Lung hypertension diagnosis and treatment simulation system based on digital twinning

Abstract

The invention discloses a pulmonary hypertension diagnosis and treatment simulation system based on digital twinning, and the system comprises the following modules: a medical examination data processing module which is used for obtaining multi-modal medical data; the cardiopulmonary structure reconstruction module is used for obtaining a three-dimensional time-varying cardiopulmonary structure parameter set; the three-branch feature and parameter generation module is used for generating hemodynamic parameter vectors; the blood flow simulation and energy fingerprint module is used for generating a pulmonary circulation energy fingerprint vector; the far-end impedance sensitivity and updating module is used for generating a new pulmonary circulation energy fingerprint vector; and the energy fingerprint comparison and result output module is used for comparing the new pulmonary circulation energy fingerprint vector with the original pulmonary circulation energy fingerprint vector and determining whether far-end impedance update is accepted or not according to a preset performance improvement condition. According to the invention, non-invasive accurate diagnosis, disease progress evaluation and treatment effect prediction of pulmonary hypertension are realized by constructing a mechanism based on multi-modal medical data and energy-impedance co-evolution.

Description

Technical Field

[0001] This invention relates to the field of medical engineering technology, and in particular to a pulmonary hypertension diagnosis and treatment simulation system based on digital twins. Background Technology

[0002] Current clinical assessments of pulmonary hypertension primarily rely on echocardiography, CTA / MRA imaging, and right heart catheterization for pressure measurement. While echocardiography is widely used due to its non-invasiveness and convenience, its results are highly dependent on acoustic window conditions and operator experience, providing only a rough estimate of pulmonary artery pressure and failing to reflect the true state of pulmonary circulation from a holistic hemodynamic perspective. Although CTA / MRA provides information on vascular and cardiac structure, it is static data and cannot reflect the hemodynamic processes that change with the cardiac cycle. Right heart catheterization, while directly measuring pressure and resistance, is invasive and risky, and cannot be performed frequently for continuous monitoring of disease progression. Therefore, current techniques generally rely on single-timepoint, single-modality, or empirical estimations to diagnose pulmonary hypertension, making it difficult to comprehensively assess pulmonary circulatory function, particularly lacking the ability to quantify microcirculatory dysfunction and trends in distal impedance changes.

[0003] In recent years, hemodynamic simulation methods based on medical imaging have attempted to assist in the assessment of pulmonary hypertension through numerical simulation. However, most techniques drive the model with fixed or one-time fitted parameters, focusing only on macroscopic pressure and velocity fields. They fail to reflect deeper hemodynamic characteristics such as pulmonary circulation energy loss, power spectrum anomalies, and vortex changes, and cannot assess distal impedance changes based on energy characteristics. More importantly, existing simulation technologies lack the ability to dynamically update with disease progression. Once the model is established, it lacks an adaptive adjustment mechanism, making it difficult to predict future disease progression and treatment response, thus limiting its ability in clinical decision support. While existing machine learning prediction methods can estimate hemodynamic values, they only focus on static indicators and lack the ability to trace segmental-level blood flow physiological changes, failing to meet the needs of continuous tracking and prognostic prediction of pulmonary hypertension.

[0004] Existing technologies face three key limitations in the diagnosis and treatment of pulmonary hypertension: they cannot non-invasively, dynamically, and accurately assess the hemodynamic status of pulmonary circulation; they cannot identify changes in microcirculatory impedance and the extent of segmental lesions from an energy perspective; and they cannot form a diagnostic and therapeutic feedback mechanism that updates with the course of the disease to reliably predict the effectiveness of treatment plans.

[0005] Therefore, how to provide a pulmonary hypertension diagnosis and treatment simulation system based on digital twins is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a digital twin-based simulation system for the diagnosis and treatment of pulmonary hypertension. This invention fully utilizes multimodal medical data fusion, a three-branch feature inversion network, and an energy fingerprint-driven hemodynamic self-updating mechanism. By constructing a digital twin pulmonary circulation that dynamically evolves with the disease progression, it enables the diagnosis of pulmonary hypertension, assessment of disease progression, and prediction of treatment plans. This invention reconstructs the patient's three-dimensional time-varying cardiopulmonary structure non-invasively, inverts key hemodynamic parameters, and utilizes deep hemodynamic features such as energy loss, power spectrum, and eddy current to form a pulmonary circulation energy fingerprint. Combined with distal impedance sensitivity adjustment, the digital twin model automatically corrects itself and continuously approximates the patient's true physiological state. This invention possesses advantages such as high diagnostic accuracy, strong ability to locate microcirculatory lesions, reliable treatment response prediction, and reduced reliance on right heart catheterization, providing a safe, efficient, and sustainable assessment solution for the intelligent and precise diagnosis and treatment of pulmonary hypertension.

[0007] A pulmonary hypertension diagnosis and treatment simulation system based on digital twins according to an embodiment of the present invention includes the following modules:

[0008] The medical examination data processing module is used to acquire and preprocess comprehensive medical examination data of patients to obtain multimodal medical data;

[0009] The cardiopulmonary structure reconstruction module is used for automatic reconstruction based on multimodal medical data to obtain a three-dimensional time-varying cardiopulmonary structure parameter set;

[0010] The three-branch feature and parameter generation module is used to generate hemodynamic parameter vectors from a three-dimensional time-varying cardiopulmonary structural parameter set and multimodal medical data.

[0011] The blood flow simulation and energy fingerprint module is used to perform blood flow simulation based on hemodynamic parameter vectors, obtain pulmonary circulation pressure and velocity distribution, and generate pulmonary circulation energy fingerprint vectors.

[0012] The distal impedance sensitivity and update module is used to determine the direction and magnitude of distal impedance adjustment based on the pulmonary circulation energy fingerprint vector and the hemodynamic parameter vector, and to generate a new pulmonary circulation energy fingerprint vector.

[0013] The energy fingerprint comparison and result output module is used to compare the new pulmonary circulation energy fingerprint vector with the original pulmonary circulation energy fingerprint vector, and determine whether to accept the remote impedance update based on the preset performance improvement conditions.

[0014] Optionally, modules can be integrated using the following methods:

[0015] Acquire comprehensive medical examination data from patients, preprocess the comprehensive medical examination data, and form multimodal medical data;

[0016] Automatic structural reconstruction of the right atrium, right ventricle, pulmonary artery trunk and multi-level branches based on MRA image data in multimodal medical data is performed to generate a three-dimensional time-varying cardiopulmonary structural parameter set containing information on the changes of vessel diameter, wall thickness, pulsation amplitude and ventricular volume over time.

[0017] Using a three-dimensional time-varying cardiopulmonary structural parameter set and multimodal medical data as input, a three-branch network consisting of morphological feature branches, dynamic feature branches, and clinical feature branches is constructed to output a hemodynamic parameter vector;

[0018] Based on hemodynamic parameter vectors, simulation calculations are performed to obtain pulmonary circulation pressure and velocity distributions. Energy loss analysis, power spectrum analysis, and vorticity analysis are then performed on the pressure and velocity distributions to construct a pulmonary circulation energy fingerprint vector.

