Thoracic cavity whole organ system level lung four-dimensional EIT detection digital motif and library building method

Abstract

The invention belongs to the technical field of medical imaging and digital twinning, and discloses a thoracic cavity whole organ system level lung four-dimensional EIT detection digital motif and a library building method, and the method comprises the following steps: S1) segmenting multiple organs such as double lungs, heart, sternum, rib and thoracic vertebra central axis skeleton, main bronchus and esophagus based on an individual CT image, and building a three-dimensional geometric model integrated with an EIT electrode array; s2) simulating breathing dynamic deformation by using a solid mechanical model; s3) establishing a coupling relationship between deformation and conductivity, and generating dynamic three-dimensional conductivity distribution; s4) solving an electromagnetic field positive problem, and generating a virtual EIT voltage database synchronous with respiration; the invention also covers pathological model integration and data enhancement strategies. According to the method, a high-fidelity, large-scale and diversified EIT virtual data set can be generated, core data support is provided for training and verifying an AI intelligent reconstruction algorithm, and a model foundation is laid for developing a lung function dynamic digital twinning system.

Description

Technical Field

[0001] The present invention relates to the fields of medical imaging, digital twins, and biomedical engineering technology. Specifically, it relates to a method for constructing a digital phantom for electrical impedance tomography (EIT) of the lung based on an individualized whole-organ system-level model of the thoracic cavity, and a method for generating a high-fidelity virtual EIT sensing database using the phantom. Background Art

[0002] Electrical impedance tomography (EIT), as an emerging, non-invasive, radiation-free, and real-time monitoring imaging technology, has shown great potential in clinical lung function assessment. EIT technology arranges multi-channel electrodes on the body surface, injects weak alternating current into the body, and measures the surface boundary voltage, thereby reconstructing the body's conductivity distribution and achieving dynamic visualization of lung ventilation and perfusion status.

[0003] However, traditional EIT technology has several drawbacks in clinical application: 1) Dimensional limitations: Traditional EIT mainly relies on two-dimensional cross-sections for reconstruction, which makes it difficult to capture the three-dimensional dynamic ventilation characteristics of the entire lung and has limited information.

[0004] 2) Lack of individualization: General reconstruction algorithms often ignore the anatomical differences of thoracic organs between individuals, such as the complex effects of cardiac compression and diaphragmatic movement on lung deformation, resulting in a mismatch between the reconstructed model and the patient's actual physiological state.

[0005] 3) Data scarcity: Obtaining large-scale, diverse, and standardized EIT measurement data in clinical practice is constrained by ethical restrictions, operational risks, and individual variability. This makes it difficult to meet the data-driven intelligent reconstruction algorithm's demand for a large number of training samples, limiting the algorithm's robustness and generalization capabilities.

[0006] With the rise of the digital twin concept, the deep integration of virtual simulation and real-world monitoring has become a hot topic in multidisciplinary research. In the medical field, digital twins can integrate personalized anatomical models with multimodal data such as CT, MRI, and EIT to create a virtual replica that reflects the patient's real-time physiological state, providing a new approach to precision medicine and personalized treatment.

[0007] However, in current lung EIT applications, there is still a lack of a complete digital twin framework that can cover the entire chest cavity, encompass multi-organ linkage, couple multi-physics field characteristics, and be linked with real-time data; existing research mostly focuses on the modeling of a single organ, while system-level digital twins for EIT need to address three major challenges: 1) the complexity of multi-organ linkage modeling; 2) cross-scale electrophysiological property mapping of tissue dielectric parameters that dynamically change with respiratory phase; 3) closed-loop calibration between virtual simulation data and real measurement data.

[0008] Therefore, there is an urgent need for a digital twin phantom that can combine real three-dimensional chest structure modeling, lung motion simulation, dynamic mapping of electrical parameters, and EIT measurement data feedback fusion to achieve high-fidelity visualization, predictable simulation, and personalized analysis of lung status, providing a solid foundation for the development and verification of a new generation of EIT intelligent algorithms. Summary of the Invention

[0009] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and library construction method, aiming to solve the challenges of three-dimensional imaging, individualized modeling and data scarcity in EIT technology.

