Portal vein state detection method based on portal vein pressure redistribution

Abstract

The invention belongs to the technical field of portal vein noninvasive detection, and provides a portal vein pressure redistribution-based portal vein state detection method, which comprises the following steps of: acquiring a to-be-detected CBCTA image and detecting a portal vein state detection model; the construction process of the portal vein state detection model comprises the steps of to-be-analyzed CBCTA image acquisition, portal vein three-dimensional model generation, CFD simulation setting, block segmentation, pressure field distribution data extraction, portal vein pressure actual data acquisition, model correction, multi-working-condition simulation and ConvNeXt model training. According to the method, the three-dimensional portal vein model is constructed through digital twinning, the model under multiple working conditions is simulated, the treatment scheme is determined according to the simulation result, the portal vein detection effect is improved, and the optimal treatment scheme is determined.

Description

Technical Field

[0001] The present invention relates to the technical field of non-invasive detection of the portal vein, and particularly to a method for detecting the status of the portal vein based on the redistribution of portal vein pressure. Background Art

[0002] The portal vein originates at the level of the second lumbar vertebra and is mostly formed by the confluence of the superior mesenteric portal vein and the splenic portal vein behind the neck of the pancreas. The tributaries of the portal vein include the superior mesenteric portal vein, the splenic portal vein, the inferior mesenteric portal vein, the left gastric portal vein, the right gastric portal vein, the cystic portal vein, and the paraumbilical portal vein, etc., and mostly run parallel to the arteries of the same name. Liver cirrhosis is the most common cause of portal hypertension; portal hypertension can cause abdominal distension, abdominal discomfort, disturbance of consciousness, and gastrointestinal bleeding. Doctors need to make a clear diagnosis based on symptoms, physical examination results, and sometimes with the help of ultrasound examination, computed tomography, magnetic resonance imaging, or liver biopsy. However, the traditional detection methods are too time-consuming and rely on the personal experience of doctors. To solve the above problems, researchers have proposed two solutions: The first technology is based on three-dimensional modeling and fluid mechanics simulation. It uses CTA images to construct a three-dimensional model of the portal vein system, simulates the blood flow characteristics of the liver through a porous medium model, and performs numerical calculations by combining Darcy's law and the Navier-Stokes equation to directly output the portal vein pressure gradient. Its innovation lies in regarding the liver as a porous medium and making the simulation results consistent with invasive measurements by dynamically iteratively correcting model parameters (such as combining ultrasound hardness data), and providing visual pressure distribution and flow field analysis; The second technology adopts the path of combining image features and statistical modeling. It classifies patients by analyzing the shunt status in hepatic portal venography and establishes different regression models for different shunt types such as early, middle, and late stages, mapping the HVPG detection value to the PPG prediction value. The advantage is that it simplifies the calculation process and is suitable for rapid clinical decision-making.

[0003] However, the above two methods only analyze the portal vein pressure in specific situations, do not perform multi-condition simulations, and do not determine treatment plans based on the simulation results, and still need to be improved. Summary of the Invention

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a method for detecting the status of the portal vein based on the redistribution of portal vein pressure, so as to realize reliable detection of the portal vein under multiple conditions.

[0005] To achieve the above purpose, the present invention provides the following solution:

[0006] A method for detecting the status of the portal vein based on the redistribution of portal vein pressure, comprising:

[0007] Collecting the CBCTA image to be detected of the abdomen of the patient to be detected;

[0008] Input the CBCTA image to be detected into the portal vein status detection model to obtain the portal vein pressure detection result including the portal vein occlusion status;

[0009] Match the portal vein pressure detection result with the preset treatment plan library to obtain the guiding treatment plan;

[0010] The construction process of the portal vein status detection model includes:

[0011] Collect the CBCTA image to be analyzed of the abdomen of the pre-screened patients;

[0012] Generate a three-dimensional model of the portal vein according to the CBCTA image to be analyzed;

[0013] Perform CFD simulation settings on the three-dimensional model of the portal vein to obtain a portal vein hydrodynamic model;

[0014] Perform block segmentation on the portal vein hydrodynamic model and create a hydrodynamic sub-model of the portacaval shunt channel to obtain multiple block hydrodynamic sub-models;

[0015] Output the pressure simulation data of each block hydrodynamic sub-model to obtain the pressure field distribution data;

[0016] Collect the intraoperative portal vein pressure measurement data of the pre-screened patients to obtain the actual portal vein pressure data;

[0017] Use the actual portal vein pressure data and the pressure field distribution data to correct each block hydrodynamic sub-model to obtain a corrected block hydrodynamic correction model;

[0018] Set the working condition status of each block hydrodynamic correction model according to the preset occlusion and shunt schemes, and collect the pressure simulation data of the block hydrodynamic correction model under different working condition statuses to obtain multi-condition simulation data with working condition labels;

[0019] Input the CBCTA image to be analyzed and the multi-condition simulation data into the pre-constructed ConvNeXt model for training to obtain a trained portal vein status detection model.