[0019] Based on the sensitivity relationship between the pulmonary circulation energy fingerprint vector and the distal impedance in the hemodynamic parameter vector, the adjustment direction and adjustment range of the distal impedance are determined, the hemodynamic parameter vector is updated, and the hemodynamic simulation is executed again to generate a new pulmonary circulation energy fingerprint vector.

[0020] The new pulmonary circulation energy fingerprint vector is compared with the pulmonary circulation energy fingerprint vector. When the new pulmonary circulation energy fingerprint vector meets the preset performance improvement conditions, the distal impedance update is accepted and the pulmonary hypertension diagnosis result, disease progression assessment result or treatment plan prediction result is output. Otherwise, the distal impedance update is rejected and the corresponding result is output.

[0021] Optionally, the comprehensive medical examination data includes MRA imaging data, echocardiogram data, electrocardiogram signals, and laboratory test data.

[0022] Optionally, the preprocessing of the comprehensive medical examination data includes time-phase registration of MRA image data, noise suppression of echocardiogram data, and signal normalization of electrocardiogram signals and laboratory test data.

[0023] Optionally, generating a three-dimensional time-varying cardiopulmonary structural parameter set containing information on the changes in vessel diameter, wall thickness, pulsation amplitude, and ventricular volume over time includes:

[0024] MRA image data was selected from multimodal medical data, and the MRA image data was read and the region of interest was cropped to obtain local cardiopulmonary MRA image data containing only the heart and pulmonary artery regions.

[0025] The cardiopulmonary local MRA image data is input into a pre-trained segmentation network, and voxels at each time phase are classified to obtain initial segmentation results that distinguish the right atrium, right ventricle, main pulmonary artery and multi-level branches.

[0026] Based on the initial segmentation results, temporal consistency analysis was performed on the boundary positions and volume changes of the right atrium, right ventricle, pulmonary artery trunk and multi-level branches under all time phases. The pre-set pulmonary artery tree anatomical topology constraints were used to automatically correct and merge discontinuous and abnormal branches, forming a temporal segmentation result of the right atrium, right ventricle, pulmonary artery trunk and multi-level branches with a consistent topological structure throughout the entire cardiac cycle.

[0027] In the time-series segmentation results, the pulmonary artery trunk and its branches at all levels are divided into several segments along the direction of the blood vessels. The changes in the inner diameter of the blood vessels, the thickness of the blood vessel walls, and the amplitude of the inner diameter over time are calculated at multiple time points in each segment to obtain cardiopulmonary structural parameters characterizing the changes in the diameter, wall thickness, and pulsation amplitude of the blood vessels in each segment over time.

[0028] In the time-series segmentation results, the volume of the right ventricle at each time point is calculated to obtain the ventricular volume parameters that change with time. The cardiopulmonary structural parameters are then associated with the ventricular volume parameters in chronological order to form a three-dimensional time-varying cardiopulmonary structural parameter set containing information on the changes in vessel diameter, wall thickness, pulsation amplitude, and ventricular volume over time.

[0029] Optionally, the output hemodynamic parameter vector includes:

[0030] From the three-dimensional time-varying cardiopulmonary structural parameter set, the vessel diameter, vessel wall thickness and pulsation amplitude are extracted according to the vessel segment and time sequence. The parameters of the right atrium, right ventricle and pulmonary artery trunk and multi-level branches in one or more complete cardiac cycles are organized into morphological input sequences.

[0031] Echocardiogram data and electrocardiogram signals are extracted from multimodal medical data. The echocardiogram data is divided into cardiac cycles and sampled. The electrocardiogram signals are identified by heartbeats and divided into rhythms. The resulting time series is organized into a dynamic input sequence that is aligned with the morphological input sequence in terms of cardiac cycles.

[0032] Laboratory test data and disease course-related information are extracted from multimodal medical data. The laboratory test data are classified according to the test time point, and the disease course-related information is organized according to the order of consultation time to form a clinical feature input sequence, which is aligned with the dynamic input sequence in the time dimension.

[0033] Construct a three-branch network including morphological feature branches, dynamic feature branches, and clinical feature branches:

[0034] The morphological input sequence is input into the morphological feature branch, which encodes the parameters of the cardiopulmonary structure changing over time according to the vascular segment and time sequence, generating morphological feature representations that characterize the temporal structural features of each vascular segment.

[0035] The dynamic input sequence is input into the dynamic feature branch, which performs multi-scale time feature extraction on cardiac ultrasound data and electrocardiogram signals to generate a dynamic feature representation corresponding to the cardiac cycle.

[0036] The clinical feature input sequence is input into the clinical feature branch, which extracts features from laboratory test data and disease course-related information to generate a clinical feature representation as an adjustment coefficient;

[0037] In the three-branch network, clinical feature representation is used to weight and regulate morphological and dynamic feature representations. The regulated morphological and dynamic feature representations are jointly encoded on multiple cardiac cycles and multiple vascular segments to obtain a fused feature representation, which is then input into the regression output layer to obtain a hemodynamic parameter vector.

[0038] Optionally, constructing the pulmonary circulation energy fingerprint vector includes:

[0039] The pulmonary circulation simulation boundary conditions are established based on hemodynamic parameter vectors. Using the hemodynamic parameter vectors, simulation calculations are performed within a complete cardiac cycle to obtain the time-varying pulmonary circulation pressure and velocity distributions.

[0040] Based on the pressure and velocity distribution in the pulmonary circulation, the main pulmonary artery and its multiple branches are divided into multiple energy analysis segments according to their anatomical orientation. Multiple cross sections are selected within each energy analysis segment, and the pressure and velocity at multiple time points of each cross section are statistically analyzed to obtain the time-series pressure data and time-series velocity data corresponding to each energy analysis segment.

[0041] Based on time-series pressure data and time-series velocity data, energy loss analysis is performed on each energy analysis segment within one or more cardiac cycles to obtain energy loss measurements for each energy analysis segment. Frequency domain analysis is also performed on the time-series pressure data and time-series velocity data to obtain power spectrum characteristics and vorticity characteristics.

[0042] Based on the upstream and downstream relationships of the main pulmonary artery and its multi-level branches, an energy transfer relationship diagram between each energy analysis segment is established. Based on energy loss measurement, power spectrum characteristics and vorticity characteristics, the energy transfer intensity and energy loss gradient between adjacent energy analysis segments are calculated. The energy characteristics of each energy analysis segment are arranged according to the topological order of the pulmonary artery tree structure.

[0043] The energy loss measurement, power spectrum characteristics, vorticity characteristics, energy transfer intensity, and energy loss gradient of each energy analysis segment within the cardiac cycle are encoded according to a preset spatial and temporal order to form a pulmonary circulation energy fingerprint vector.