[0010] The core technical solution of this invention is to construct a digital thoracic phantom for four-dimensional EIT detection of the lungs, and use this phantom to achieve the following core functions: Generation of high-quality virtual databases: Through multi-organ coupled simulation, a standardized EIT data pool covering different anatomical variations, pathological conditions, and respiratory patterns is generated, providing a sufficient and high-quality training and verification foundation for AI-driven intelligent reconstruction algorithms.

[0011] Digital twin system development support: Provides a sophisticated dynamic chest model framework to support rapid migration and personalized adaptation from virtual simulation to real clinical monitoring systems, providing a core model foundation for the development of dynamic digital twin systems.

[0012] To achieve the above object, the present invention provides the following technical solutions: A method for building a database of a thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom includes the following steps: Step S1, constructing an individualized 3D geometric model: Based on chest medical imaging data such as CT images obtained from an individual, multiple major organs within the chest cavity, such as the lungs, heart, axial skeleton, main bronchi, and esophagus, are segmented, and 3D geometric models of each organ are generated. These models are then integrated into a complete, individualized 3D chest geometry model; at the same time, an EIT electrode array model is integrated into a preset position on the body surface of the model.

[0013] Step S2, multi-physics field dynamic model construction: Assign electrical parameters from a public database to each organ and tissue in the chest model. Specifically, the electrical parameters are usually conductivity and dielectric constant; using a solid mechanics model, by applying dynamic loads and boundary conditions that simulate the respiratory process, the dynamic deformation process of the lungs and related organs during the respiratory cycle is simulated, thereby obtaining organ geometric morphological data that changes with time.

[0014] Step S3, multi-physics field coupling and dynamic conductivity mapping: establish a coupling relationship model between the dynamic deformation of the organ and its electrical characteristics; map the dynamic geometric data such as lung volume change obtained in step S2 into a spatial three-dimensional conductivity distribution model that changes dynamically over time.

[0015] Step S4, virtual EIT sensor data generation: Based on a dynamic three-dimensional conductivity distribution model and an EIT electrode array model, the electromagnetic field forward problem is solved using formulas such as the Laplace equation. The EIT current excitation and voltage measurement process is simulated, and the voltage sequence generated on the boundary electrodes is calculated. This voltage data is synchronized with the dynamic deformation process of the organ and together constitutes a virtual EIT sensor database.

[0016] Preferably, in step S1, the thoracic medical imaging data is CT imaging data; the segmented multiple organs include at least the lungs, the heart, the axial skeleton, the main bronchi and the esophagus.

[0017] Preferably, in step S1, when integrating the EIT electrode array model, the electrode-skin contact impedance characteristics are simulated, and the electrode array model is arranged at a height between the fourth rib and the fifth rib along the circumference of the chest model.

[0018] Preferably, in step S2, the solid mechanics model simulates the effect of respiratory internal pressure by applying a dynamic distributed force on the lung surface, applies a fixed constraint in the lung apex area, and applies a multi-axis dynamic force load in the lung base area to simulate the deformation driven by diaphragm movement.

[0019] Preferably, the method also includes a state model integration step: introducing at least one lung state model into the three-dimensional geometric model of the lung, the lung state model including a healthy model and a pathological model, and assigning specific geometric constraints and electrical parameters to the corresponding areas of the state model.

[0020] Preferably, when the lung state model is a pathological model, the pathological model is selected from at least one of a pneumonia model, a pneumothorax model, and a lung tumor model, wherein: When the pathological model is a pneumonia model, a conductivity and a dielectric constant higher than those of normal lung tissue are assigned to the delineated pneumonia area, and a fixed constraint is applied to limit its strain; When the pathological model is a pneumothorax model, an invisible air cavity is set in the lung apex area, and electrical parameters close to those of air are assigned to the air cavity; When the pathological model is a lung tumor model, the segmented tumor geometric model is imported into the normal lung model, and electrical parameters higher than those of normal lung tissue are assigned to it, and fixed constraints are applied to limit its strain.