[0020] Preferably, inputting the CBCTA image to be analyzed and the multi-condition simulation data into the pre-constructed ConvNeXt model for training to obtain a trained portal vein status detection model includes:

[0021] Use the N5 bias field correction algorithm introducing hemodynamic priors to eliminate artifacts from the CBCTA image to be analyzed to obtain a preprocessed CBCTA image; the bias field estimation expression of the N5 bias field correction algorithm is:

[0022]

[0023] Among them, B(x) is the partial field estimation value; λ0 is the initial regularization coefficient; V(x) is the observed image intensity value; T(x) is the hemodynamic prior term; x is the original coordinate point;

[0024] Resample the multi-condition simulation data to obtain resampled data with the same resolution as the preprocessed CBCTA image;

[0025] Use the least squares method to match and locally deform and compensate the preprocessed CBCTA image and the resampled data to obtain a 4D input tensor; the expression of the local deformation compensation is:

[0026]

[0027] Among them, is the image intensity value; A is the affine transformation matrix; b is the translation vector; w i is the weight coefficient of the i-th control point; is the radial basis function; x i is the i-th control point; n is the predefined number of control points;

[0028] Design the ConvNeXt architecture according to the 4-stage structure; the number of channels of the 4-stage structure are 96, 192, 384, and 768 respectively;

[0029] Set a cross-modal attention gate between the third and fourth layers of the ConvNeXt architecture to obtain an initial network model;

[0030] Determine a loss function including a working condition classification loss, a pressure regression loss, and an anatomical constraint loss; the expression of the working condition classification loss is:

[0031]

[0032] Among them, is the output value of the working condition classification loss; w c is the weight of the c-th category; C is the total number of working condition categories; τ is the temperature coefficient; is the output score of the model for the input on the c-th category; λ ortho is the working condition orthogonal regularization coefficient; F is the feature matrix extracted by the network; I is the identity matrix with the same dimension as F; E[·] is the expectation operator;

[0033] The expression of the pressure regression loss is:

[0034]

[0035] Among them, δ = 0.1mmH; is the output value of the pressure regression loss; α l is the weight coefficient of the l-th layer; is a piecewise function; a is the piecewise variable; δ is the preset threshold; p l is the simulated pressure; is the true pressure; λ mass is the pressure regularization coefficient; ρ is the density; u is the velocity; λ dir is the direction consistency regularization coefficient;

[0036] The expression of the anatomical constraint loss is:

[0037]

[0038] Among them, g3 = max(0, 8 - R min ); is the output value of the anatomical constraint loss; λ topo is the topological constraint regularization coefficient; is the persistent homology distance; λ geo is the geometric constraint regularization coefficient; g z is the angle constraint of the z-th blood vessel branch; λ wasser is the distance regularization coefficient; W1(P seg , P prior ) is the first-order Wasserstein distance between the segmentation result distribution P seg and the prior distribution P prior ; Dgm h (f) is the h-th persistent homology diagram of the model f; Dgm h (S is the h-th persistent homology diagram of the true anatomical structure S; σ is the smoothing coefficient; μ is the integral measure; θ e is the angle of the e-th blood vessel branch; is the set of blood vessel branch edges; d z is the diameter of the z-th segment of the blood vessel; Z is the total number of blood vessels; R min is the minimum curvature radius of the blood vessel; t b is the predicted branch point coordinate; is the prior coordinate of the branch point; B is the set of branch points;

[0039] Input the 4D input tensor into the initial network model for learning, calculate the loss function, and use the gradient coordination mechanism to optimize and iterate the parameters of the initial network model according to the loss function to obtain the trained portal vein state detection model.

[0040] Preferably, collect the CBCTA images to be analyzed of the abdomen of pre-screened patients, including:

[0041] Select a cone-beam CT angiography device with a detector size less than 0.3 mm, and set the scanning parameters for the cone-beam CT angiography device;

[0042] Instruct the pre-screened patients to take a negative contrast agent and wear a vacuum negative pressure pad;

[0043] Use a double-barrel high-pressure syringe to inject non-ionic iodine agent and normal saline into the pre-screened patients in sequence and evenly;

[0044] During the injection of the pre-screened patients, use the cone-beam CT angiography device to scan the abdomen of the pre-screened patients to obtain the original scan images;

[0045] Perform contrast agent artifact filtering and iterative reconstruction on the original scan images to obtain the CBCTA images to be analyzed with a resolution error less than 5%.