[0044] Optionally, generating a new pulmonary circulation energy fingerprint vector includes:

[0045] Extract distal impedance parameters corresponding to the pulmonary artery trunk and multi-level branches from the hemodynamic parameter vector, and establish a correspondence between each energy analysis segment in the pulmonary circulation energy fingerprint vector and each distal impedance parameter.

[0046] Based on the correspondence, a perturbation of a preset amplitude is applied to each distal impedance parameter, and the simulation operation is called while keeping other hemodynamic parameters unchanged to obtain the pulmonary circulation energy fingerprint vector change results of each energy analysis segment before and after the perturbation. The sensitivity index corresponding to each distal impedance parameter is determined according to the energy change direction and amplitude of each energy analysis segment.

[0047] Based on the sensitivity index and the preset energy optimization criteria, the sensitivity indices of each far-end impedance parameter are weighted and combined to determine the adjustment direction of each far-end impedance parameter. According to the magnitude of the sensitivity index of each far-end impedance parameter and the preset adjustment step range, the adjustment range of each far-end impedance parameter is determined.

[0048] The distal impedance parameters in the hemodynamic parameter vector are updated according to the adjustment direction and adjustment range to generate an updated hemodynamic parameter vector, while keeping the non-distal impedance parameters in the updated hemodynamic parameter vector consistent with those before the update.

[0049] The updated hemodynamic parameter vector is input into the simulation process, and the hemodynamic simulation and energy analysis are re-executed to generate a new pulmonary circulation energy fingerprint vector.

[0050] Optionally, the output of pulmonary hypertension diagnosis results, disease progression assessment results, or treatment plan prediction results includes:

[0051] The pulmonary circulation energy fingerprint vector and a new pulmonary circulation energy fingerprint vector are obtained. The energy loss characteristics, power spectrum characteristics and vorticity characteristics of the two are compared at the same cardiac cycle and the same pulmonary artery segment to obtain the energy characteristic comparison results.

[0052] Based on the energy feature comparison results, the neopulmonary circulation energy fingerprint vector is judged according to the preset performance improvement conditions. When the neopulmonary circulation energy fingerprint vector meets the preset performance improvement conditions, the update of the distal impedance in the hemodynamic parameter vector is determined as an accepted distal impedance update; otherwise, the update is determined as a rejected distal impedance update.

[0053] When the distal impedance update is accepted, the diagnosis of pulmonary hypertension, the assessment of disease progression, or the prediction of treatment plan are output based on the new pulmonary circulation energy fingerprint vector and its corresponding hemodynamic parameter vector. When the distal impedance update is rejected, the diagnosis of pulmonary hypertension, the assessment of disease progression, or the prediction of treatment plan are output based on the original pulmonary circulation energy fingerprint vector and its corresponding hemodynamic parameter vector.

[0054] The beneficial effects of this invention are:

[0055] This invention constructs a digital twin pulmonary circulation system that can be dynamically updated as the disease progresses, enabling continuous tracking and high-precision characterization of the patient's individualized hemodynamic state. This overcomes the limitations of existing technologies that rely on single examination data, static structural indicators, or empirical inferences for diagnosis. This invention unifies the processing of multimodal medical data and maps it to a three-dimensional time-varying cardiopulmonary structure, achieving a full-link correlation from anatomical structure and hemodynamic changes to pathophysiological characteristics. This allows pulmonary hypertension assessment to move beyond traditional pressure estimation-based methods, more accurately reflecting the overall pulmonary circulation load level and vascular system stress state, thus improving the reliability and sensitivity of diagnosis.

[0056] This invention generates individualized hemodynamic parameter vectors through a three-branch feature inversion network based on morphology, dynamics, and clinical indicators. Furthermore, it constructs a pulmonary circulation energy fingerprint using deep hemodynamic indicators such as energy loss, power spectrum, and eddy current, effectively overcoming the limitations of existing technologies in identifying distal impedance changes and microcirculatory damage areas. The sensitivity analysis and self-updating mechanism between the energy fingerprint and hemodynamic parameters enable the digital twin model to automatically adjust distal impedance based on the patient's actual condition, thereby achieving self-calibration evolution of the pulmonary circulation state. This expands the assessment beyond a single time point, enabling disease progression tracking and perception of abnormal spread trends, significantly improving lesion segment localization, disease outcome assessment, and prognostic evaluation.

[0057] This invention achieves multi-round iterative updates of hemodynamic parameters through energy fingerprint feedback regulation, enabling the digital twin model to not only be used for diagnosis but also to predict the effects of drug or therapeutic interventions. Clinical trial results show that this invention can predict the degree of improvement in blood flow redistribution and energy loss before treatment, and provide physicians with treatment strategies that are most likely to improve patient prognosis, improving the targeting of treatment options and helping to reduce ineffective drug use and reliance on right heart catheterization. In summary, this invention not only improves the diagnostic accuracy of pulmonary hypertension but also enhances the ability to identify disease progression and predict treatment response, providing a significant technological breakthrough and clinical value for the precise, non-invasive, and intelligent diagnosis and treatment of pulmonary hypertension. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0059] Figure 1 This is a schematic diagram of the structure of a pulmonary hypertension diagnosis and treatment simulation system based on digital twin proposed in this invention;

[0060] Figure 2 This is a flowchart illustrating a digital twin-based simulation method for pulmonary hypertension diagnosis and treatment proposed in this invention. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0062] refer to Figure 1 A digital twin-based simulation system for the diagnosis and treatment of pulmonary hypertension includes the following modules:

[0063] The medical examination data processing module is used to acquire and preprocess comprehensive medical examination data of patients to obtain multimodal medical data;

[0064] The cardiopulmonary structure reconstruction module is used for automatic reconstruction based on multimodal medical data to obtain a three-dimensional time-varying cardiopulmonary structure parameter set;

[0065] The three-branch feature and parameter generation module is used to generate hemodynamic parameter vectors from a three-dimensional time-varying cardiopulmonary structural parameter set and multimodal medical data.

[0066] The blood flow simulation and energy fingerprint module is used to perform blood flow simulation based on hemodynamic parameter vectors, obtain pulmonary circulation pressure and velocity distribution, and generate pulmonary circulation energy fingerprint vectors.

[0067] The distal impedance sensitivity and update module is used to determine the direction and magnitude of distal impedance adjustment based on the pulmonary circulation energy fingerprint vector and the hemodynamic parameter vector, and to generate a new pulmonary circulation energy fingerprint vector.

[0068] The energy fingerprint comparison and result output module is used to compare the new pulmonary circulation energy fingerprint vector with the original pulmonary circulation energy fingerprint vector, and determine whether to accept the remote impedance update based on the preset performance improvement conditions.