[0021] Preferably, in step S3, the change in the electrical conductivity of the lung tissue with lung deformation satisfies the following relationship: in, for Time and space point The conductivity at is the conductivity at the end of inspiration, is the maximum lung volume, for Lung volume at the moment, is the change in conductivity.

[0022] Preferably, in step S4, the EIT electrode array model includes a single-layer 16-electrode system or a double-layer 32-electrode system; for the double-layer 32-electrode system, the virtual EIT sensing data includes upper-layer electrode independent measurement data, lower-layer electrode independent measurement data, and cross-layer joint measurement data.

[0023] Preferably, the method further comprises data enhancement and labeling steps: Simulating breathing patterns under different tidal volumes, respiratory rates, or inspiration-expiration ratios by adjusting respiratory parameters in the solid mechanics model to expand the virtual EIT sensor database; Injecting Gaussian white noise or contact impedance fluctuation noise into the generated virtual EIT sensing data; The generated boundary voltage measurements were time-stamped and synchronized with the corresponding respiratory phase and three-dimensional conductivity distribution to construct a triplet dataset containing respiratory phase, conductivity distribution, and boundary voltage.

[0024] A method for training an artificial intelligence model based on the thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and the database generated by the database construction method, characterized by comprising: The artificial intelligence model is trained using boundary voltage measurements in the virtual EIT sensor database as input and corresponding three-dimensional conductivity distributions as output to establish a nonlinear mapping relationship from EIT boundary voltages to internal three-dimensional conductivity distributions; During the training process, the anatomical structure parameters in the individualized thoracic three-dimensional geometric model are also incorporated into the artificial intelligence model as physical constraints.

[0025] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention overcomes the limitations of traditional EIT two-dimensional and universal models and constructs a patient-specific three-dimensional chest model that includes multi-organ linkage, significantly improving the physiological authenticity of the model.

[0026] 2. The present invention simulates lung deformation during breathing and its dynamic impact on conductivity distribution by coupling solid mechanics and electromagnetic fields, making the generated virtual data closer to the real physiological process.

[0027] 3. The present invention can systematically and cost-effectively generate EIT datasets covering normal conditions, various pathological conditions, and different respiratory modes, solving the problem of difficult clinical data collection and providing the possibility for training and verifying high-performance AI reconstruction algorithms.

[0028] 4. The digital model and database construction method constructed by the present invention provide core technology and model framework for the development of dynamic digital twin systems of lung function for clinical applications, and promote the development of EIT technology towards precision and intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for describing the embodiments.

[0030] Figure 1 Schematic diagram of a patient's chest CT image used for organ segmentation in the present invention; Figures 2 to 6 They are respectively the three-dimensional geometric models of the thoracic cavity, axial skeleton, lungs, heart, main bronchi and esophagus obtained after segmentation and processing from the CT images in the present invention; Figure 7 Schematic diagram of a single-layer 16-electrode EIT sensor array model in the present invention; Figure 8 Schematic diagram of the double-layer 32-electrode EIT sensor array model in the present invention; Figure 9 Schematic diagram of the geometric model of the thoracic digital phantom in which the organ model is integrated with the single-layer and double-layer EIT sensors respectively in the present invention; Figure 10 This is a virtual measurement voltage data diagram of a single-layer 16-electrode EIT system generated in the present invention; Figure 11 This is a diagram of virtual measured voltage data of a double-layer 32-electrode EIT system generated in the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] Example 1 This embodiment describes the digital phantom construction and database generation process. This section describes in detail the complete process of a thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method.

[0033] Step S1: Construction of an individualized thoracic whole-organ system-level geometric model First, the organ segmentation operation is performed. The purpose of this step is to extract the accurate three-dimensional structure of the organ from the original medical image, such as Figure 1 As shown, using a patient's chest CT image as input data, medical image processing software such as Mimics and 3D Slicer is used, and a semi-automatic or fully automatic segmentation algorithm is employed to accurately segment the major organs in the thoracic cavity, including but not limited to: the lungs, heart, the axial skeleton including the sternum, ribs, and thoracic vertebrae, the main bronchi, and the esophagus; denoising and smoothing processing are performed on the segmented three-dimensional models of each organ. Specifically, the denoising processing may be a median filtering operation, and the smoothing processing may be a Laplace smoothing operation. The purpose is to remove image noise and segmentation defects without affecting the key geometric features of the organs, thereby ensuring the correct topological structure and smooth surface of the model, and providing a high-quality geometric foundation for subsequent finite element meshing and physical simulation; Figures 2 to 6 The processed 3D geometric models of the thoracic cavity, axial skeleton, lungs, heart, and main bronchi and esophagus are displayed.

