Causal device and causal method thereof

Abstract

The invention discloses a causal device and a causal method thereof. The causal device includes a causal module and a causal feature learning module coupled to the causal module. The causal module is used for identifying or utilizing a causal relationship among a plurality of variables; the causal feature learning module is used for extracting at least one causal feature of one of the plurality of variables. Therefore, the accurate fusion of combined and mixed imaging can be ensured or the imaging examination priority triage can be improved.

Description

Technical Field

[0001] The present invention relates to a causal device and a causal method thereof, and more particularly to a causal device and a causal method thereof that can ensure precise fusion of hybrid imaging or improve the priority triage of imaging examinations. Background Art

[0002] Hybrid Imaging refers to combining more than two imaging modalities in a single imaging process, such as PET / MRI (Positron Emission Tomography / Magnetic Resonance Imaging) or PET / CT (Positron Emission Tomography / Computed Tomography), to obtain complementary information about the anatomical structure and function of the imaged tissue or organ. The composite image can provide more accurate and comprehensive information than any single imaging modality alone. One of the main challenges of PET / MRI imaging is accurate attenuation correction because MRI (Magnetic Resonance Imaging) images do not provide direct information about photon attenuation. The attenuation map of PET / MRI is usually generated by combining multiple methods. However, these methods are prone to errors, especially in regions with high tissue heterogeneity or metal implants. Another challenge of PET / MRI imaging is to correct the motion and registration errors between the PET (Positron Emission Tomography) image and the MRI image. PET images and MRI images are usually acquired separately and then registered with each other, but differences in imaging geometry and physiological state may introduce errors and misalignments. In addition, another challenge of PET / MRI images is that they are affected by various artifacts, such as radiofrequency interference, sensitivity, and chemical shift artifacts. Although PET / MRI imaging can improve soft tissue contrast and reduce radiation exposure compared to PET / CT imaging, there is still a need to improve the accuracy and reliability of the image data.

[0003] In addition, priority triage is the process of prioritizing patients based on the severity of their medical conditions and the urgency of their medical needs. For example, the urgency of further diagnostic imaging (such as a CT scan) for a patient is determined based on the results of previous imaging studies (such as an X-ray). However, existing priority triage methods cannot accurately predict which patients require immediate attention, resulting in treatment delays for some patients and unnecessary interventions for others. Moreover, existing priority triage methods are based on fixed rules or algorithms and cannot adapt to different patients or clinical environments, leading to poor efficiency. Existing priority triage methods usually rely on limited clinical features or imaging modalities and cannot fully capture the complexity of a patient's condition. When existing priority triage methods are used to determine the use of limited resources (such as imaging equipment or intensive care beds), if the algorithm design or validation is improper, bias may be introduced, raising ethical issues. Existing priority triage methods require the opinions of radiologists or other medical professionals to interpret imaging results and make decisions regarding patient prioritization, which is very time-consuming and resource-intensive. Summary of the Invention

[0004] Therefore, the present invention mainly provides a causal device and a causal method to improve the deficiencies of the prior art.

[0005] An embodiment of the present invention discloses a causal device, including a causal module for identifying or exploiting causal relationships between a plurality of variables; and a causal feature learning module coupled to the causal module for extracting at least one causal feature of one of the plurality of variables.

[0006] An embodiment of the present invention discloses a causal method for a causal device, including identifying or exploiting causal relationships between a plurality of variables; and extracting at least one causal feature of one of the plurality of variables. Brief Description of the Drawings

[0007] Figure 1 It is a schematic diagram of an image reconstruction device according to Embodiment 1 of the present invention;

[0008] Figure 2 It is a schematic diagram of an image reconstruction method according to Embodiment 1 of the present invention;

[0009] Figure 3 It is a schematic diagram of a causal graph corresponding to a structural causal model according to an embodiment of the present invention;

[0010] Figure 4 It is a schematic diagram of a causal graph corresponding to a continuous-time structural causal model according to an embodiment of the present invention;

[0011] Figure 5 It is a schematic diagram of a causal feature learning module according to an embodiment of the present invention;

[0012] Figure 6 Schematic diagram of the reconstruction module in the first embodiment of the present invention;

[0013] Figure 7 Schematic diagram of the priority triage device in the first embodiment of the present invention;

[0014] Figure 8 Schematic diagram of the priority triage method in the first embodiment of the present invention;

[0015] Figure 9 Schematic diagram of the causal model in the first embodiment of the present invention;

[0016] Figure 10 Schematic diagram of the causal device in the first embodiment of the present invention.

[0017] Symbol description

[0018] 10: Image reconstruction device

[0019] 10CG: Causal relationship

[0020] 11: Causal device

[0021] 1142: Causal module

[0022] 120: Preprocessing module

[0023] 140: Extraction module

[0024] 142: Causal reasoning module

[0025] 144,544,744,1144: Causal feature learning module

[0026] 160,660: Reconstruction module

[0027] 20: Image reconstruction method

[0028] 30CG,40CG: Causal graph

[0029] 544C,544C1,544C2: Clustering module

[0030] 544D: Density estimation module

[0031] 5X,5Y: Data

[0032] 660D: Discriminator network

[0033] 660G: Generator network

[0034] 70: Priority triage device

[0035] 740: Establishment module

[0036] 742: Causal model establishment module

[0037] 780: Decision analysis module

[0038] 790: Judgment module

[0039] 90: Continuous-time structural causal model

[0040] CF1~CFr, CF, cf, cf1~cfy: Causal features

[0041] CL1~CLz: Confidence levels

[0042] CTMCDA: Continuous-time multi-criteria decision analysis

[0043] DCPGt0~DCPGt3: Dynamic causal planning graph

[0044] DT: Input data

[0045] FD: Feedback

[0046] IMG: Image

[0047] IV1~IVq: Input variables

[0048] IV11~IV5: Input variables

[0049] ivd, IVD: Input variable data

[0050] OV: Output variable

[0051] rIMG: Reconstructed image

[0052] S200~S208, S800~S809: Steps

[0053] SV1~SVx: State variables

[0054] SV11~SV4p: State variables

[0055] t0~t3: Time points Detailed implementation manners

[0056] Figure 1 It is a schematic diagram of the image reconstruction device 10 according to the first embodiment of the present invention. The image reconstruction device 10 may include a preprocessing module 120, an extraction module 140, and a reconstruction module 160. The extraction module 140 may include a causal reasoning module 142 and a Causal Feature Learning (CFL) module 144.

