and converted into a structural feature vector, and a structural feature set is obtained by integration, including: Performing structural segmentation processing according to the three-dimensional model set, determining the relative positions and boundary differences of the cardia, esophageal wall and lower esophageal sphincter by combining three-dimensional curvature analysis, and obtaining a structural segmentation map of the detection area; Performing geometric feature extraction and gridding processing according to the structural segmentation map, by converting the cardia, esophageal wall and lower esophageal sphincter into regular grids, extracting the dynamic range of the opening diameter in the cardia area, extracting the local gradient of the thickness in the esophageal wall area, and extracting the curvature and texture structure of the surface in the lower esophageal sphincter area, to obtain a geometric feature matrix; Performing deep feature extraction processing according to the geometric feature matrix, and obtaining a structural feature vector by using a convolutional neural network to learn the contraction and expansion patterns of the cardia and lower esophageal sphincter, the thickness gradient and elasticity distribution of the esophageal wall; Dimensionality reduction processing is performed on the structural feature vector, and features that are sensitive to pressure changes are screened to obtain a set of structural features.

4. The method for constructing an esophageal pressure prediction model based on big data according to claim 1, characterized in that: According to the structural feature set, simulation processing is performed, and the esophageal pressure distribution is simulated using the finite element analysis method and the Navier-Stokes equation. By calculating the effect of the opening diameter change of the cardia area on the esophageal pressure and evaluating the pressure distribution in the stenotic area, an esophageal pressure annotation data set is obtained, including: Initialization processing is performed according to the structural feature set, the boundary conditions of the pressure simulation are set by using the discrete boundary method, the pressure and velocity boundaries of the cardia area are set by introducing a dynamic opening and closing control function according to the change characteristics of the cardia opening diameter, the elastic coefficient of the esophageal wall is set according to the thickness characteristics of the esophageal wall, and the gradient change of the pressure boundary value is defined in the lower esophageal sphincter to simulate the airflow resistance, so as to obtain the simulation initial condition matrix; The model is constructed according to the simulation initial condition matrix, and the dynamic influence of airflow and cardia opening and closing on the esophageal peri-esophageal pressure is simulated by the lattice Boltzmann method to obtain an esophageal pressure simulation model; Performing pressure field solution processing according to the esophageal pressure simulation model, decomposing the pressure field by applying fast Fourier transform, extracting high-frequency pressure changes and low-frequency trend components in the cardia area, and generating pressure field distribution data; Feature annotation processing is performed according to the pressure field distribution data, and the pressure change trend of the cardia and the lower esophageal sphincter is analyzed by wavelet decomposition, and the dynamic change characteristics are extracted and converted into an esophageal pressure annotation data set.

5. The method for constructing an esophageal pressure prediction model based on big data according to claim 4, characterized in that: The model is constructed according to the simulation initial condition matrix, and the dynamic influence of airflow and cardia opening and closing on the esophageal peri-esophageal pressure is simulated by the lattice Boltzmann method to obtain an esophageal pressure simulation model, including: Performing grid initialization processing according to the simulation initial condition matrix, obtaining a discrete lattice Boltzmann simulation grid by spatially gridding the cardia, esophageal wall and lower esophageal sphincter region and setting boundary nodes; Performing velocity distribution function initialization processing according to the lattice Boltzmann simulation grid, and obtaining an initial velocity distribution matrix by setting an initial airflow velocity and density distribution function to reflect the initial esophageal pressure state of the cardia and esophageal wall; Performing airflow evolution processing according to the initial velocity distribution matrix, simulating the dynamic evolution of airflow in the cardia and lower esophageal sphincter region through the fluid collision and migration process in the lattice Boltzmann method, and obtaining velocity field-pressure field distribution data; The velocity field-pressure field distribution data is analyzed and processed, and the esophageal pressure change pattern is obtained by analyzing the high-frequency and low-frequency components in the pressure distribution according to the dynamic response characteristics of the cardiac contraction, esophageal wall elasticity and lower esophageal sphincter, thereby constructing an esophageal pressure simulation model.