[0046] Preferably, generate a three-dimensional model of the portal vein according to the CBCTA images to be analyzed, including:

[0047] Extract the DICOM sequence from the CBCTA images to be analyzed, and use the bilinear interpolation algorithm to unify the voxel size of the DICOM sequence;

[0048] Use histogram equalization to adjust the window width of the DICOM sequence within a preset window width range, and use non-local means denoising to remove the noise of the DICOM sequence to obtain a preprocessed sequence;

[0049] Segment the preprocessed sequence through a 3D region growing algorithm and a 3D U-Net model to obtain a portal vein voxel probability map;

[0050] Use the centerline tracking algorithm to repair the broken branches of the portal vein voxel probability map to obtain an optimized probability map;

[0051] Generate a triangular mesh according to the optimized probability map using the Marching Cubes algorithm to obtain an original three-dimensional model, and perform sawtooth smoothing elimination, vertex clustering simplification, and normal consistency check on the original three-dimensional model to obtain an optimized three-dimensional model;

[0052] Perform boundary layer setting and volume mesh generation on the optimized three-dimensional model to obtain the three-dimensional model of the portal vein.

[0053] Preferably, perform CFD simulation settings on the three-dimensional model of the portal vein to obtain a hydrodynamic model of the portal vein, including:

[0054] Perform fluid property settings and turbulence model selection on the three-dimensional model of the portal vein to obtain the three-dimensional model of the portal vein with determined properties;

[0055] Perform inlet condition setting, outlet condition setting, and wall condition setting on the three-dimensional model of the portal vein with determined properties to obtain the three-dimensional model of the portal vein with determined boundary conditions;

[0056] Perform numerical format configuration, convergence criterion setting, and parallel computing optimization on the three-dimensional model of the portal vein with determined boundary conditions to obtain the hydrodynamic model of the portal vein.

[0057] Preferably, the block hydrodynamic sub-model includes: a gastric coronary vein hydrodynamic sub-model, a splenic vein hydrodynamic sub-model, and an intrahepatic portocaval vein hydrodynamic sub-model.

[0058] Preferably, use the actual portal vein pressure data and the pressure field distribution data to correct each block hydrodynamic sub-model to obtain a corrected block hydrodynamic correction model, including:

[0059] Set the value ranges of the inlet flow rate, outlet pressure, vascular wall elastic modulus, blood viscosity, and blood density to obtain limiting conditions;

[0060] Construct an error function and a regularization function based on the mean square error, and perform weighted fusion on the error function and the regularization function to obtain an optimization objective;

[0061] Formulate several simulation parameter schemes according to the limiting conditions, and perform simulations on each block hydrodynamic sub-model according to the simulation parameter schemes to obtain temporary simulation data;

[0062] Calculate the optimization objective based on the temporary simulation data and the actual portal vein pressure data to obtain optimization degree data;

[0063] Update the simulation parameter scheme corresponding to the smallest optimization degree data to the block hydrodynamic sub-model to obtain a corrected block hydrodynamic correction model.

[0064] Preferably, the working conditions include: the number of occluded vascular accesses, the occlusion ratio, the vascular diameter size, and the shunt channel diameter size; the occlusion ratios include: 40%, 50%, 60%, and 70%.

[0065] Preferably, the preset window width range is 200 to 400 HU.

[0066] Preferably, the error function is:

[0067]

[0068] where E(θ) is the output value of the error function; w L is the L-th error weight; P sin,L (θ) is the simulated pressure value of the L-th measurement point under the parameter θ; P real,L is the true pressure value of the L-th measurement point; N is the total number of pressure measurement points;

[0069] The regularization function is:

[0070] R(θ) = ∥θ Geo - θ orig ∥ 2 + ∥θ BC - θ BC,orig ∥ 2 ;

[0071] where R(θ) is the output value of the regularization function; θ Geo represents the set of geometric parameters; θ orig represents the original value of the geometric parameters; θ BC is the set of boundary condition parameters; θ BC,orig represents the original value of the boundary condition parameters.

[0072] The present invention discloses the following technical effects:

[0073] The present invention provides a method for detecting the portal vein state based on portal vein pressure redistribution. By using digital twins to construct a three-dimensional portal vein model, simulating the model under multiple working conditions, and determining the treatment plan according to the simulation results, the defect that the detection effect of the existing detection technology varies greatly under different working conditions is solved, and reliable detection of the portal vein under multiple working conditions is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0075] Figure 1 It is a schematic diagram of the detection process of the portal vein state based on portal vein pressure redistribution provided by the embodiment of the present invention;

[0076] Figure 2 Schematic diagram of the training process of the portal vein status detection model provided by the embodiment of the present invention;

[0077] Figure 3 Schematic diagram of the acquisition process of the CBCTA image to be analyzed provided by the embodiment of the present invention. Detailed implementation manners

[0078] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0079] The purpose of the present invention is to provide a method for detecting the portal vein status based on the redistribution of portal vein pressure, so as to realize the reliable detection of the portal vein under multiple working conditions.