[0057] The preprocessing module 120 can be used to perform necessary preprocessing on the input variable data ivd. The input variable data IVD obtained by preprocessing the input variable data ivd may include input variables IV1 to IVq, and the input variables IV1 to IVq may have different types and dimensions. In one embodiment, the input variable data ivd or IVD may include, for example, PET data (such as a PET sinogram or a PET image), MRI data (such as an MRI sequence or an MRI image), patient demographics, imaging protocols, or scanner characteristics. A PET sinogram is the raw data obtained from a PET scan and is used to create a PET image. An MRI sequence is a set of MRI images captured at different times during a scan.

[0058] The causal inference module 142 can be used to identify the causal relationship 10CG between the preprocessed input variables IV1 to IVq and the outcome variable OV. In one embodiment, the outcome variable OV may be a PET / MRI reconstructed image. The CFL module 144 can be used to extract causal features CF1 to CFr from the preprocessed input variables IV1 to IVq. The causal features CF1 to CFr may correspond to or include, for example, a part of an image or certain information, but are not limited thereto. The reconstruction module 160 can be used to generate a PET / MRI reconstructed image rIMG based on the causal features CF1 to CFr.

[0059] As can be seen from the above, the PET / MRI reconstructed image (such as rIMG) refers to a PET / MRI image generated by an image reconstruction algorithm using input variable data such as a PET sinogram and an MRI sequence. Among them, the PET sinogram and the MRI sequence can be registered or corrected. For example, the MRI image can provide anatomical information for the PET image through attenuation correction. In this case, at least the PET sinogram and the MRI sequence have a causal relationship with the PET / MRI reconstructed image, and the causal inference module 142 can learn or determine the causal relationship 10CG. The CFL module 144 can extract the causal features CF1 to CFr corresponding to the causal relationship 10CG, and the reconstruction module 160 generates the PET / MRI reconstructed image rIMG based on the causal features CF1 to CFr. In other words, the causal features CF1 to CFr imply the features that will affect or cause the outcome variable OV. Therefore, compared with the existing hybrid imaging PET / MRI images, the PET / MRI reconstructed image rIMG has higher accuracy.

[0060] Figure 2 FIG. 20 is a schematic diagram of an image reconstruction method 20 according to Embodiment 1 of the present invention. The image reconstruction method 20 is applicable to the image reconstruction apparatus 10 and may include the following steps:

[0061] Step S200: Start.

[0062] Step S202: Preprocess the input variable data (e.g., ivd), where the input variable data includes input variables (e.g., IV1 to IVq).

[0063] Step S204: Extract causal features (e.g., CF1 to CFr) from each of the preprocessed input variables.

[0064] Step S206: Perform image reconstruction based on the causal features to generate a PET / MRI reconstructed image (e.g., rIMG).

[0065] Step S208: End.

[0066] The image reconstruction method 20 will be described in detail below. In step S202, the image reconstruction device 10 can receive the input variable data, and the input variable data can be loaded into the memory of the image reconstruction device 10.

[0067] In one embodiment, in step S202, the preprocessing module 120 can perform necessary preprocessing on the input variable data (e.g., ivd) or the input variables (e.g., IV1 to IVq), such as attenuation correction, motion correction, registration, or normalization, standardization.

[0068] In one embodiment, the attenuation correction of the PET sinogram can adopt the transmission-based attenuation correction (TAC) method, which can improve accuracy or quantification without relying on external factors such as patient body size or body habitus, and the additional radiation exposure of the transmission scan is relatively low compared to the radiation exposure of the PET scan itself. The motion correction of the PET sinogram can adopt the motion-compensation reconstruction (MCR) method, which can ensure accurate motion correction for large motions and can be implemented with or without accessing motion tracking data. The normalization or standardization of the PET sinogram can correct variations in scanner sensitivity and other factors that may affect image quality, and for example, the standard uptake value (SUV) method can be used.

[0069] In one embodiment, the motion correction of the MRI sequence can adopt the non-rigid registration method to handle non-linear deformations.

[0070] In one embodiment, the registration of the PET sinogram or the MRI sequence can adopt the Gradient Correlation (GC) method, and can be robust to intensity differences or noise. The registration of the PET sinogram of the PET / MRI image and the MRI sequence refers to aligning the PET sinogram and the MRI sequence to the same coordinate space for joint analysis. The PET sinogram and the MRI sequence are obtained by using different imaging modalities, and the spatial resolution or image quality of the PET sinogram and the MRI sequence may be different. Therefore, registration is required to compensate for the differences, so as to accurately integrate the two imaging modalities to improve diagnosis and treatment planning.

[0071] In one embodiment, the normalization or standardization of patient demographic data (or demographic variables) can eliminate any confounding effects of demographic variables on image data. For example, the Robust scaling method can be used, which is robust to outliers to preserve the distribution of the data. Patient demographic data can include gender or age, etc.

[0072] In one embodiment, the normalization or standardization of the imaging protocol can adopt the method of Use of phantom studies, which can be used to verify and optimize the imaging protocol and continuously monitor the imaging quality.

[0073] In one embodiment, the preprocessing of scanner characteristics can be performed by the scanner manufacturer.

[0074] In one embodiment, in step S202, the image reconstruction device 10 (such as the preprocessing circuit 120) can also divide the preprocessed input variable data into a training set and a testing set.

[0075] In one embodiment, at least a part of the image reconstruction method 20 can be compiled into a pseudocode. For example, the corresponding step S202 may include:

[0076]

[0077] In one embodiment, in step S202, since each input variable data is collected at a specific time point, the input variable data may be discrete, but interpolation methods (linear or non-linear) can be used to convert the input variable data into continuous-time input variable data.

[0078] In step S204, the causal inference module 142 may utilize a causal inference algorithm to identify or determine the (potential) causal relationship (e.g., 10CG) between the preprocessed input variables (e.g., IV1 to IVq) and the outcome variable (e.g., OV).