6. The method for constructing an esophageal pressure prediction model based on big data according to claim 1, characterized in that: The model building process is performed according to the esophageal pressure annotated data set and the structural feature set, and the esophageal pressure prediction model is obtained by analyzing the mapping relationship between the esophageal structural features and the local pressure and training the neural network model, including: Performing data alignment processing on the esophageal pressure annotated data set and the structural feature set, aligning the pressure data and the structural features on the time axis and space by time synchronization and space interpolation methods, and obtaining a spatiotemporal feature matrix; According to the spatiotemporal feature evidence, feature extraction processing is performed, and a one-dimensional convolutional neural network is used to perform convolution operations on multiple time windows to respectively extract the cardia opening and closing frequency, the contraction pattern of the lower esophageal sphincter, and the effects of the esophageal wall elastic gradient and thickness changes on the esophageal pressure, and perform vector stacking to obtain a multi-scale feature vector; A nonlinear mapping construction process is performed according to the multi-scale feature vector, and the contraction frequency of the cardia, the expansion characteristics of the lower esophageal sphincter, and the nonlinear effect of the elasticity change of the esophageal wall on the local pressure are modeled by a bidirectional long short-term memory network to obtain a preliminary structure-pressure mapping model; The preliminary structure-pressure mapping model is optimized and trained and tuned through a multi-layer deep neural network to obtain a final esophageal pressure prediction model.

7. A system for building an esophageal pressure prediction model based on big data, characterized in that: include: An acquisition module, used to acquire historical detection data, wherein the historical detection data includes an endoscopic detection video of at least one patient and a corresponding esophageal pressure value; A reconstruction module is used to perform three-dimensional reconstruction processing according to the historical detection data, extract depth information from the video image, and perform frame alignment in combination with feature point registration to obtain a three-dimensional model set; An extraction module, used to perform feature extraction based on the three-dimensional model set, perform feature extraction on each three-dimensional model using a convolutional neural network, extract geometric information of the cardia, esophageal wall and lower esophageal sphincter and convert it into a structural feature vector, and integrate to obtain a structural feature set; A simulation module, used to perform simulation processing according to the structural feature set, use finite element analysis and Navier-Stokes equation to simulate the esophageal pressure distribution, calculate the influence of the opening diameter change of the cardia area on the esophageal pressure and evaluate the distribution of pressure in the stenosis area, so as to obtain an esophageal pressure annotated data set; A construction module is used to perform model construction processing according to the esophageal pressure annotation data set and the structural feature set, and obtain an esophageal pressure prediction model by analyzing the mapping relationship between esophageal structural features and local pressure and training a neural network model.

8. The esophageal pressure prediction model construction system based on big data according to claim 7, characterized in that: The reconstruction module comprises: A first computing unit is used to extract depth information in the video image using structured light analysis based on the image sequence of the endoscope video in the historical detection data, generate a fringe map by performing difference calculation on the grayscale images of adjacent frames, and obtain a depth map sequence; A first detection unit is used to perform feature point detection processing according to the depth map sequence, and obtain a feature point set by performing SIFT feature point detection and generating descriptors on the cardia, esophageal wall and lower esophageal sphincter in the esophageal detection area; A first matching unit is used to perform feature point matching processing according to the feature point set, align feature points of adjacent frames through FLANN matching and rigid registration based on the least squares method, and obtain aligned feature point pairs; The first reconstruction unit is used to perform three-dimensional reconstruction processing according to the aligned feature point pairs, generate a three-dimensional structure of the detection area through Delaunay triangulation and interpolation, and integrate to obtain a three-dimensional model set.

9. A device for constructing an esophageal pressure prediction model based on big data, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the method for constructing an esophageal pressure prediction model based on big data as described in any one of claims 1 to 6 when executing the computer program.

Images