[0080] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0081] Figure 1 Schematic diagram of the portal vein status detection process based on the redistribution of portal vein pressure provided by the embodiment of the present invention. As Figure 1 shown, the present invention provides a method for detecting the portal vein status based on the redistribution of portal vein pressure, including:

[0082] Step 100: Collect the CBCTA image to be detected of the abdomen of the patient to be detected;

[0083] Step 200: Input the CBCTA image to be detected into the portal vein status detection model to obtain the portal vein pressure detection result including the portal vein occlusion status;

[0084] Step 300: Match the portal vein pressure detection result with the preset treatment plan library to obtain the guiding treatment plan;

[0085] The construction process of the portal vein status detection model includes:

[0086] Step 201: Collect the CBCTA image to be analyzed of the abdomen of the pre-screened patient;

[0087] Step 202: Generate a three-dimensional model of the portal vein according to the CBCTA image to be analyzed;

[0088] Step 203: Perform CFD simulation settings on the three-dimensional model of the portal vein to obtain the portal vein hydrodynamic model;

[0089] Step 204: Perform block segmentation on the portal vein hydrodynamic model and create a hydrodynamic sub-model for the portacaval shunt channel to obtain multiple block hydrodynamic sub-models;

[0090] Step 205: Output the pressure simulation data of each block hydrodynamic sub-model to obtain pressure field distribution data;

[0091] Step 206: Collect intraoperative portal vein pressure measurement data of pre-screened patients to obtain actual portal vein pressure data;

[0092] Step 207: Use the actual portal vein pressure data and the pressure field distribution data to correct each block hydrodynamic sub-model to obtain a corrected block hydrodynamic correction model;

[0093] Step 208: Set the working condition status for each block hydrodynamic correction model according to the preset occlusion and shunt schemes, and collect the pressure simulation data of the block hydrodynamic correction model under different working condition statuses to obtain multi-condition simulation data with working condition labels;

[0094] Step 209: Input the CBCTA image to be analyzed and the multi-condition simulation data into the pre-constructed ConvNeXt model for training to obtain a trained portal vein status detection model.

[0095] Reference Figure 2 , input the CBCTA image to be analyzed and the multi-condition simulation data into the pre-constructed ConvNeXt model for training to obtain a trained portal vein status detection model, including:

[0096] Step 20901: Use the N5 bias field correction algorithm with hemodynamic priors to eliminate artifacts from the CBCTA image to be analyzed to obtain a preprocessed CBCTA image; the bias field estimation expression of the N5 bias field correction algorithm is:

[0097]

[0098] where B(x) is the bias field estimate; λ0 is the initial regularization coefficient; V(x) is the observed image intensity value; T(x) is the hemodynamic prior term; x is the original coordinate point;

[0099] Step 20902: Resample the multi-condition simulation data to obtain resampled data with the same resolution as the preprocessed CBCTA image;

[0100] Step 20903: Use the least squares method to match and perform local deformation compensation on the preprocessed CBCTA image and the resampled data to obtain a 4D input tensor; the expression for local deformation compensation is:

[0101]

[0102] Among them, is the image intensity value; A is the affine transformation matrix; b is the translation vector; w i is the weight coefficient of the i-th control point; is the radial basis function; x i is the i-th control point; n is the predefined number of control points;

[0103] Step 20904 designs the ConvNeXt architecture according to the 4-stage structure; the number of channels of the 4-stage structure are 96, 192, 384, and 768 respectively;

[0104] Step 20905 sets a cross-modal attention gate between the third and fourth layers of the ConvNeXt architecture to obtain the initial network model;

[0105] Step 20906 determines the loss function including the working condition classification loss, the pressure regression loss, and the anatomical constraint loss; the expression of the working condition classification loss is:

[0106]

[0107] Among them, is the output value of the working condition classification loss; w c is the weight of the c-th category; C is the total number of working condition categories; τ is the temperature coefficient; is the output score of the model for the input on the c-th category; λ ortho is the working condition orthogonal regularization coefficient; F is the feature matrix extracted by the network; I is the identity matrix with the same dimension as F; E[·] is the expectation operator;

[0108] The expression of the pressure regression loss is:

[0109]

[0110] Among them, δ = 0.1 mmHg; is the output value of the pressure regression loss; α l is the weight coefficient of the l-th layer; is the piecewise function; a is the piecewise variable; δ is the preset threshold; p l is the simulated pressure; is the true pressure; λ mass is the pressure regularization coefficient; ρ is the density; u is the velocity; λ dir is the direction consistency regularization coefficient;

[0111] The expression of the anatomical constraint loss is:

[0112]

[0113] Among them, g3 = max(0, 8 - R min ); is the output value of the anatomical constraint loss; λ topo is the topological constraint regularization coefficient; is the persistent homology distance; λ geo is the geometric constraint regularization coefficient; g z is the z-th blood vessel branch angle constraint; λ wasser is the distance regularization coefficient; W1(P seg , P prior ) is the first-order Wasserstein distance between the segmentation result distribution P seg and the prior distribution P prior ; Dgm h (f) is the h-th persistent homology diagram of the model f; Dgm h (S) is the h-th persistent homology diagram of the true anatomical structure S; σ is the smoothing coefficient; μ is the integration measure; θ e is the e-th blood vessel branch angle; is the blood vessel branch edge set; d z is the diameter of the z-th section of blood vessel; Z is the total number of blood vessels; R min is the minimum curvature radius of the blood vessel; t b is the predicted branch point coordinate; is the prior coordinate of the branch point; B is the branch point set;

[0114] Step 20907 inputs the 4D input tensor into the initial network model for learning, calculates the loss function, and uses the gradient coordination mechanism to optimize and iterate the parameters of the initial network model to obtain the trained portal vein state detection model.