[0069] refer to Figure 2 A simulation method for pulmonary hypertension diagnosis and treatment based on digital twins, comprising:

[0070] Acquire comprehensive medical examination data from patients, preprocess the comprehensive medical examination data, and form multimodal medical data;

[0071] Automatic structural reconstruction of the right atrium, right ventricle, pulmonary artery trunk and multi-level branches based on MRA image data in multimodal medical data is performed to generate a three-dimensional time-varying cardiopulmonary structural parameter set containing information on the changes of vessel diameter, wall thickness, pulsation amplitude and ventricular volume over time.

[0072] Using a three-dimensional time-varying cardiopulmonary structural parameter set and multimodal medical data as input, a three-branch network consisting of morphological feature branches, dynamic feature branches, and clinical feature branches is constructed to output a hemodynamic parameter vector;

[0073] Based on hemodynamic parameter vectors, simulation calculations are performed to obtain pulmonary circulation pressure and velocity distributions. Energy loss analysis, power spectrum analysis, and vorticity analysis are then performed on the pressure and velocity distributions to construct a pulmonary circulation energy fingerprint vector.

[0074] Based on the sensitivity relationship between the pulmonary circulation energy fingerprint vector and the distal impedance in the hemodynamic parameter vector, the adjustment direction and adjustment range of the distal impedance are determined, the hemodynamic parameter vector is updated, and the hemodynamic simulation is executed again to generate a new pulmonary circulation energy fingerprint vector.

[0075] The new pulmonary circulation energy fingerprint vector is compared with the pulmonary circulation energy fingerprint vector. When the new pulmonary circulation energy fingerprint vector meets the preset performance improvement conditions, the distal impedance update is accepted and the pulmonary hypertension diagnosis result, disease progression assessment result or treatment plan prediction result is output. Otherwise, the distal impedance update is rejected and the corresponding result is output.

[0076] In this embodiment, the comprehensive medical examination data includes MRA imaging data, echocardiography data, electrocardiogram signals, and laboratory test data.

[0077] In this embodiment, the preprocessing of comprehensive medical examination data includes time-phase registration of MRA image data, noise suppression of echocardiogram data, and signal normalization of electrocardiogram signals and laboratory test data.

[0078] In this embodiment, generating a three-dimensional time-varying cardiopulmonary structural parameter set containing information on the changes in vessel diameter, wall thickness, pulsation amplitude, and ventricular volume over time includes:

[0079] MRA image data was selected from multimodal medical data, and the MRA image data was read and the region of interest was cropped to obtain local cardiopulmonary MRA image data containing only the heart and pulmonary artery regions.

[0080] The cardiopulmonary local MRA image data is input into a pre-trained segmentation network, and voxels at each time phase are classified to obtain initial segmentation results that distinguish the right atrium, right ventricle, main pulmonary artery and multi-level branches.

[0081] Based on the initial segmentation results, temporal consistency analysis was performed on the boundary positions and volume changes of the right atrium, right ventricle, pulmonary artery trunk, and multi-level branches at all time phases. Furthermore, a pre-defined pulmonary artery tree-like anatomical topological constraint was used to automatically correct and merge discontinuous and abnormal branches, resulting in a temporal segmentation result of the right atrium, right ventricle, pulmonary artery trunk, and multi-level branches with a consistent topological structure throughout the entire cardiac cycle. The pre-defined pulmonary artery tree-like anatomical topological constraint is as follows:

[0082] Parent-child hierarchical connectivity constraint: Any branch at any level must be continuously connected to its upstream parent branch, and the Euclidean distance between the end coordinate of the upstream parent branch and the starting coordinate of the downstream child branch in three-dimensional space must not exceed the preset tolerance.

[0083] Branch diameter decreasing constraint: According to the anatomical rule of "parent tube diameter is greater than child tube diameter", if the diameter of a child branch is detected to exceed 70% of the diameter of the corresponding parent branch, the correction is automatically triggered, and the abnormal child branch is merged into the adjacent level or the branch affiliation is recalculated.

[0084] Branch angle and number constraints: At the same branching point, the angle between the child branch and the parent branch must be between 15° and 75°, and the number of direct child branches of each parent branch must not exceed three. Excess small branches exceeding the threshold are merged into one equivalent branch according to the minimum diameter principle.

[0085] In the temporal segmentation results, the pulmonary artery trunk and its branches at all levels are divided into several segments along the vessel's course. At multiple time points within each segment, the changes in vessel diameter, vessel wall thickness, and the amplitude of change over time are calculated to obtain cardiopulmonary structural parameters characterizing the changes in vessel diameter, wall thickness, and pulsation amplitude over time in each segment. Specifically, the calculation of the changes in vessel diameter, vessel wall thickness, and the amplitude of change over time involves:

[0086] The vascular lumen boundary is extracted from the original segmentation plane of each segment, and the equivalent circle diameter is measured by the least circumscribed circle fitting method, which is used as the vascular diameter of the segment at the current time point.

[0087] Simultaneously extract the outer wall boundary of the blood vessel on the same cross-section, calculate the average radial distance between the outer wall and the inner lumen boundary, and obtain the blood vessel wall thickness of the cross-section at the current time point;

[0088] Multiple time points are taken along the same segment within a complete cardiac cycle. The maximum and minimum values ​​of the inner diameter of the same section are statistically analyzed. The difference between the two values ​​is the amplitude of the change of the inner diameter of the section over time. The average of all section results by segment is used to obtain the pulsation amplitude of that segment.

[0089] In the time-series segmentation results, the volume of the right ventricle at each time point is calculated to obtain the ventricular volume parameters that change with time. The cardiopulmonary structural parameters are then associated with the ventricular volume parameters in chronological order to form a three-dimensional time-varying cardiopulmonary structural parameter set containing information on the changes in vessel diameter, wall thickness, pulsation amplitude, and ventricular volume over time.

[0090] In this embodiment, the output hemodynamic parameter vector includes:

[0091] From the three-dimensional time-varying cardiopulmonary structural parameter set, the vessel diameter, vessel wall thickness and pulsation amplitude are extracted according to the vessel segment and time sequence. The parameters of the right atrium, right ventricle and pulmonary artery trunk and multi-level branches in one or more complete cardiac cycles are organized into morphological input sequences.

[0092] Echocardiogram data and electrocardiogram signals are extracted from multimodal medical data. The echocardiogram data is divided into cardiac cycles and sampled. The electrocardiogram signals are identified by heartbeats and divided into rhythms. The resulting time series is organized into a dynamic input sequence that is aligned with the morphological input sequence in terms of cardiac cycles.

[0093] Laboratory test data and disease course-related information are extracted from multimodal medical data. The laboratory test data are classified according to the test time point, and the disease course-related information is organized according to the order of consultation time to form a clinical feature input sequence, which is aligned with the dynamic input sequence in the time dimension.