[0079] In one embodiment, the causal inference algorithm may include, for example, a Continuous Time Structured Equation Modeling (CTSEM) framework. In one embodiment, the causal inference algorithm may be applied to a Continuous Time Structural Causal Model (CTSCM). In one embodiment, the CTSEM framework may be part of the CTSEM. In one embodiment, the CTSCM may be processed by software such as Python, Gephi, DAGitty or other software, and the CTSEM may be analyzed by, for example, LISREL, Knitr, OpenMx, Onyx, Stata or other software.

[0080] In one embodiment, the causal inference module 142 may define or generate a Causal graph in step S204 to define, present or describe the causal relationship between the preprocessed input variables and the outcome variable. In one embodiment, the Causal graph may be drawn using domain knowledge or previous research.

[0081] For example, Figure 3 FIG. 30CG is a schematic diagram of a Causal graph corresponding to the SCM according to an embodiment of the present invention, Figure 4 FIG. 40CG is a schematic diagram of a Causal graph corresponding to the CTSCM according to an embodiment of the present invention. In Figure 3 and Figure 4 , the preprocessed input variables IV11 to IV5 may be used to implement the preprocessed input variables IV1 to IVq. In one embodiment, the preprocessed input variables IV11 to IV1i may respectively correspond to or be different angles of the PET sinogram, the preprocessed input variables IV21 to IV2j may respectively correspond to or be T1-weighted or T2-weighted MRI sequences, and the preprocessed input variables IV3 to IV5 may respectively correspond to or be patient demographic data, imaging protocol and scanner characteristics. In other words, the Causal graph 30CG or 40CG may give the causal relationship (e.g., 10CG) between the input variables and the outcome variable OV.

[0082] In Figure 3 and Figure 4, the causal graph 30CG can describe the causal relationship between the preprocessed input variables IV11-IV5 and the outcome variable OV, and the causal graph 40CG can describe the causal relationship between the preprocessed input variables IV11-IV5 and the outcome variable OV at different time points t0-t3. From Figure 3 and Figure 4 it can be seen that the SCM or its causal graph 30CG only corresponds to a specific instant of the CTSCM or its causal graph 40CG, and the CTSCM can be composed of countless instants. In other words, the CTSCM can be regarded as including multiple SCMs, and the CTSCM can capture the causal relationship between the input variables and the outcome variable OV changing over time. Similarly, the SEM framework only analyzes a specific instant of the CTSEM framework.

[0083] Although the causal graph 40CG only describes the causal relationship between the preprocessed input variables IV11-IV5 and the outcome variable OV at a certain time point (such as t0), in one embodiment, the outcome variable OV at time point t1 may also be affected by the input variables at time point t0. The specific causal relationship depends on specific imaging protocols, patient characteristics, or other factors.

[0084] In one embodiment, at step S204, after determining the CTSCM, the preprocessed input variables (and the outcome variable) can be used to train the CTSCM. And during the training process, the CTSCM can learn the causal relationship between the preprocessed input variables and the PET / MRI reconstructed image, and be used for subsequent causal feature extraction (step S204) and image reconstruction (step S206), so as to improve the accuracy and robustness of image reconstruction.

[0085] In one embodiment, in step S204, the training of the CTSCM involves finding the model parameter values that best fit the preprocessed input variables (and result variables). The algorithms used to train the CTSCM may include Maximum Likelihood Estimation (MLE), Bayesian inference, Expectation-Maximization (EM), or other algorithms. MLE is a statistical method for finding the model parameter values that maximize the probability of the observed data (input variables or result variables). MLE is relatively easy to implement and can generally effectively find the parameters that produce good predictions. Bayesian inference is a statistical method for training the CTSCM by iteratively updating the model parameter values when new preprocessed input variables (and result variables) are collected. EM is an iterative algorithm for finding the model parameter values that maximize the likelihood of the observed data. EM can effectively handle incomplete or data including missing values. Which algorithm to use to train the CTSCM can be selected according to various factors (such as the size and quality of the available input variables or result variables, the specific characteristics of the model, and the available computing resources).

[0086] For example, the image reconstruction method 20 may further involve training the CTSCM using MLE, and may further include the following steps:

[0087] Step S500: Start.

[0088] Step S502: Initialize the model parameter values of the CTSCM. In one embodiment, the initial values of the model parameter values can be set by randomly generating the model parameter values or by adopting expertise. Then, step S502 is performed.

[0089] Step S504: Use the CTSCM to predict the predicted values of the variables (such as the result variable OV, the input variable IVq, the causal relationship 10CG, the probability, or the conditional probability). Then, step S506 is performed.

[0090] Step S506: Compare the predicted values of the variables with the actual values of the variables (such as the true (ground truth) PET / MRI image, the actual input variable IVq, the causal relationship, the probability, or the conditional probability). Then, step S508 is performed.

[0091] Step S508: Update the model parameter values to make the predicted values of the CTSCM closer to the actual values.

[0092] Step S510: Determine whether the model parameter values have changed significantly. If they have changed significantly, perform step S502; if not, perform step S512.

[0093] Step S512: End.

[0094] For example, in step S204, the image reconstruction device 10 obtains a set of PET sinograms and MRI sequences, as well as patient demographic data, imaging protocols, and scanner characteristics related to the PET sinograms and MRI sequences. The image reconstruction device 10 can use these preprocessed input variables (such as IV1 to IVq) to train the CTSCM to understand the causal relationship between the preprocessed input variables and the PET / MRI reconstructed images. Alternatively, for example, in step S204, the image reconstruction device 10 can use the CTSCM to learn the causal relationship between the PET sinograms and the MRI sequences. The PET sinograms can be used as preprocessed input variables, and the MRI sequences can be used as result variables. By training the CTSCM, the CTSCM can learn how the PET sinograms affect the MRI sequences, thereby understanding the correlation between the PET sinograms and the MRI sequences. It can be seen that the training of the CTSCM involves training using preprocessed input variables and result variables to understand the causal relationship between the preprocessed input variables and the result variables.

[0095] In one embodiment, the CTSCM may include time-related model parameter values, so the training of the CTSCM may require more input variable data to determine the function of the model parameter values with respect to time. In one embodiment, a linear regression model (such as θ(t) = α + βt, where θ(t) is the model parameter value at time point t and α is the initial model parameter value) or a non-linear regression model (such as θ(t) = f(t), where f(t) is a non-linear function) can be used to estimate the function of the model parameter values with respect to time according to whether the time-related model parameter values of the CTSCM are linear or non-linear.