[0115] Reference Figure 3 , collect the CBCTA images to be analyzed of the abdomen of pre-screened patients, including:

[0116] Step 20101: Select a cone beam CT angiography device with a detector size less than 0.3 mm and set the scanning parameters for the cone beam CT angiography device;

[0117] Step 20102: Instruct the pre-screened patients to take a negative contrast agent and wear a vacuum negative pressure pad;

[0118] Step 20103: Use a double-barrel high-pressure syringe to uniformly inject non-ionic iodine agent and normal saline into the pre-screened patients in sequence;

[0119] Step 20104: During the injection of the pre-screened patients, use a cone-beam CT angiography device to scan the abdomen of the pre-screened patients to obtain the original scan images;

[0120] Step 20105: Perform contrast agent artifact filtering and iterative reconstruction on the original scan images to obtain the CBCTA images to be analyzed with a resolution error less than 5%.

[0121] Specifically, generating a three-dimensional model of the portal vein based on the CBCTA images to be analyzed includes:

[0122] Extract the DICOM sequence from the CBCTA images to be analyzed, and use the bilinear interpolation algorithm to unify the voxel size of the DICOM sequence;

[0123] Use histogram equalization to adjust the window width of the DICOM sequence within a preset window width range, and use non-local means denoising to remove the noise of the DICOM sequence to obtain a preprocessed sequence;

[0124] Segment the preprocessed sequence through the 3D region growing algorithm and the 3D U-Net model to obtain the voxel probability map of the portal vein;

[0125] Use the centerline tracking algorithm to repair the broken branches of the voxel probability map of the portal vein to obtain an optimized probability map;

[0126] Generate a triangular mesh according to the optimized probability map using the Marching Cubes algorithm to obtain the original three-dimensional model, and perform sawtooth smoothing elimination, vertex clustering simplification, and normal consistency check on the original three-dimensional model to obtain an optimized three-dimensional model;

[0127] Perform boundary layer setting and volume mesh division on the optimized three-dimensional model to obtain a three-dimensional model of the portal vein.

[0128] Furthermore, perform CFD simulation settings on the three-dimensional model of the portal vein to obtain a hydrodynamic model of the portal vein, including:

[0129] Perform fluid property setting and turbulence model selection on the three-dimensional model of the portal vein to obtain a three-dimensional model of the portal vein with determined properties;

[0130] Perform inlet condition setting, outlet condition setting, and wall condition setting on the three-dimensional model of the portal vein with determined properties to obtain a three-dimensional model of the portal vein with determined boundary conditions;

[0131] Perform numerical format configuration, convergence criterion setting, and parallel computing optimization on the three-dimensional model of the portal vein with determined boundary conditions to obtain a hydrodynamic model of the portal vein.

[0132] Preferably, the block hydrodynamic sub-model includes: a gastric coronary vein hydrodynamic sub-model, a splenic vein hydrodynamic sub-model, and an intrahepatic portacaval vein hydrodynamic sub-model.

[0133] Furthermore, the actual data of the portal vein pressure and the pressure field distribution data are used to correct each block hydrodynamic sub-model, and a corrected block hydrodynamic correction model is obtained, including:

[0134] Set the value ranges of the inlet flow rate, outlet pressure, blood vessel wall elastic modulus, blood viscosity, and blood density to obtain limiting conditions;

[0135] Construct an error function and a regularization function according to the mean square error, and perform weighted fusion on the error function and the regularization function to obtain an optimization objective; the weight ratio of the error function and the regularization function is set to 3:1;

[0136] Formulate several simulation parameter schemes according to the limiting conditions, and perform simulations on each block hydrodynamic sub-model according to the simulation parameter schemes to obtain temporary simulation data;

[0137] Calculate the optimization objective according to the temporary simulation data and the actual data of the portal vein pressure to obtain optimization degree data;

[0138] Update the simulation parameter scheme corresponding to the minimum optimization degree data to the block hydrodynamic sub-model to obtain a corrected block hydrodynamic correction model.

[0139] Optionally, the working condition states include: the number of occluded blood vessel passages, the occlusion ratio, the blood vessel diameter size, and the shunt channel diameter size; the occlusion ratios include: 40%, 50%, 60%, and 70%.

[0140] Preferably, the preset window width range is from 200 to 400 HU.

[0141] Specifically, the error function is:

[0142]

[0143] where E(θ) is the output value of the error function; w L is the error weight of the L-th; P sin,L (θ) is the simulated pressure value of the L-th measurement point under the parameter θ; P real,L is the true pressure value of the L-th measurement point; N is the total number of pressure measurement points;

[0144] The regularization function is:

[0145] R(θ) = ∥θ Geo -θ orig ∥ 2+∥θ BC -θ BC,orig ∥ 2 ;

[0146] where R(θ) is the output value of the regularization function; θ Geo represents the set of geometric parameters; θ orig represents the original value of the geometric parameters; θ BC is the set of boundary condition parameters; θ BC,orig represents the original value of the boundary condition parameters.