[0094] Construct a three-branch network including morphological feature branches, dynamic feature branches, and clinical feature branches:

[0095] The morphological input sequence is fed into the morphological feature branch. The morphological feature branch encodes the parameters of the cardiopulmonary structure changing over time according to vascular segments and time sequence, generating morphological feature representations characterizing the temporal structural features of each vascular segment. Specifically, the morphological feature branch encodes the parameters of the cardiopulmonary structure changing over time according to vascular segments and time sequence as follows:

[0096] Equidistant cross sections were sequentially taken along the main pulmonary artery and its branches at various levels. The vessel diameter, wall thickness, and cross-sectional area within the same segment were arranged in the order of cardiac cycle time to form a segment-time two-dimensional matrix.

[0097] For a two-dimensional matrix, one-dimensional convolution is first used to extract local temporal variation patterns, and then gated recurrent units are used to cyclically encode the segment sequence to generate a temporal embedding vector that reflects the expansion-contraction law of a single segment.

[0098] The temporal embedding vectors of all segments are positionally encoded according to the parent-child topological order of the pulmonary artery tree, and upstream and downstream vascular dependence information is fused through a self-attention layer to obtain a morphological feature representation of the dynamic structural relationship of the entire pulmonary artery tree.

[0099] The dynamic input sequence is input into the dynamic feature branch, which performs multi-scale time feature extraction on cardiac ultrasound data and electrocardiogram signals to generate a dynamic feature representation corresponding to the cardiac cycle.

[0100] The clinical feature input sequence is input into the clinical feature branch, which extracts features from laboratory test data and disease course-related information to generate a clinical feature representation as an adjustment coefficient;

[0101] In the three-branch network, clinical feature representation is used to weight and regulate morphological and dynamic feature representations. The regulated morphological and dynamic feature representations are jointly encoded on multiple cardiac cycles and multiple vascular segments to obtain a fused feature representation, which is then input into the regression output layer to obtain a hemodynamic parameter vector.

[0102] In this embodiment, constructing the pulmonary circulation energy fingerprint vector includes:

[0103] The pulmonary circulation simulation boundary conditions are established based on hemodynamic parameter vectors. Using the hemodynamic parameter vectors, simulation calculations are performed within a complete cardiac cycle to obtain the time-varying pulmonary circulation pressure and velocity distributions.

[0104] Based on the pressure and velocity distribution in the pulmonary circulation, the main pulmonary artery and its multiple branches are divided into multiple energy analysis segments according to their anatomical orientation. Multiple cross sections are selected within each energy analysis segment, and the pressure and velocity at multiple time points of each cross section are statistically analyzed to obtain the time-series pressure data and time-series velocity data corresponding to each energy analysis segment.

[0105] Based on time-series pressure and velocity data, energy loss analysis is performed on each energy analysis segment within one or more cardiac cycles to obtain energy loss measurements for each segment. Frequency domain analysis is then performed on the time-series pressure and velocity data to obtain power spectral characteristics and vorticity characteristics. Specifically, the energy loss analysis for each energy analysis segment within one or more cardiac cycles involves:

[0106] Based on the velocity field, the velocity gradient of each voxel in the segment is spatially discretized, the instantaneous viscous shear power is calculated and integrated over the entire cross-sectional area to obtain the instantaneous internal dissipation power of the segment.

[0107] Multiply the pressure difference between the upstream and downstream sections of the segment by the volumetric flow rate to obtain the external power loss caused by the pressure drop per unit time, and add it to the instantaneous internal power loss to obtain the instantaneous total energy loss rate of the segment.

[0108] The instantaneous total energy loss rate is integrated over the complete cardiac cycle and normalized to a cardiac cycle average value, which is used as the energy loss measure for that segment. In multi-cardiac cycle scenarios, the average value of each cycle is taken to smooth out the effects of respiratory and load fluctuations.

[0109] Based on the upstream and downstream relationships of the pulmonary artery trunk and its multi-level branches, an energy transfer relationship diagram is established between each energy analysis segment. The energy transfer intensity and energy loss gradient between adjacent energy analysis segments are calculated based on energy loss measurement, power spectrum characteristics, and vorticity characteristics. The energy characteristics of each energy analysis segment are arranged according to the topological order of the pulmonary artery tree structure. Specifically, the calculation of the energy transfer intensity and energy loss gradient between adjacent energy analysis segments based on energy loss measurement, power spectrum characteristics, and vorticity characteristics involves:

[0110] Integrate the instantaneous kinetic energy and pressure potential energy along the cross section for adjacent upstream and downstream segments respectively, calculate the net energy flux difference per unit time, and divide the difference by the cross-sectional area of ​​the upstream segment to obtain the normalized energy transfer intensity.

[0111] In the frequency domain, the energy ratio of the main peak frequency band of the power spectrum of adjacent segments is extracted. Combined with the ratio of the root mean square value of the vorticity of the two segments, a frequency-vortex joint attenuation coefficient is constructed to correct the normalized energy transfer intensity and obtain the final energy transfer intensity index.

[0112] Accumulate the energy loss measurement of each segment during the cardiac cycle, calculate the difference in energy loss measurement between adjacent segments and divide it by the distance between the centerlines of the two segments. The resulting ratio is the energy loss gradient, and the positive or negative gradient indicates the direction of energy dissipation.

[0113] The energy loss measurement, power spectrum characteristics, vorticity characteristics, energy transfer intensity, and energy loss gradient of each energy analysis segment within the cardiac cycle are encoded according to a preset spatial and temporal order to form a pulmonary circulation energy fingerprint vector. The preset spatial and temporal order is as follows:

[0114] The spatial sequence follows the rule of from near to far and from left to right within the same layer. That is, starting from the main pulmonary artery, the branches of each level are traversed sequentially along the direction of blood flow. The main trunk is recorded first, followed by the first-level branches on the left and right. The downstream branches of each branch continue to expand in layers until the terminal microcirculation.

[0115] Within the same level, if there are multiple parallel branches, they are ordered according to their anatomical position from left to right in the coronal plane to ensure that the energy feature encoding order of the left and right pulmonary artery segments is consistent.

[0116] The time sequence is based on the triggering of the R wave in the electrocardiogram signal as the zero moment. The complete cardiac cycle is evenly divided into twenty time windows. The energy loss measurement, power spectrum characteristics, vorticity characteristics, energy transfer intensity and energy loss gradient of adjacent segments are serially encoded according to the window number.

[0117] In this embodiment, generating a new pulmonary circulation energy fingerprint vector includes:

[0118] Extract distal impedance parameters corresponding to the pulmonary artery trunk and multi-level branches from the hemodynamic parameter vector, and establish a correspondence between each energy analysis segment in the pulmonary circulation energy fingerprint vector and each distal impedance parameter.

[0119] Based on the correspondence, a perturbation of a preset amplitude is applied to each distal impedance parameter, and the simulation operation is called while keeping other hemodynamic parameters unchanged to obtain the pulmonary circulation energy fingerprint vector change results of each energy analysis segment before and after the perturbation. The sensitivity index corresponding to each distal impedance parameter is determined according to the energy change direction and amplitude of each energy analysis segment.