[0096] In step S204, the CFL module 144 can extract causal features from each of the preprocessed input variables using the CFL algorithm. The causal features can be used to describe the causal relationship between the preprocessed input variables and the result variables. Generally, causal features are more robust to noise than ordinary features, especially when the noise affects the statistical laws of the data but does not affect the potential causal relationship between the input variables and the output variables. Causal features can be used to extract the potential causes leading to the observed data (such as input variables or output variables), rather than just the statistical laws of the data itself, and thus are more robust to noise.

[0097] In one embodiment, the CFL algorithm can be an unsupervised CFL algorithm or a CFL neural network. In one embodiment, the CFL algorithm can be a CFL neural network in the form of CTSCM. In other words, CTSCM can be a machine learning framework that uses an unsupervised CFL algorithm to extract causal features from preprocessed input variables. In fact, extracting causal features from input variables using CTSCM means identifying the potential causal relationships between preprocessed input variables and result variables, and the causal features obtained by CTSCM can point out the control mechanism for the relationship between preprocessed input variables and result variables, and can be used to improve the accuracy and robustness of image reconstruction.

[0098] In one embodiment, the CFL algorithm can minimize the mutual information between latent variables at different time steps while maximizing the mutual information between latent variables and observed variables to learn disentangled and causally related features. In one embodiment, the causal features extracted by the CFL algorithm can be used to cluster input variables and improve the accuracy of PET / MRI reconstructed images.

[0099] For example, Figure 5 FIG. is a schematic diagram of a CFL module 544 according to an embodiment of the present invention. The CFL module 544 can be used to implement the CFL module 144. The CFL module 544 can be used to execute the CFL algorithm, and can include at least one Density Estimation Block 544D and at least one Clustering Block 544C.

[0100] In one embodiment, the density estimation module 544D can be used to receive data 5X, 5Y including microscopic variables and can estimate the probability density function of data 5X, 5Y (such as input variables IV1-IVq, IV11-IV5 or result variable OV). In one embodiment, the density estimation module 544D can calculate, for example, the conditional probability P(5Y|5X) of data 5Y occurring under the condition that data 5X occurs, but is not limited thereto.

[0101] In one embodiment, the clustering module 544C is used to divide data into different clusters. In one embodiment, the clustering module 544C can be at least divided into clustering modules 544C1 and 544C2. The clustering module 544C1 can divide the data 5X into different clusters according to the conditional probability P(5Y|5X), and group the data 5X that predicts similar data 5Y into the same cluster. The clustering module 544C2 can divide the data 5Y into different clusters according to the conditional probability P(5Y|5X), and group the data 5Y with similar responses to interventions into the same cluster. In one embodiment, the data in the same cluster may correspond to a certain or certain causal features, so that the CFL module 544 can extract at least one causal feature (such as CF or cf) from each preprocessed input variable using the CFL algorithm.

[0102] In the CFL algorithm, density estimation and clustering are two main steps for generating macro variables. These macro variables can be interpreted as meaningful scientific quantities and can be used for causal explanations. Among them, in the CFL algorithm, density estimation is the first step in generating macro variables, and clustering is the second step in generating macro variables. Therefore, the CFL module 544 encapsulates a series of modules, covers the main categories of the CFL algorithm, and is used to coordinate the data transformation pipeline.

[0103] In one embodiment, the CFL algorithm adopted by the CFL module 544 may include, for example, Table 1:

[0104] (Table 1)

[0105]

[0106] In one embodiment, the unsupervised CFL algorithm adopted by the CFL module 544 is applicable to the Discrete Time Structural Causal Model (DTSCM). In one embodiment, the continuous time dimension can be considered by, for example, merging time-dependent parameters or modifying the optimization objective, so as to cooperate with the CTSCM. For example, for continuous-time input variables, then, depending on the nature of the input variables and the complexity of the CFL algorithm, one or more of the following methods 1 to 4 can be used to consider the continuous time dimension to improve accuracy and flexibility.

[0107] For example, in Method 1, the CFL algorithm needs to be modified to include time-varying parameters to introduce time-related parameters into the CFL algorithm. In one embodiment, this can be achieved by using a dynamic model of the system (such as a differential equation model). Subsequently, the same optimization procedure as for other parameters of the CFL algorithm can be used to estimate the time-related parameters. In one embodiment, the CFL algorithm can be made to consider the parameters as a function of time. For example, a linear function (such as θ(t) = α + βt, where θ(t) is the parameter value at time point t and α is the initial parameter value) or a non-linear function (such as θ(t) = f(t), where f(t) is a non-linear function) can be used to represent the variation of the parameter over time to incorporate time-related parameters into the CFL algorithm. In one embodiment, a time-varying parameter model can be adopted to represent the influence of the PET sinogram and the MRI sequence on the PET / MRI reconstructed image. For example, the influence of the PET sinogram on the PET / MRI reconstructed image can be regarded as a function of time.

[0108] For example, in Method 2, a time-varying error model can be adopted. In one embodiment, the CFL algorithm can consider the error as a function of time. For example, a noise model can be used to represent the variation of the error over time to incorporate time-related parameters into the CFL algorithm. For example, the noise in the PET sinogram and the MRI sequence can be regarded as a function of time. For example, the noise model can satisfy y(t) = θ(t) + ε(t), where y(t) is the value of the input variable at time point t and ε(t) is the error at time point t.

[0109] For example, in Method 3, a time-related loss function can be adopted to modify the optimization objective of the CFL algorithm in a continuous-time setting, thereby considering the continuous-time dimension. The loss function is related to the optimization objective. The loss function is a measure of the CFL algorithm performing a specific task, while the optimization objective can be the overall objective of the CFL algorithm. The optimization objective can be for the CFL algorithm to learn to accurately capture the potential causal relationship between the preprocessed input variables IV1 to IVq and the result variable under the continuous-time setting. Even in the presence of time-varying signals, the time-related loss function can be designed to encourage the CFL algorithm to learn disentangled and causally related causal features, so that the loss function indirectly affects the mutual information between potential variables or the mutual information between potential variables and observed variables (such as input variables). For example, the loss function can be designed to minimize the mutual information between potential variables at different time steps while maximizing the mutual information between potential variables and observed variables, where the potential variables can be part of the input or output of the CTSCM. In one embodiment, the time-related loss function can include a combination of the standard loss function (such as mean squared error) of the CFL algorithm and a time-related regularization term. The exact form of the time-related loss function depends on the specific application, the attributes of the input variable data being analyzed, and the attributes of the desired causal features.