[0147] Preferably, a tubular shunt channel is established between the vein and the intrahepatic portal vein, namely the hydrodynamic sub-model of the portacaval shunt, which is used to evaluate the effect of pressure reduction.

[0148] Specifically, the 4-stage structure adopted in this embodiment includes:

[0149] Block1 (number of channels 96): depthwise separable convolution (Kernel = 7×7×7), LayerScale coefficient = 0.000001;

[0150] Block2 (number of channels 192): introducing an inverted bottleneck structure (ExpansionRatio = 4);

[0151] Block3 (number of channels 384): adding a Squeeze-and-Excitation attention module;

[0152] Block4 (number of channels 768): using a GroupNormalization module to replace the traditional LayerNormalization module.

[0153] Specifically, the scanning parameters of the cone-beam CT angiography (CBCTA) device include: slice thickness: 0.625 - 1.25 mm; scanning time window: coordinated with the peak time of the contrast agent; rotation speed: 0.5 seconds / rotation; tube voltage: 80 - 120 kV; tube current: 150 - 300 mA.

[0154] Preferably, some pre-treatments are also required before collecting CT images of the patient, such as: supine position, raising both arms to ensure the use effect of the vacuum negative pressure pad; guiding the patient to perform breath-holding training to reduce the generation of noise during shooting. In addition, patients with a history of iodine contrast agent allergy are pre-treated with glucocorticoids in advance.

[0155] Specifically, during the device scanning process, the ROI (Region of Interest) area will be automatically detected, and the diffusion trajectory of the contrast agent will be automatically detected. The triggering threshold during this period is 150 HU. The upper and lower limits of the device scanning are set to 2 cm above the diaphragm top (upper limit) and the iliac crest horizontal position (lower limit), respectively.

[0156] Furthermore, the DICOM sequence used in this embodiment is matrix data of 512×512, and the voxel size is 0.5 mm. 3 . Before segmentation, basic noise reduction can also be performed through non-local means denoising to improve data quality.

[0157] Optionally, after three-dimensional reconstruction, centerline extraction, branch annotation, and blood flow connectivity verification are performed on the blood vessel distribution. The centerline is used for calculating parameters such as branch angle and curvature radius; blood flow connectivity verification is used to ensure the reliability of the model.

[0158] Preferably, boundary layer settings: Generate 5 layers of prism boundary layers (the thickness of the first layer is 0.1 mm, and the growth rate is 1.2), and the near-wall y+ value is controlled within 1 - 5 (meeting the requirements of the low Reynolds number turbulence model). Volume mesh division: Use unstructured tetrahedral meshes (Tetrahedral), the maximum element size: 3 mm in the main trunk area, 1 mm at the branch ends, and the total number of meshes is approximately 5 million.

[0159] Furthermore, some model parameters of this embodiment are set as follows: The blood density is 1060 kg / m 3 ; The viscosity model adopts the Carreau non-Newtonian model; The turbulence model selects the k-ω SST model: Inlet turbulence intensity: 5%, specific dissipation rate: 0.1.

[0160] Optionally, the model simulation convergence conditions of this embodiment are that the residual of the continuity equation is less than 0.0001 and the residual of the momentum equation is less than 0.000001; Under-relaxation factor: 0.3 for pressure, 0.7 for momentum.

[0161] Preferably, the inlet conditions include the velocity profile and turbulence parameters (set to 5% as before); Velocity profile: Based on the patient's portal vein blood flow rate, a parabolic velocity distribution is set; The outlet conditions include pressure outlet and reverse flow suppression (stabilization of outlet reflux); The reference pressure of the pressure outlet is set to the average portal vein pressure (10 - 15 mmHg), and it is dynamically adjusted through intraoperative pressure measurement data. The wall conditions include the blood vessel wall elastic modulus (calibrated according to intraoperative ultrasound elastography data) and no-slip boundary (the roughness height is set to 0.01 mm).

[0162] Preferably, in this embodiment, the surgical plan is optimized according to the portal vein pressure, and different working conditions are combined to detect the portal vein pressure. The simplest interventional surgical treatment plan is preferably selected in this embodiment to effectively reduce the portal vein pressure.

[0163] The beneficial effects of the present invention are as follows:

[0164] By using digital twins to construct a three-dimensional portal vein model, simulating the model under multiple working conditions, and determining the treatment plan according to the simulation results, the present invention improves the detection effect of the portal vein and realizes the determination of the optimal treatment plan.