[0120] Based on the sensitivity index and the preset energy optimization criteria, the sensitivity indices of each far-end impedance parameter are weighted and combined to determine the adjustment direction of each far-end impedance parameter. According to the magnitude of the sensitivity index of each far-end impedance parameter and the preset adjustment step size range, the adjustment range of each far-end impedance parameter is determined. The preset adjustment step size range is 0.5% to 5% of the current value of the far-end impedance. When the sensitivity index is in the low range, a step size of 0.5% is used for fine adjustment. When it is in the high range, a step size of up to 5% can be used for rapid adjustment. Adjustments exceeding this range will be limited to the value close to the upper limit.

[0121] The distal impedance parameters in the hemodynamic parameter vector are updated according to the adjustment direction and adjustment range to generate an updated hemodynamic parameter vector, while keeping the non-distal impedance parameters in the updated hemodynamic parameter vector consistent with those before the update.

[0122] The updated hemodynamic parameter vector is input into the simulation process, and the hemodynamic simulation and energy analysis are re-executed to generate a new pulmonary circulation energy fingerprint vector.

[0123] In this embodiment, the output of pulmonary hypertension diagnosis results, disease progression assessment results, or treatment plan prediction results includes:

[0124] The pulmonary circulation energy fingerprint vector and a new pulmonary circulation energy fingerprint vector are obtained. The energy loss characteristics, power spectrum characteristics and vorticity characteristics of the two are compared at the same cardiac cycle and the same pulmonary artery segment to obtain the energy characteristic comparison results.

[0125] Based on the energy feature comparison results, the neopulmonary circulation energy fingerprint vector is judged according to the preset performance improvement conditions. When the neopulmonary circulation energy fingerprint vector meets the preset performance improvement conditions, the update of the distal impedance in the hemodynamic parameter vector is determined as an accepted distal impedance update; otherwise, the update is determined as a rejected distal impedance update.

[0126] When the distal impedance update is accepted, the diagnosis of pulmonary hypertension, the assessment of disease progression, or the prediction of treatment plan are output based on the new pulmonary circulation energy fingerprint vector and its corresponding hemodynamic parameter vector. When the distal impedance update is rejected, the diagnosis of pulmonary hypertension, the assessment of disease progression, or the prediction of treatment plan are output based on the original pulmonary circulation energy fingerprint vector and its corresponding hemodynamic parameter vector.

[0127] Example 1:

[0128] To verify the feasibility of this invention in practice, it was applied to Patient B, a patient with a high suspicion of pulmonary hypertension, admitted to the cardiology department of a tertiary hospital. The patient had a long history of shortness of breath, palpitations, fatigue, and mild orthopnea at night. Previous echocardiograms had suggested elevated right ventricular systolic pressure, but the results fluctuated significantly, lacking a dynamic trend assessment. The patient repeatedly refused invasive evaluations due to concerns about the risks of right heart catheterization, making the question of disease progression and the need for immediate treatment adjustments crucial. The digital twin-based pulmonary hypertension diagnosis and treatment simulation method proposed in this invention was applied to the diagnosis and treatment prediction decision-making process for this patient.

[0129] In clinical practice, the patient's 4D-MRA images, transthoracic echocardiogram, 12-lead electrocardiogram data, and laboratory test indicators are first collected through the hospital's medical imaging and physiological signal acquisition system. All data are synchronized in time, noise suppressed, and format standardized through the integrated medical examination data processing module of this invention. Subsequently, based on the MRA images, the right atrium, right ventricle, main pulmonary artery, and branch vessels are automatically reconstructed to form a three-dimensional time-varying cardiopulmonary structural parameter set that changes with the cardiac cycle, including the change of vessel radius over time, the distribution of vessel wall thickness, the pulmonary artery pulsation amplitude curve, and the right ventricular volume change curve.

[0130] This invention employs a three-branch neural network to invert hemodynamic parameters in patients. The morphological feature branch extracts curves of vessel diameter changes, wall thickness, and pulsation amplitude; the dynamic branch analyzes ultrasound and electrocardiogram signals to capture hemodynamic fluctuations and cardiac electrophysiological status; and the clinical feature branch analyzes biochemical indicators such as NT-proBNP and patient disease progression information. After multimodal fusion, a hemodynamic parameter vector is generated, including key indicators such as vascular elasticity, pulmonary artery resistance, blood flow inertia, compliance, and distal impedance. Subsequently, this invention performs blood flow simulation and constructs a pulmonary circulation energy fingerprint vector, representing the pulmonary circulation energy state using energy loss, power spectrum, and eddy current variation patterns.

[0131] Simulation results indicated that the patient exhibited significant energy loss abnormalities in the 5th-7th grade branches of the pulmonary artery, with distal impedance 47% higher than that of the normal control group, representing a typical microcirculation restriction pattern. Subsequently, the model initiated distal impedance sensitivity analysis and parameter self-updating mechanisms. Based on the sensitivity relationship between energy fingerprint and impedance, the model converged after multiple simulation iterations, predicting a mean pulmonary artery pressure of 41 mmHg and pulmonary artery resistance of 8.2 Wood units, indicating a moderate to severe pulmonary hypertension distribution pattern. This invention predicts the improvement effects of different treatment strategies, with the phosphodiesterase-5 inhibitor combined with an endothelin receptor antagonist regimen showing significantly better energy improvement than the single-drug regimen, and the most significant reduction in right ventricular load is expected within 8 weeks.

[0132] To compare the accuracy of the model's predictions with actual clinical outcomes, patients were followed up for 3 months, during which time they followed the combination treatment regimen recommended by this invention.

[0133] Table 1. Comparison of Traditional Assessment Methods and Simulation Assessment of the Invention—Patient B Follow-up Validation Data Table

[0134] Indicator Type Traditional ultrasound results Simulation prediction results of this invention Catheter test results after 3 months Comparison with measured deviation Mean pulmonary artery pressure (mmHg) 33 41 43 The error of this invention is ↓67% Pulmonary artery resistance (Wood units) Unable to measure 8.2 8.5 High consistency Remote impedance identification Unrecognized Significantly increased (+47%) The inspection confirmed Locable Microcirculation energy loss location Unable to locate Level 5–7 branch anomalies Post-treatment follow-up improvement Consistent Treatment response prediction Unquantifiable Combined treatment regimens predict effectiveness Clinically validated Significant advantages NT-proBNP improves prediction Unquantifiable A decrease of approximately 30% Actual decline of 35% High credibility 6-minute walking distance improvement prediction Unquantifiable Lift 60–80 meters The actual lift was 73 meters. High accuracy

[0135] As shown in Table 1, this invention is significantly superior to traditional ultrasound in assessing core indicators of pulmonary hypertension. Traditional ultrasound estimated the average pulmonary artery pressure at 33 mmHg, while the actual measurement by right heart catheterization three months later was 43 mmHg, indicating a significant deviation and a potential underestimation of the patient's condition. The simulation prediction result of this invention was 41 mmHg, differing from the measured value by only 2 mmHg, demonstrating a significant reduction in error. This indicates that hemodynamic simulation based on digital twins can more accurately reflect the patient's true pulmonary circulation pressure level, providing a more reliable basis for disease grading and intervention timing.