[0110] For example, in Method 4, a time-related regularization term can be adopted to modify the optimization objective of the CFL algorithm in a continuous-time setting, thereby considering the continuous-time dimension. The regularization term can be a penalty term that can be added to the optimization objective and can be used to penalize the CFL algorithm when it learns causal features that are not causally related or not invariant to time-shifts. In other words, the regularization term can be used to prevent the CFL algorithm from learning causal features that are unimportant for predicting the result variable, causal features that change over time in a manner unrelated to the result variable, or causal features that are inconsistent at different time points. For example, the regularization term can be designed to penalize a model that learns causal features related to the time derivative of the observed variable. In one embodiment, the loss function can include a time-related regularization term. The exact form of the time-related regularization term depends on the specific application, the attributes of the input variable data being analyzed, the CFL algorithm, and the desired causal features to be learned.

[0111] For example, the image reconstruction method 20 may also involve modifying the CFL algorithm adopted by the CFL module 544 for the CTSCM, and may further include the following steps:

[0112] Step S600: Start.

[0113] Step S602: Use a dynamic model (such as a differential equation model) for modeling. Then, proceed to step S604.

[0114] Step S604: Define a time-related loss function to encourage the CFL algorithm to learn disentangled and causally related features. Then, proceed to step S606.

[0115] Step S606: Define a time-related regularization term that penalizes the CFL algorithm for learning features that are not causally related or time-invariant. Then, proceed to step S608.

[0116] Step S608: Train the CFL algorithm using the modified optimization objective.

[0117] Step S610: End.

[0118] In one embodiment, the differential equation model may involve a stochastic differential equation and may satisfy or dη h (t) = (Aη h (t) + ξ h + Bz h + M∑ u x h,u δ(t - u))dt + GdW h (t) (Equation 2). The vector η h (t) is a function of time and can be used to implement model parameter values, parameters, errors, loss functions, or regularization terms. The matrix A can characterize the temporal relationship of the vector η h (t) with self-effects on the diagonal and cross-effects off the diagonal. The matrix I is the identity matrix. The random vector ξ h can determine the long-term trend of the vector η h (t) and can satisfy ξ h ~N(κ, φ ξ ), the vector κ can represent the continuous-time intercept, and the matrix φ ξ can be the covariance. The matrix B can represent the (fixed) time-independent influence of the prediction vector z h on the vector η h (t), and its number of rows may not be equal to the number of columns. The time-related prediction vector x h,u can be observed at time point u and only affects the vector η h (t) at time point u. The impulse formed by x h,u δ(t - u) on the vector η h (t) can be the matrix M. The vector W h(s) can be an independent random walk over a continuous time (e.g., can be a Wiener process), dW h (s) can be a random error term. The lower triangular matrix G represents the influence on the change of the vector η h (t). The matrix Q satisfying Q = GG T represents the variance-covariance matrix of the diffusion process over a continuous time. In one embodiment, the CTSCM can also satisfy Equation 1 or Equation 2.

[0119] In one embodiment, at step S204, the CFL algorithm can convert discrete variables into continuous variables, and thus can be used to preprocess the input variables of the CTSCM, and use the continuous input variables as the input of the CTSCM, so that the CTSCM can learn the causal relationship between the preprocessed input variables and the PET / MRI reconstructed image. The CFL algorithm can first calculate the empirical distribution of the discrete variables, and then use the empirical distribution to generate continuous variables having the same distribution as the discrete variables, and repeat the above procedure for each discrete variable of the input variable data. Since the CTSCM relies on accurate and consistent inputs to understand the causal relationship between the preprocessed input variables and the PET / MRI reconstructed image, by converting the discrete variables into continuous variables, it can be ensured that the input variables meet the requirements of the CTSCM.

[0120] At step S204, the image reconstruction device 10 (e.g., the extraction module 140) can combine (e.g., concatenate) the causal features of all input variables, and establish a unified set of causal features for the input variable data IVD (e.g., PET image and MRI image).

[0121] In one embodiment, the pseudo code corresponding to step S204 of the image reconstruction method 20 can, for example, include:

[0122]

[0123] In step S206, the reconstruction module 160 may use the preprocessed input variable data (such as PET images or MRI images) of the training set with completed causal feature extraction as training data to train a Generative Adversarial Network (GAN) model. Among them, the real (ground truth) PET / MRI images or the existing hybrid imaging PET / MRI images can be used as labels. The reconstruction module 160 may use the GAN model as an image reconstruction algorithm, and based on the causal features (such as CF1~CFr) extracted from the preprocessed input variables (such as IV1~IVq), generate high-quality PET / MRI reconstructed images (such as rIMG) with details, which can play a role in applications that attach importance to visual accuracy, and learn to generate PET / MRI reconstructed images from highly complex and diverse data distributions, which can play a role under the existing difficulties of hybrid imaging, and can fill in the missing data or interpolate between existing data points to reduce the amount of data required for reconstruction.

[0124] For example, Figure 6 FIG. 660 is a schematic diagram of a reconstruction module 660 according to Embodiment 1 of the present invention. The reconstruction module 660 can be used to implement the reconstruction module 160. The reconstruction module 660 may include a generator network 660G and a discriminator network 660D. The generator network 660G and the discriminator network 660D may be neural networks (NNs) respectively.

[0125] In one embodiment, in the training phase, the generator network 660G and the discriminator network 660D may be trained together (in the adversarial process). In one embodiment, the generator network 660G may receive and utilize the preprocessed input variable data IVD (or the causal features CF1~CFr extracted by the CFL module 144, the random noise vector) with completed causal feature extraction in step S206, and attempt to generate a PET / MRI reconstructed image rIMG1 that is used to deceive the discriminator network 660D. The random noise vector is used to introduce randomness into the generator network 660G, which helps to generate diverse and realistic PET / MRI reconstructed images rIMG1. The PET / MRI reconstructed image rIMG1 may be similar to the real (ground truth) PET / MRI image IMG in terms of causality.