[0165] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0166] Specific examples are used in this article to elaborate on the principles and implementation methods of the present invention. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A portal vein status detection method based on portal vein pressure redistribution, characterized in that: include: Acquire the CBCTA image of the abdomen of the patient to be tested; Inputting the CBCTA image to be detected into a portal vein status detection model to obtain a portal vein pressure measurement result including a portal vein obstruction status; Matching the portal vein pressure measurement result with a preset treatment plan library to obtain a guiding treatment plan; The construction process of the portal vein status detection model includes: Collect CBCTA images of the abdomen of pre-screened patients to be analyzed; generating a three-dimensional model of the portal vein according to the CBCTA image to be analyzed; Performing CFD simulation settings on the portal vein three-dimensional model to obtain a portal vein fluid dynamics model; The portal vein fluid dynamics model is segmented into blocks and a portal-caval shunt fluid dynamics sub-model is newly created to obtain a plurality of block fluid dynamics sub-models; Outputting the pressure simulation data of each block fluid dynamic sub-model to obtain pressure field distribution data; Collecting intraoperative portal vein pressure measurement data of the pre-screened patients to obtain actual portal vein pressure data; Using the actual portal vein pressure data and the pressure field distribution data, each block fluid dynamics sub-model is corrected to obtain a corrected block fluid dynamics correction model; According to the preset blocking and diversion scheme, the working condition state of each block fluid dynamic correction model is set, and the pressure simulation data of the block fluid dynamic correction model under different working conditions are collected to obtain multi-working condition simulation data with working condition labels; The CBCTA image to be analyzed and the multi-condition simulation data are input into a pre-built ConvNeXt model for training to obtain a trained portal vein status detection model.

2. A portal vein status detection method based on portal vein pressure redistribution according to claim 1, characterized in that: The CBCTA image to be analyzed and the multi-condition simulation data are input into a pre-built ConvNeXt model for training to obtain a trained portal vein status detection model, including: The N5 bias field correction algorithm with hemodynamic priors is used to eliminate artifacts from the CBCTA image to be analyzed, thereby obtaining a preprocessed CBCTA image. The bias field estimation expression of the N5 bias field correction algorithm is: Wherein, B(x) is the bias field estimation value; λ0 is the initial regularization coefficient; V(x) is the observed image intensity value; T(x) is the hemodynamic prior; x is the original coordinate point; Resampling the multi-operating condition simulation data to obtain resampled data with the same resolution as the preprocessed CBCTA image; The pre-processed CBCTA image and the resampled data are matched and locally deformed by using the least squares method to obtain a 4D input tensor; the expression for the local deformation compensation is: in, is the image intensity value; A is the affine transformation matrix; b is the translation vector; w i is the weight coefficient of the i-th control point; is the radial basis function; x i is the i-th control point; n is the number of predefined control points; Design the ConvNeXt architecture based on a 4-stage structure; the number of channels of the 4-stage structure are 96, 192, 384 and 768 respectively; A cross-modal attention gate is set between the third and fourth network layers of the ConvNeXt architecture to obtain an initial network model; Determine a loss function including working condition classification loss, pressure regression loss and anatomical constraint loss; the expression of the working condition classification loss is: in, is the output value of the classification loss of the working condition; w c is the weight of the cth category; C is the total number of working condition categories; τ is the temperature coefficient; Input for the model Output score on category c; λ ortho is the orthogonal regularization coefficient of the working condition; F is the feature matrix extracted by the network; I is the unit matrix of the same dimension as F; E[·] is the expected value operator; The expression of the pressure regression loss is: in, δ=0.1mmH; is the output value of the pressure regression loss; α l is the weight coefficient of the lth layer; is a piecewise function; a is a piecewise variable; δ is a preset threshold; p l To simulate pressure; is the real pressure; mass is the pressure regularization coefficient; ρ is the density; u is the velocity; λ dir is the direction consistency regularization coefficient; The expression of the anatomical constraint loss is: in, g3=max(0,8-R min ); is the output value of the anatomical constraint loss; topo is the topological constraint regularization coefficient; is the persistent homology distance; λ geo is the geometric constraint regularization coefficient; g z is the z-th blood vessel branch angle constraint; wasser is the distance regularization coefficient; W1(P seg ,P prior ) is the segmentation result distribution P seg and the prior distribution P prior The first-order Wasserstein distance of h (f) is the h-th permanent homology graph of model f; Dgm h (S) is the h-th permanent homology graph of the real anatomical structure S; σ is the smoothing coefficient; μ is the integral measure; θ e is the e-th blood vessel branch angle; is the set of blood vessel branch edges; d z is the diameter of the zth segment of blood vessels; Z is the total number of blood vessels; R min is the minimum curvature radius of the blood vessel; t b is the predicted branch point coordinates; is the prior coordinate of the branch point; B is the set of branch points; The 4D input tensor is input into the initial network model for learning, the loss function is calculated, and the parameters of the initial network model are optimized and iterated according to the loss function using a gradient coordination mechanism to obtain the trained portal vein status detection model.