[0136] In the assessment of deep hemodynamic function, traditional methods cannot provide quantitative information on pulmonary artery resistance, distal impedance, and microcirculatory energy loss, and can only make rough inferences. This invention not only provides near-measured pulmonary artery resistance but also identifies significantly elevated distal impedance and precisely locates abnormal energy loss in the 5th–7th grade branches. Subsequent examinations and follow-ups confirmed that this branch region was indeed the focus of the lesion. This demonstrates that this invention can reveal microcirculatory burden and lesion distribution at the segmental level, achieving a diagnostic elevation from "whether there is a problem" to "where the problem is."

[0137] In terms of predicting and following up on treatment effects, traditional methods cannot quantify the expected benefits of different treatment regimens, relying solely on experience and post-treatment observation. This invention, however, quantitatively predicts the effectiveness of combined treatment regimens and the extent of improvement in NT-proBNP and 6-minute walk distance before treatment. Three months later, the actual decreases and increases closely matched the predictions, demonstrating the high reliability and clinical guidance value of this invention in efficacy prediction. This invention not only improves diagnostic accuracy but also enhances lesion localization and treatment response prediction capabilities, providing strong technical support for the non-invasive and precise diagnosis and treatment of pulmonary hypertension.

[0138] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A pulmonary hypertension diagnosis and treatment simulation system based on digital twins, characterized in that, Includes the following modules: The medical examination data processing module is used to acquire and preprocess comprehensive medical examination data of patients to obtain multimodal medical data; The cardiopulmonary structure reconstruction module is used for automatic reconstruction based on multimodal medical data to obtain a three-dimensional time-varying set of cardiopulmonary structure parameters. The three-branch feature and parameter generation module is used to generate hemodynamic parameter vectors from a three-dimensional time-varying cardiopulmonary structural parameter set and multimodal medical data. The blood flow simulation and energy fingerprint module is used to perform blood flow simulation based on hemodynamic parameter vectors, obtain pulmonary circulation pressure and velocity distribution, and generate pulmonary circulation energy fingerprint vectors. The distal impedance sensitivity and update module is used to determine the direction and magnitude of distal impedance adjustment based on the pulmonary circulation energy fingerprint vector and the hemodynamic parameter vector, and to generate a new pulmonary circulation energy fingerprint vector. The energy fingerprint comparison and result output module is used to compare the new pulmonary circulation energy fingerprint vector with the original pulmonary circulation energy fingerprint vector, and determine whether to accept the remote impedance update based on the preset performance improvement conditions.

2. A digital twin-based simulation method for pulmonary hypertension diagnosis and treatment, applied to the digital twin-based simulation system for pulmonary hypertension diagnosis and treatment as described in claim 1, characterized in that, include: Acquire comprehensive medical examination data from patients, preprocess the comprehensive medical examination data, and form multimodal medical data; Automatic structural reconstruction of the right atrium, right ventricle, pulmonary artery trunk and multi-level branches based on MRA image data in multimodal medical data is performed to generate a three-dimensional time-varying cardiopulmonary structural parameter set containing information on the changes of vessel diameter, wall thickness, pulsation amplitude and ventricular volume over time. Using a three-dimensional time-varying cardiopulmonary structural parameter set and multimodal medical data as input, a three-branch network consisting of morphological feature branches, dynamic feature branches, and clinical feature branches is constructed to output a hemodynamic parameter vector; Based on hemodynamic parameter vectors, simulation calculations are performed to obtain pulmonary circulation pressure and velocity distributions. Energy loss analysis, power spectrum analysis, and vorticity analysis are then performed on the pressure and velocity distributions to construct a pulmonary circulation energy fingerprint vector. Based on the sensitivity relationship between the pulmonary circulation energy fingerprint vector and the distal impedance in the hemodynamic parameter vector, the adjustment direction and adjustment range of the distal impedance are determined, the hemodynamic parameter vector is updated, and the hemodynamic simulation is executed again to generate a new pulmonary circulation energy fingerprint vector. The new pulmonary circulation energy fingerprint vector is compared with the pulmonary circulation energy fingerprint vector. When the new pulmonary circulation energy fingerprint vector meets the preset performance improvement conditions, the distal impedance update is accepted and the pulmonary hypertension diagnosis result, disease progression assessment result or treatment plan prediction result is output. Otherwise, the distal impedance update is rejected and the corresponding result is output.

3. The method for pulmonary hypertension diagnosis and treatment simulation based on digital twins according to claim 2, characterized in that, The comprehensive medical examination data includes MRA imaging data, echocardiogram data, electrocardiogram signals, and laboratory test data.

4. The method for pulmonary hypertension diagnosis and treatment simulation based on digital twins according to claim 2, characterized in that, The preprocessing of comprehensive medical examination data includes time-phase registration of MRA image data, noise suppression of echocardiogram data, and signal normalization of electrocardiogram signals and laboratory test data.

5. The method for pulmonary hypertension diagnosis and treatment simulation based on digital twins according to claim 2, characterized in that, The generation of a three-dimensional time-varying cardiopulmonary structural parameter set, including information on the changes in vessel diameter, wall thickness, pulsation amplitude, and ventricular volume over time, includes: MRA image data was selected from multimodal medical data, and the MRA image data was read and the region of interest was cropped to obtain local cardiopulmonary MRA image data containing only the heart and pulmonary artery regions. The cardiopulmonary local MRA image data is input into a pre-trained segmentation network, and voxels at each time phase are classified to obtain initial segmentation results that distinguish the right atrium, right ventricle, main pulmonary artery and multi-level branches. Based on the initial segmentation results, temporal consistency analysis was performed on the boundary positions and volume changes of the right atrium, right ventricle, pulmonary artery trunk and multi-level branches under all time phases. The pre-set pulmonary artery tree anatomical topology constraints were used to automatically correct and merge discontinuous and abnormal branches, forming a temporal segmentation result of the right atrium, right ventricle, pulmonary artery trunk and multi-level branches with a consistent topological structure throughout the entire cardiac cycle. In the time-series segmentation results, the pulmonary artery trunk and its branches at all levels are divided into several segments along the direction of the blood vessels. The changes in the inner diameter of the blood vessels, the thickness of the blood vessel walls, and the amplitude of the inner diameter over time are calculated at multiple time points in each segment to obtain cardiopulmonary structural parameters characterizing the changes in the diameter, wall thickness, and pulsation amplitude of the blood vessels in each segment over time. In the time-series segmentation results, the volume of the right ventricle at each time point is calculated to obtain the ventricular volume parameters that change with time. The cardiopulmonary structural parameters are then associated with the ventricular volume parameters in chronological order to form a three-dimensional time-varying cardiopulmonary structural parameter set containing information on the changes in vessel diameter, wall thickness, pulsation amplitude, and ventricular volume over time.