[0126] In one embodiment, during the training phase, the discriminator network 660D is used to receive real PET / MRI images IMG (or existing hybrid imaging PET / MRI images IMG) and the PET / MRI reconstructed images rIMG1 generated by the generator network 660G. The discriminator network 660D can be used to compare the features (or visual appearance) of the PET / MRI images IMG and the PET / MRI reconstructed images rIMG1, and assign a probability score to the PET / MRI reconstructed images rIMG1 to indicate the likelihood that the PET / MRI reconstructed images rIMG1 are real, so as to learn or distinguish real PET / MRI images or PET / MRI reconstructed images. The discriminator network 660D does not need to directly understand the causal features. The discriminator network 660D can evaluate the PET / MRI reconstructed images rIMG1 and provide feedback FD to the generator network 660G to improve performance. Over time, the generator network 660G learns to generate PET / MRI reconstructed images rIMG1 that are increasingly similar to the real PET / MRI images IMG, while the discriminator network 660D can more accurately distinguish real PET / MRI images or PET / MRI reconstructed images.

[0127] In step S206, the reconstruction module 160 can also use the trained GAN model and synthesize images according to the causal features of the test set, so as to generate PET / MRI reconstructed images rIMG2. In one embodiment, during the test phase, the generator network 660G can receive and use the input variable data IVD (or the causal features CF1 to CFr extracted by the CFL module 144) that has been preprocessed and completed with causal feature extraction in step S206, and generate PET / MRI reconstructed images rIMG2. The PET / MRI reconstructed images rIMG2 can be similar to the real PET / MRI images IMG in terms of causality, which helps to improve the accuracy and robustness of image reconstruction.

[0128] In other words, the causal features (such as CF1 to CFr) extracted by the CTSCM can be input into the GAN model, and the GAN model can correspondingly generate PET / MRI reconstructed images (such as rIMG2). And since the GAN has been trained, it can generate PET / MRI reconstructed images rIMG2 that are similar to the real PET / MRI images IMG according to the causality learned from the extracted causal features. Using the causal features extracted from the preprocessed input variables to generate synthetic images can help reduce the possible loss of important features of the PET / MRI reconstructed images, because the causal features provide the GAN model with more information about the potential causal mechanisms driving the relationship between the input variables and the result variables, enabling the GAN model to generate images that are more faithful to the underlying data and retain important features, thereby improving the accuracy and robustness of image reconstruction.

[0129] In step S206, the image reconstruction device 10 can also use suitable metrics (such as mean squared error, peak signal-to-noise ratio (PSNR), or structural similarity index (SSIM)) to evaluate the accuracy and quality of the PET / MRI reconstructed images rIMG1 or rIMG2.

[0130] In one embodiment, the virtual code corresponding to step S206 of the image reconstruction method 20 may, for example, include:

[0131]

[0132] In one embodiment, the topology enables an artificial intelligence (AI) server of a local data network to run a medical AI application for intelligent healthcare. In one embodiment, the image reconstruction device 10 may be disposed in an AI server, an imaging device, a computer, or a mobile phone; alternatively, the image reconstruction device 10 may be implemented using a particular machine and externally connected to an imaging device. The modules (such as 120, 142, 144, or 160), networks (such as 660G or 660D), or modules (such as 544D, 544C1, or 544C2) of the image reconstruction device 10 may be implemented using hardware (such as circuits), software, or firmware.

[0133] Figure 7 It is a schematic diagram of a priority triage device 70 according to Embodiment 1 of the present invention. The priority triage device 70 may include a creation module 740, a decision analysis module 780, and a judgment module 790. The creation module 740 may include a causal model creation module 742 and a CFL module 744.

[0134] The causal model creation module 742 can be used to receive input data DT and create a causal model, and the causal model may include state variables SV1 to SVx. The CFL module 744 can be used to extract causal features cf1 to cfy from imaging examinations. The decision analysis module 780 can be used to receive the causal features cf1 to cfy and output confidence levels CL1 to CLz.

[0135] Figure 8 It is a schematic diagram of a priority triage method 80 according to Embodiment 1 of the present invention. The priority triage method 80 is applicable to the priority triage device 70 and may include the following steps:

[0136] Step S800: Start.

[0137] Step S801: Determine the initial state.

[0138] Step S802: Establish a causal model.

[0139] Step S803: Input for Continuous Time multi-criteria decision analysis (CTMCDA).

[0140] Step S804: Output of CTMCDA.

[0141] Step S805: Human intervention.

[0142] Step S806: Update the causal model.

[0143] Step S807: Determine whether the update of the causal model is completed. If the update of the causal model is completed, proceed to Step S808; if not, proceed to Step S803.

[0144] Step S808: Maximize the objective function.

[0145] Step S809: End.

[0146] The priority triage method 80 will be described in detail below. In Step S801, the state variable can start from the initial state, and the initial state represents the medical condition and medical history of the patient.

[0147] In Step S802, the causal model building module 742 can use the Dynamic Causal Planning Graph (DCPG) to create a causal model. In one embodiment, the causal model can include CTSCM. In other words, the priority triage method 80 can adopt a causal AI planning including CTSCM.

[0148] For example, Figure 9 is a schematic diagram corresponding to the continuous time structural causal model 90 in the embodiment of the present invention. In Figure 9 , the dynamic causal planning graphs DCPGt0 to DCPGt3 at different time points t0 to t3 respectively include state variables SV11, SV21 to SV2m, SV31 to SV3n, and SV41 to SV4p. Among them, the state variable SVx can be implemented by using the state variables SV11,..., or SV4p, and the state variable SV11 can be the initial state. As Figure 9As shown, the dynamic causal planning graph (e.g., DCPGt0) only corresponds to a specific instant of the CTSCM 90, while the CTSCM 90 can be composed of countless instants. In other words, the CTSCM 90 can be regarded as including or represented as multiple dynamic causal planning graphs DCPGt0 to DCPGt3, and each of the series of dynamic causal planning graphs DCPGt0 to DCPGt3 can represent the state of the system at a given time point.