[0034] Secondly, the organ model and EIT sensor integration operation is performed. The multiple three-dimensional organ models processed in step S1 are imported into a multi-physics field simulation software platform, such as COMSOL Multiphysics. In the software, these independent organ models are precisely assembled to construct a complete, anatomically accurate, individualized thoracic whole-organ geometric model.

[0035] Then, the EIT sensor integration operation is performed, such as Figure 7 and Figure 8 As shown, the present invention can construct EIT electrode array models with different configurations; one embodiment is to construct an Figure 7 Another preferred embodiment is to construct a single-layer 16-electrode annular array as shown; Figure 8 The double-layer 32-electrode array shown in FIG. 1 is used to embed the selected electrode array model into the surface of the chest cavity model, as shown in FIG. Figure 9 As shown in the figure, the final form of the integration of the organ model with single-layer (left) and double-layer (right) electrode arrays is shown; the arrangement of the electrodes follows clinical practice, for example, it surrounds the patient's torso and is located at a height between the fourth and fifth ribs of the model; the final generated thoracic full-organ system-level digital phantom constitutes an accurate anatomical basis for all subsequent simulations.

[0036] Step S2: Construction of an individualized thoracic whole-organ system-level physical model First, the electrical parameter assignment operation is performed. Assigning accurate electrical parameters to each tissue in the model is the key to ensuring simulation accuracy. This embodiment queries and extracts the electrical parameters of each biological tissue at a specific frequency from a public biomedical database, such as the ITIS Database for Thermal and Electromagnetic Parameters of Biological Tissues. Specifically, the electrical parameters are conductivity and relative permittivity For example, at the 125kHz frequency commonly used by EIT, assign values ​​to different parts of the model. Example parameters are shown in the following table: Next, we simulate the dynamic lung deformation during breathing using the Solid Mechanics module in COMSOL Multiphysics. The specific settings are as follows: Load, a time-varying dynamic distributed force is applied to the inner surfaces of the segmented left and right lungs. This force is equivalent to the change in intrapulmonary pressure during breathing.

[0037] Boundary conditions: A three-dimensional fixed constraint is applied to the apex region to limit its displacement and simulate its connection with the top of the thorax, with strain controlled below 5%. At the same time, a multi-axial dynamic force load is applied to the lung base region along the anatomical coordinate system to simulate the pushing effect of the diaphragm on the lung base during exhalation. This setting is intended to achieve a principal strain of 10%-15% in the lung base region, which is consistent with the physiological expiratory retraction characteristics.

[0038] Constitutive model: Lung tissue is assumed to be an isotropic linear elastic material, and its stress-strain relationship satisfies Hooke's law, which is defined by the Lame constants λ and μ. By solving the linear elastic mechanics equation , the displacement field and deformation data of the lungs and surrounding organs at a series of equally spaced time nodes during the respiratory cycle can be obtained. These data accurately capture the main deformation process of the lungs from maximum volume to minimum volume.

[0039] Subsequently, a disease model integration operation is performed. To enhance the diversity of the database, this embodiment also integrates a variety of typical lung disease models.

[0040] Pneumonia model: In the lung model, the inflammation area is delineated, such as partial inflammation of the right upper lobe or complete inflammation of the right lung; the electrical parameters different from those of normal lung tissue are assigned to this area, such as the conductivity is set to 0.704S / m and the dielectric constant is set to In the mechanical simulation, a three-way fixed constraint was imposed on the inflamed area, limiting its strain to less than 5% to simulate the characteristic of decreased compliance of lung tissue caused by inflammation.