6. The method for pulmonary hypertension diagnosis and treatment simulation based on digital twins according to claim 2, characterized in that, The output hemodynamic parameter vector includes: From the three-dimensional time-varying cardiopulmonary structural parameter set, the vessel diameter, vessel wall thickness and pulsation amplitude are extracted according to the vessel segment and time sequence. The parameters of the right atrium, right ventricle and pulmonary artery trunk and multi-level branches in one or more complete cardiac cycles are organized into morphological input sequences. Echocardiogram data and electrocardiogram signals are extracted from multimodal medical data. The echocardiogram data is divided into cardiac cycles and sampled. The electrocardiogram signals are identified by heartbeats and divided into rhythms. The resulting time series is organized into a dynamic input sequence that is aligned with the morphological input sequence in terms of cardiac cycles. Laboratory test data and disease course-related information are extracted from multimodal medical data. The laboratory test data are classified according to the test time point, and the disease course-related information is organized according to the order of consultation time to form a clinical feature input sequence, which is aligned with the dynamic input sequence in the time dimension. Construct a three-branch network including morphological feature branches, dynamic feature branches, and clinical feature branches: The morphological input sequence is input into the morphological feature branch, which encodes the parameters of the cardiopulmonary structure changing over time according to the vascular segment and time sequence, generating morphological feature representations that characterize the temporal structural features of each vascular segment. The dynamic input sequence is input into the dynamic feature branch, which performs multi-scale time feature extraction on cardiac ultrasound data and electrocardiogram signals to generate a dynamic feature representation corresponding to the cardiac cycle. The clinical feature input sequence is input into the clinical feature branch, which extracts features from laboratory test data and disease course-related information to generate a clinical feature representation as an adjustment coefficient; In the three-branch network, clinical feature representation is used to weight and regulate morphological and dynamic feature representations. The regulated morphological and dynamic feature representations are jointly encoded on multiple cardiac cycles and multiple vascular segments to obtain a fused feature representation, which is then input into the regression output layer to obtain a hemodynamic parameter vector.

7. The method for pulmonary hypertension diagnosis and treatment simulation based on digital twins according to claim 2, characterized in that, The construction of the pulmonary circulation energy fingerprint vector includes: The pulmonary circulation simulation boundary conditions are established based on hemodynamic parameter vectors. Using the hemodynamic parameter vectors, simulation calculations are performed within a complete cardiac cycle to obtain the time-varying pulmonary circulation pressure and velocity distributions. Based on the pressure and velocity distribution in the pulmonary circulation, the main pulmonary artery and its multiple branches are divided into multiple energy analysis segments according to their anatomical orientation. Multiple cross sections are selected within each energy analysis segment, and the pressure and velocity at multiple time points of each cross section are statistically analyzed to obtain the time-series pressure data and time-series velocity data corresponding to each energy analysis segment. Based on time-series pressure data and time-series velocity data, energy loss analysis is performed on each energy analysis segment within one or more cardiac cycles to obtain energy loss measurements for each energy analysis segment. Frequency domain analysis is also performed on the time-series pressure data and time-series velocity data to obtain power spectrum characteristics and vorticity characteristics. Based on the upstream and downstream relationships of the main pulmonary artery and its multi-level branches, an energy transfer relationship diagram between each energy analysis segment is established. Based on energy loss measurement, power spectrum characteristics and vorticity characteristics, the energy transfer intensity and energy loss gradient between adjacent energy analysis segments are calculated. The energy characteristics of each energy analysis segment are arranged according to the topological order of the pulmonary artery tree structure. The energy loss measurement, power spectrum characteristics, vorticity characteristics, energy transfer intensity, and energy loss gradient of each energy analysis segment within the cardiac cycle are encoded according to a preset spatial and temporal order to form a pulmonary circulation energy fingerprint vector.

8. The pulmonary hypertension diagnosis and treatment simulation method based on digital twins according to claim 2, characterized in that, The generation of the new pulmonary circulation energy fingerprint vector includes: Extract distal impedance parameters corresponding to the pulmonary artery trunk and multi-level branches from the hemodynamic parameter vector, and establish a correspondence between each energy analysis segment in the pulmonary circulation energy fingerprint vector and each distal impedance parameter. Based on the correspondence, a perturbation of a preset amplitude is applied to each distal impedance parameter, and the simulation operation is called while keeping other hemodynamic parameters unchanged to obtain the pulmonary circulation energy fingerprint vector change results of each energy analysis segment before and after the perturbation. The sensitivity index corresponding to each distal impedance parameter is determined according to the energy change direction and amplitude of each energy analysis segment. Based on the sensitivity index and the preset energy optimization criteria, the sensitivity indices of each far-end impedance parameter are weighted and combined to determine the adjustment direction of each far-end impedance parameter. According to the magnitude of the sensitivity index of each far-end impedance parameter and the preset adjustment step range, the adjustment range of each far-end impedance parameter is determined. The distal impedance parameters in the hemodynamic parameter vector are updated according to the adjustment direction and adjustment range to generate an updated hemodynamic parameter vector, while keeping the non-distal impedance parameters in the updated hemodynamic parameter vector consistent with those before the update. The updated hemodynamic parameter vector is input into the simulation process, and the hemodynamic simulation and energy analysis are re-executed to generate a new pulmonary circulation energy fingerprint vector.

9. A simulation method for pulmonary hypertension diagnosis and treatment based on digital twins according to claim 2, characterized in that, The output of pulmonary hypertension diagnosis results, disease progression assessment results, or treatment plan prediction results includes: The pulmonary circulation energy fingerprint vector and a new pulmonary circulation energy fingerprint vector are obtained. The energy loss characteristics, power spectrum characteristics and vorticity characteristics of the two are compared at the same cardiac cycle and the same pulmonary artery segment to obtain the energy characteristic comparison results. Based on the energy feature comparison results, the neopulmonary circulation energy fingerprint vector is judged according to the preset performance improvement conditions. When the neopulmonary circulation energy fingerprint vector meets the preset performance improvement conditions, the update of the distal impedance in the hemodynamic parameter vector is determined as an accepted distal impedance update; otherwise, the update is determined as a rejected distal impedance update. When the distal impedance update is accepted, the diagnosis of pulmonary hypertension, the assessment of disease progression, or the prediction of treatment plan are output based on the new pulmonary circulation energy fingerprint vector and its corresponding hemodynamic parameter vector. When the distal impedance update is rejected, the diagnosis of pulmonary hypertension, the assessment of disease progression, or the prediction of treatment plan are output based on the original pulmonary circulation energy fingerprint vector and its corresponding hemodynamic parameter vector.

Images