[0149] In one embodiment, the dynamic causal planning graph (e.g., DCPGt1) can represent the causal relationship between different state variables (e.g., SV11 and SV2m), where the edge DG of the DCPG represents the causal relationship between state variables and the actions that can be taken to affect the system. In one embodiment, the state variables can, for example, include or be, for example, the patient diagnosis, treatment plan, or overall health outcome of a certain patient; in one embodiment, the state variables can, for example, include or be, for example, imaging examinations, medical conditions, medical histories, patient diagnoses, treatment plans, or overall health outcomes.

[0150] In one embodiment, the DCPG can be a planning graph that allows the causal graph to evolve during planning. In other words, in the DCPG, the causal relationship between state variables is not fixed and can change over time according to the actions taken. In one embodiment, the DCPG can replace the existing planning tree for existing AI planning.

[0151] In one embodiment, the selection of an imaging examination can be regarded as or used as an action that affects the causal relationship between state variables. Selecting an appropriate imaging examination can affect the causal relationship between different state variables, such as affecting patient diagnosis, treatment plan, or overall health outcome, because the accuracy of the information provided by the imaging examination may affect the actions subsequently taken by the healthcare provider.

[0152] In step S802, the CFL module 744 can extract the causal characteristics of the precondition state variables from the previous imaging examination (which can be called the first imaging examination). In one embodiment, a state variable (e.g., SVx) can include at least one causal characteristic (e.g., cfy).

[0153] In one embodiment, causal features may refer to any relevant information obtained from previous imaging examinations. In one embodiment, causal features extracted from imaging examinations include anatomy-based features, contrast agent-based features, texture-based features, shape-based features, or space-based features. In one embodiment, anatomy-based features may refer to features that capture the anatomical structures presented in the imaging examination. For example, anatomy-based features may include the size, density, or location of bones, organs, or blood vessels. In one embodiment, contrast-based features may refer to features that capture the contrast differences in the imaging examination. For example, contrast agent-based features may include the presence or absence of soft tissue or fluid in the image. In one embodiment, texture-based features may refer to features that capture the texture or pattern of the image. For example, texture-based features may include the presence of microcalcifications or lesions. In one embodiment, shape-based features may refer to features that capture the shape or contour of an object in the image. For example, shape-based features may include the curvature of bones or organs. In one embodiment, space-based features may refer to features that capture the spatial relationships between objects in the image. For example, space-based features may include the relative positions of bones or organs.

[0154] In one embodiment, causal features may, for example, include medical imaging examination selection factors or the presence of certain medical conditions or abnormalities. In one embodiment, medical imaging examination selection factors may, for example, include the patient's medical history (such as including any relevant previous medical conditions, surgeries, or procedures), symptoms (such as including the nature, severity, or duration of the patient's symptoms), the patient's current medical condition (such as the patient's current physiological or clinical state), allergies (such as any known drug or contrast agent allergies or adverse reactions), whether the patient is pregnant, the patient's age, the risks or benefits of the imaging examination (such as including contrast agent exposure, radiation exposure, or invasiveness), cost (such as the cost of the imaging examination or related follow-up procedures), the availability or accessibility of imaging equipment or personnel.

[0155] In one embodiment, at step S802, the CFL module 744 may extract causal features using a CFL algorithm. In one embodiment, the CFL algorithm may be an unsupervised CFL algorithm, a CFL neural network, or a CFL neural network in the form of CTSCM.

[0156] For example, Figure 5The CFL module 544 shown can be used to implement the CFL module 744. In one embodiment, the density estimation module 544D of the CFL module 544 can be used to receive data 5X, 5Y and can estimate the probability density function of the data 5X, 5Y. The data 5X or 5Y can be, for example, one or more of the state variables SV1~SVx, SV11~SV4p. For example, the data 5X can be a medical condition and the data 5Y can be an imaging examination. In one embodiment, the clustering module 544C of the CFL module 544 is used to divide the data 5X or 5Y into different clusters. The clustering of the data 5X or 5Y can be performed according to causal features, so that the CFL module 544 can extract causal features (such as CF or cf) from the data 5X or 5Y using the CFL algorithm.

[0157] In one embodiment, the priority triage method 80 may further involve modifying the CFL algorithm used by the CFL module 544 for the CTSCM, and may further include steps S600~S610.

[0158] In one embodiment, the establishment module 740 (such as the CFL module 744) can incorporate causal features into the DCPG, which helps to capture the potential causal relationship between the medical condition and the imaging examination, thus making more accurate and efficient decisions. In one embodiment, the medical condition can include, for example, the health condition or disease, infection, injury, chronic disease or other medical problems that the patient may have or be suspected of having, which may affect the patient's health.

[0159] In step S803, the causal features extracted from the previous imaging examination (or medical imaging test selection factors) are input to the CTMCDA of the decision analysis module 780. In other words, the priority triage method 80 can apply at least one DCPG and CTMCDA in at least one CTSCM.

[0160] In one embodiment, after the causal features are extracted, a causal inference algorithm can be used to identify potential causal mappings. In one embodiment, the Structural Causal Modeling framework can be used as the causal inference algorithm to estimate the causal relationship between the medical condition and the imaging examination.

[0161] In step S804, the CTMCDA of the decision analysis module 780 can calculate the confidence level (e.g., CLz) of the plan for each action or calculate the confidence level (e.g., CL1) of each imaging examination (e.g., each second imaging examination). In one embodiment, the plan of an action can be an effect state variable. In one embodiment, the confidence level can represent whether the next imaging examination (which can be called the second imaging examination) needs to be performed or the necessity of performing the next imaging examination. The confidence level can be related to or based on relevant medical imaging examination selection factors or the causal characteristics of previous imaging examinations.

[0162] For example, if a patient has just undergone a CT scan with an injected contrast agent (which can be used as the first imaging examination) and a brain tumor is detected in the patient, the CTMCDA may calculate a relatively high confidence level for a subsequent MRI imaging (which can be used as the second imaging examination) to further evaluate the size or location of the brain tumor. If the previous CT scan did not show any abnormalities, the CTMCDA may calculate a relatively low confidence level for subsequent imaging examinations, unless there are further changes in the patient's medical condition or relevant factors.