[0041] Pneumothorax model: A non-deformable air cavity occupying 5%-30% of the lung volume is created at the apex of the lung. This cavity does not mechanically participate in deformation and is assigned electrical parameters close to those of air, specifically a conductivity of approximately 0.001 S / m and a dielectric constant of approximately 1.

[0042] Tumor model: The pre-segmented tumor geometry model was imported into the normal lung model, and the electrical parameters of lung cancer tissue were assigned to it, for example, the conductivity was 0.3597 S / m, which is about 1.6-3.3 times that of normal lung; the dielectric constant was 1.11×10 4 , which is about 3-5 times that of normal lungs; in the mechanical simulation, fixed constraints are also imposed on the tumor tissue to limit its strain to less than 5%.

[0043] Finally, a multi-physics coupling simulation of the respiratory process is performed. This step is one of the core of the present invention. It couples mechanical deformation with electrical properties. At the end of inspiration, the alveoli are fully expanded, the air content is high, and the conductivity is low. With expiration, the lung volume decreases, the air content decreases, and the conductivity increases accordingly. This dynamic change is modeled by the following formula: in, yes The conductivity distribution at time is the baseline conductivity distribution at the end of inspiration, It is obtained through mechanical simulation Lung volume at a moment, is the maximum lung volume, also known as the end-inspiratory volume. is the total conductivity change during the entire respiratory cycle.

[0044] Through this coupling model, the geometric deformation data of each time node obtained from the mechanical simulation is converted into a dynamically changing, full-field three-dimensional conductivity distribution map.

[0045] Step S3: Method for building a dynamic virtual lung EIT monitoring database First, we generate high-fidelity EIT sensor output voltage data. Based on the dynamic digital phantom established above, we simulate the EIT data acquisition process. In the AC / DC module of COMSOL, we solve the current conservation equation, namely the Laplace equation. ,in is the time-varying conductivity distribution obtained in the previous step.

[0046] The simulation adopts the adjacent excitation-measurement mode. For a single-layer 16-electrode EIT system, the adjacent electrode pairs are used as excitation electrodes to inject current, while the voltage is measured on the remaining unexcited electrode pairs. A complete rotation measurement can obtain 16×13 / 2=104 sets of independent voltage observation values, such as Figure 10 As shown in FIG, the figure shows the virtual measured voltage data of a single-layer 16-electrode EIT system generated at a certain moment.

[0047] For a double-layer 32-electrode EIT system, the measurement mode is more complex and can include upper-layer independent measurement, lower-layer independent measurement, and cross-layer joint measurement. For example, by designing a specific inter-layer excitation-measurement protocol, a single measurement can form 328 sets of comprehensive observation data sets, such as Figure 11 As shown, the figure shows the virtual measured voltage data generated by the two-layer system.

[0048] Specifically, the comprehensive observation data set includes 104 sets of data independently measured at the upper layer, 104 sets of data independently measured at the lower layer, and 120 sets of data jointly measured across layers.

[0049] By repeating this process for each time node in the respiratory cycle, EIT voltage time series data that is completely synchronized with the dynamic changes of the lungs can be generated.

[0050] First, a multi-dimensional data enhancement strategy is implemented. To improve the coverage and generalization ability of the database, the following enhancement strategies are adopted: Breathing pattern control: Different breathing parameter combinations are set in the mechanical model, such as tidal volume 300-800mL, respiratory rate 12-30 breaths / min, and inspiration-expiration ratio 1:1 to 1:3, to simulate various normal / abnormal breathing patterns such as calm breathing, deep breathing, and rapid breathing.

[0051] Pathological model injection: Various pathological models described in the third step of step S2 are combined with different breathing patterns to generate a large amount of EIT data under pathological conditions.

[0052] Secondly, virtual data is standardized and labeled. To make the virtual data closer to clinical practice and easier to use, the following processing is performed: Noise injection: Based on the signal-to-noise ratio (SNR) characteristics of clinical EIT equipment, which is typically 30-80dB, Gaussian white noise is added to the generated clean voltage data. At the same time, poor electrode contact can be simulated to introduce contact impedance fluctuation noise.