3. The method for detecting portal vein status based on portal vein pressure redistribution according to claim 1, characterized in that: Acquire CBCTA images of the abdomen of pre-screened patients for analysis, including: Selecting a cone beam CT angiography device having a detector size of less than 0.3 mm, and setting scanning parameters for the cone beam CT angiography device; Instructing the pre-screened patient to take a negative contrast agent and wear a vacuum negative pressure pad; Using a double-barrel high-pressure syringe, uniformly injecting non-ionic iodine and normal saline into the pre-screened patient in sequence; During the injection of the pre-screened patient, scanning the abdomen of the pre-screened patient using the cone beam CT angiography device to obtain an original scan image; The original scanned image is subjected to contrast agent artifact filtering and iterative reconstruction to obtain the CBCTA image to be analyzed with a resolution error of less than 5%.

4. The method for detecting portal vein status based on portal vein pressure redistribution according to claim 1, characterized in that: Generating a portal vein three-dimensional model according to the CBCTA image to be analyzed includes: Extracting the DICOM sequence in the CBCTA image to be analyzed, and unifying the voxel size of the DICOM sequence using a bilinear interpolation algorithm; The window width of the DICOM sequence is adjusted to a preset window width range by using histogram equalization, and the DICOM sequence is de-noised by using non-local mean denoising to obtain a pre-processed sequence; Segmenting the preprocessed sequence by a 3D region growing algorithm and a 3D U-Net model to obtain a portal vein voxel probability map; Repairing broken branches of the portal vein voxel probability map using a centerline tracking algorithm to obtain an optimized probability map; Generating a triangular mesh using the Marching Cubes algorithm according to the optimized probability graph to obtain an original three-dimensional model, and performing anti-aliasing, vertex clustering simplification, and normal consistency check on the original three-dimensional model to obtain an optimized three-dimensional model; The optimized three-dimensional model is subjected to boundary layer setting and volume mesh division to obtain the portal vein three-dimensional model.

5. The method for detecting portal vein status based on portal vein pressure redistribution according to claim 1, characterized in that: The portal vein three-dimensional model is subjected to CFD simulation setting to obtain a portal vein fluid dynamic model, including: Setting fluid properties and selecting a turbulence model for the three-dimensional portal vein model to obtain the three-dimensional portal vein model with determined properties; Setting inlet conditions, outlet conditions and wall conditions for the portal vein three-dimensional model with determined attributes to obtain the portal vein three-dimensional model with determined boundary conditions; The portal vein three-dimensional model determined by boundary conditions is subjected to numerical format configuration, convergence standard setting and parallel calculation optimization to obtain the portal vein fluid dynamics model.

6. The method for detecting portal vein status based on portal vein pressure redistribution according to claim 1, characterized in that: The block fluid dynamics sub-model includes: a gastric coronary vein fluid dynamics sub-model, a splenic vein fluid dynamics sub-model and an intrahepatic portal vein fluid dynamics sub-model.

7. The method for detecting portal vein status based on portal vein pressure redistribution according to claim 1, characterized in that: Using the actual portal vein pressure data and the pressure field distribution data to correct each block fluid dynamics sub-model, a corrected block fluid dynamics correction model is obtained, including: The value ranges of inlet flow rate, outlet pressure, vascular wall elastic modulus, blood viscosity, and blood density are set to obtain limiting conditions; An error function and a regularization function are constructed according to the mean square error, and the error function and the regularization function are weightedly fused to obtain an optimization target; Formulate a number of simulation parameter schemes according to the limiting conditions, and simulate each of the block fluid dynamic sub-models according to the simulation parameter schemes to obtain temporary simulation data; Calculating the optimization target according to the temporary simulation data and the actual portal vein pressure data to obtain optimization degree data; The simulation parameter scheme corresponding to the minimum optimization degree data is updated to the block fluid dynamics sub-model to obtain a corrected block fluid dynamics correction model.

8. The method for detecting portal vein status based on portal vein pressure redistribution according to claim 1, characterized in that: The working conditions include: the number of blocked vascular pathways, the occlusion ratio, the vascular caliber size, and the shunt caliber size; the occlusion ratio includes: 40%, 50%, 60%, and 70%.

9. The method for detecting portal vein status based on portal vein pressure redistribution according to claim 4, characterized in that: The preset window width ranges from 200 to 400 HU.

10. The method for detecting portal vein status based on portal vein pressure redistribution according to claim 7, characterized in that: The error function is: Wherein, E(θ) is the output value of the error function; w L is the Lth error weight; P sin,L (θ is the simulated pressure value of the Lth measurement point under parameter θ; P real,L is the actual pressure value of the Lth measuring point; N is the total number of pressure measuring points; The regularization function is: R(θ)=∥θ Geo -θ orig ∥ 2 +∥θ BC -θ BC,orig ∥ 2 ; Wherein, R(θ) is the output value of the regularization function; θ Geo represents the set of geometric parameters; θ orig Represents the original value of the geometric parameters; θ BC is the boundary condition parameter set; θ BC,orig Represents the original value of the boundary condition parameter.

Images