[0053] Spatiotemporal alignment and labeling: The virtual voltage measurement data at each time point is strictly time-stamped and aligned with its corresponding simulation moment, respiratory phase label, pathology type label, and the three-dimensional conductivity distribution map as the gold standard.

[0054] Data format: All data including geometric, physical, and EIT measurement data are organized and stored in HDF5 format files. This format supports a hierarchical structure for easy management and call. A typical file structure is as follows: / SubjectID ├──Geometry (CT grid coordinates, electrode position matrix) ├──Physics (conductivity distribution tensor, dielectric constant mapping table) ├──EIT_Data (voltage measurement timing data, excitation mode parameters) └──Metadata (respiratory phase label, pathology type label, mechanical parameter summary) Example 2 This embodiment specifically describes the application of a database in the training of an AI reconstruction algorithm. This section describes in detail how to use the database generated in Example 1 to train a data-driven intelligent reconstruction algorithm.

[0055] The goal of this application is to build a deep neural network AI model to achieve direct and rapid reconstruction of the internal three-dimensional conductivity change distribution from EIT boundary voltage measurements.

[0056] First, a virtual database was constructed. Using the method of Example 1, different respiratory states, such as calm breathing and deep breathing, and pathological states such as pneumonia of varying locations and severity, were batch simulated on the digital twin platform. This generated a database containing tens of thousands of sets of virtual EIT data, each of which represented a data pair: a boundary voltage vector and a three-dimensional conductivity distribution map.

[0057] Secondly, AI model training is carried out using a deep neural network architecture, such as U-Net, Transformer, or a custom encoder-decoder network; the boundary voltage vector in the database is used as the input of the network, and the corresponding three-dimensional conductivity distribution map is used as the output of the network, that is, the label or true value; during the training process, the individualized anatomical structure parameters in the digital twin are incorporated into the network's loss function as physical constraints or prior information. Specifically, the position can be the chest contour or organ position; this guides the mapping relationship learned by the network to not only fit the data, but also must conform to the individual's anatomical laws; through backpropagation and optimization algorithms, the optimization algorithm can specifically use Adam to minimize the difference between the conductivity distribution predicted by the network and the true value distribution in the database until the model converges.

[0058] Virtual / real real-time monitoring is then carried out. After training is completed, the AI ​​model has the ability to reconstruct from voltage to image. In virtual monitoring, the newly generated voltage data from the digital twin can be input into the model to directly output a visual distribution of the three-dimensional conductivity changes in the lungs during dynamic breathing.

[0059] In actual applications, the EIT voltage data collected from real patients, together with the digital twin of the patient that provides parameters such as electrode configuration and tissue properties, are input into the trained AI model. The model can then directly output dynamic images of the patient's three-dimensional lung conductivity changes in real time, realizing efficient and intelligent bedside monitoring.

[0060] In summary, this invention constructs a highly realistic, dynamically evolving, personalized digital chest phantom and generates a large-scale, high-quality virtual EIT database based on it. This not only provides an innovative technical path to address the bottlenecks of traditional EIT technology, but also lays a solid foundation for the development of the next generation of AI-based EIT systems that truly realize three-dimensional, personalized, and real-time monitoring.

[0061] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A digital phantom and database construction method for 4D EIT detection of the lungs at the whole-organ system level in the thoracic cavity, characterized by: The following steps are involved: Step S1, constructing an individualized 3D geometric model: Based on chest medical imaging data acquired from an individual, multiple organs within the chest cavity are segmented, a 3D geometric model of each organ is generated, and the 3D geometric models are integrated into an individualized 3D chest cavity geometric model containing the multiple organs; and an EIT electrode array model is integrated at a preset position on the body surface of the 3D chest cavity geometric model; Step S2, multi-physics field dynamic model construction: assigning preset electrical parameters to each organ and tissue in the three-dimensional thoracic cavity geometric model; and using a solid mechanics model to simulate the dynamic deformation process of the lungs and related organs caused by respiratory movement during the respiratory cycle, to obtain the geometric morphological data of the organs that change over time; Step S3, multi-physics field coupling and dynamic conductivity mapping: establishing a coupling relationship between the dynamic deformation of the organ and the electrical parameters, and mapping the time-varying organ geometry data obtained in step S2 into a time-varying, spatially distributed three-dimensional conductivity distribution model; Step S4, generating virtual EIT sensor data: Based on the time-varying three-dimensional conductivity distribution model and the EIT electrode array model, the EIT measurement process is simulated by solving an electromagnetic field forward problem, and the boundary voltage measurement values ​​generated on the EIT electrode array under a preset excitation mode are calculated, thereby generating a virtual EIT sensor database synchronized with the dynamic deformation process.

2. The thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to claim 1 is characterized in that: In step S1, the thoracic medical image data is CT image data; the segmented multiple organs include at least the lungs, the heart, the axial skeleton, the main bronchi and the esophagus.

3. The thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to claim 1, characterized in that: In step S1 , when integrating the EIT electrode array model, the electrode-skin contact impedance characteristic is simulated, and the electrode array model is arranged at a height between the fourth rib and the fifth rib along the circumference of the chest model.

4. The thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to claim 2, characterized in that: In step S2, the solid mechanics model simulates the effect of respiratory internal pressure by applying a dynamic distributed force on the lung surface, applies a fixed constraint in the lung apex area, and applies a multi-axis dynamic force load in the lung base area to simulate the deformation driven by diaphragm movement.

5. The thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to claim 4, characterized in that: The method also includes a state model integration step: introducing at least one lung state model into the three-dimensional geometric model of the lung, the lung state model including a healthy model and a pathological model, and assigning specific geometric constraints and electrical parameters to the corresponding areas of the state model.

6. The thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to claim 5, characterized in that: When the lung state model is a pathological model, the pathological model is selected from at least one of a pneumonia model, a pneumothorax model, and a lung tumor model, wherein: When the pathological model is a pneumonia model, a conductivity and a dielectric constant higher than those of normal lung tissue are assigned to the delineated pneumonia area, and a fixed constraint is applied to limit its strain; When the pathological model is a pneumothorax model, an invisible air cavity is set in the lung apex area, and electrical parameters close to those of air are assigned to the air cavity; When the pathological model is a lung tumor model, the segmented tumor geometric model is imported into the normal lung model, and electrical parameters higher than those of normal lung tissue are assigned to it, and fixed constraints are applied to limit its strain.

7. The thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to claim 3, characterized in that: In step S3, the change in the electrical conductivity of the lung tissue with lung deformation satisfies the following relationship: in, for Time and space point The conductivity at is the conductivity at the end of inspiration, is the maximum lung volume, for Lung volume at the moment, is the change in conductivity.

8. The thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to claim 1, characterized in that: In step S4, the EIT electrode array model includes a single-layer 16-electrode system or a double-layer 32-electrode system; for the double-layer 32-electrode system, the virtual EIT sensing data includes upper-layer electrode independent measurement data, lower-layer electrode independent measurement data, and cross-layer joint measurement data.

9. The thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to claim 1, characterized in that: The method also includes data enhancement and labeling steps: Simulating breathing patterns under different tidal volumes, respiratory rates, or inspiration-expiration ratios by adjusting respiratory parameters in the solid mechanics model to expand the virtual EIT sensor database; Injecting Gaussian white noise or contact impedance fluctuation noise into the generated virtual EIT sensing data; The generated boundary voltage measurements were time-stamped and synchronized with the corresponding respiratory phase and three-dimensional conductivity distribution to construct a triplet dataset containing respiratory phase, conductivity distribution, and boundary voltage.

10. A method for training an artificial intelligence model based on the database generated by the thoracic whole-organ system-level lung four-dimensional EIT detection digital phantom and database construction method according to any one of claims 1 to 9, characterized in that: include: The artificial intelligence model is trained using boundary voltage measurements in the virtual EIT sensor database as input and corresponding three-dimensional conductivity distributions as output to establish a nonlinear mapping relationship from EIT boundary voltages to internal three-dimensional conductivity distributions; During the training process, the anatomical structure parameters in the individualized thoracic three-dimensional geometric model are also incorporated into the artificial intelligence model as physical constraints.

Images