[0163] After the CTMCDA provides the confidence level for the plan of each action in step S804, in step S805, the expert can intervene according to his medical judgment, expertise, or the confidence level (e.g., CLz) to select the best plan of action. Once the best plan of action is determined through the human intervention in step S805, the causal relationship between the prerequisite state variables and the effect state variables is fixed, and then the DCPG can be used to simulate the effects of subsequent actions according to the selected plan of action.

[0164] Accordingly, after the expert selects the best plan of action in step S805, the causal model can be updated in step S806 to reflect the effects of the selected actions. In one embodiment, in step S806, the state variables can be updated according to the selected actions, or new state variables can be generated. For example, in Figure 9 , at time point t1, the dynamic causal planning graph DCPGt1 includes state variables SV11, SV21~SV2m. After determining or selecting the state variable SV2m (which can be used as a candidate imaging examination) from the state variables SV21, …, or SV2m or their corresponding actions (which can be used as the second imaging examination) according to the confidence levels of the state variables SV21, …, or SV2m in step S805, the causal model is updated so that the dynamic causal planning graph DCPGt2 at time point t2 includes state variables SV11, SV21~SV2m, SV31~SV3n, and can reflect the state variables (e.g., newly generated state variables SV31~SV3n) corresponding to the candidate imaging examination (e.g., the state variable SV2m).

[0165] In step S807, the prioritized triage device 70 (e.g., the determination module 790) may determine whether the update of the causal model is completed. If the update of the causal model is completed, steps S803 - S806 are repeated, and the updated causal model is used as the new starting point for the next iteration. For example, the dynamic causal planning graph DCPGt2 at time point t2 includes the dynamic causal planning graph DCPGt1 at time point t1.

[0166] In step S808, the prioritized triage device 70 (e.g., the determination module 790) may maximize the objective function. By using DCPG and CTMCDA under CTSCM, the causal effects of actions on the system over time can be inferred, and the best decision that maximizes the objective function can be made, such as reducing unnecessary imaging examinations while ensuring that patients receive appropriate care.

[0167] In one embodiment, at least a part of the prioritized triage method 80 may be compiled into a virtual code, for example, it may include:

[0168]

[0169]

[0170] In one embodiment, i, j, k, m, n, p, x, y, or z may be positive integers, but are not limited thereto.

[0171] Figure 10 It is a schematic diagram of a causal device 11 according to Embodiment 1 of the present invention. Figure 1 The illustrated image reconstruction device 10 or Figure 7 The illustrated prioritized triage device 70 can be used to implement the causal device 11.

[0172] The causal device 11 may include a causal module 1142 and a CFL module 1144. The causal module 1142 can be used to identify or utilize the causal relationships between multiple variables. Figure 1 The illustrated causal inference module 142 or Figure 7 The illustrated causal model establishment module 742 can be used to implement the causal module 1142.

[0173] The causal feature learning module 1144 can be used to extract at least one causal feature of one of the multiple variables. Figure 1 The illustrated CFL module 144 or Figure 7 The illustrated CFL module 744 can be used to implement the causal feature learning module 1144. The CFL module 1144 can extract causal features by using the CFL algorithm. For example, Figure 5 The illustrated CFL module 544 can be used to implement the CFL module 1144.

[0174] In summary, PET data (such as PET sinograms or PET images) and MRI data (such as MRI sequences or MRI images) can be regarded as input variables of a continuous-time structural causal model, and the CTSCM describes the causal relationships among PET data, MRI data, and PET / MRI reconstructed images. The CTSCM can be used to model system dynamics and capture causal relationships among variables, thereby enabling more accurate and robust image reconstruction.

[0175] In summary, adopting the DCPG and CTMCDA models under the CTSCM is an effective and feasible method for inferring the causal effects of medical imaging examinations and making optimal decisions. By representing the CTSCM as a series of DCPGs, the time-varying causal relationships among medical state variables can be modeled, and the potential impacts of different imaging examinations on these causal relationships can be evaluated. Therefore, the present invention can assist healthcare providers in arranging appropriate imaging examinations for patients based on the best available evidence and clinical judgment.

[0176] The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made according to the claims of the present invention shall fall within the scope covered by the present invention.

Claims

1. A causal device, comprising: A causal module for identifying or exploiting causal relationships among multiple variables; and A causal feature learning module coupled to the causal module for extracting at least one causal feature of one of the multiple variables.

2. The causal device according to claim 1, wherein, The causal device is an image reconstruction device, The causal module identifies the causal relationship between multiple input variables of the multiple variables and the reconstructed image of the multiple variables; The causal feature learning module extracts at least one causal feature from each of the multiple input variables; the causal device further includes a reconstruction module for generating the reconstructed image according to the multiple causal features.

3. The causal device according to claim 1, wherein, One of the multiple input variables of the multiple variables corresponds to a positron emission tomography sinogram, a nuclear magnetic resonance sequence, patient demographic data, an imaging protocol, or scanner characteristics.

4. The causal device according to claim 1, wherein, The causal device further includes: A preprocessing module for preprocessing at least one first input variable data into at least one second input variable data, wherein the at least one second input variable data includes multiple input variables of the multiple variables, and the preprocessing includes attenuation correction, motion correction, registration, normalization, or standardization.

5. The causal device according to claim 1, wherein, The causal module uses a continuous-time structural equation model framework to identify the causal relationship between multiple input variables of the multiple variables and the reconstructed image of the multiple variables; The multiple causal features of the multiple input variables are combined and input into the reconstruction module of the causal device to generate a reconstructed image according to the multiple causal features using a generative adversarial network.

6. The causal device according to claim 1, wherein, The causal feature learning module includes: A density estimation module for estimating multiple probability density functions of the multiple variables; and A clustering module coupled to the density estimation module for classifying the multiple variables into different clusters according to the multiple probability density functions to extract the at least one causal feature.

7. The causal device according to claim 1, wherein, The causal feature learning module extracts the at least one causal feature according to a causal feature learning algorithm, and the causal feature learning algorithm includes at least one parameter that changes over time, an error model that changes over time, a loss function related to time, or a regularization term related to time.

8. The causal device according to claim 1, wherein, The causal device is a priority triage device, and any one of the multiple variables is a first state variable; The causal module uses the causal relationship between the multiple first state variables to create a first causal model; The causal feature learning module extracts the at least one causal feature from a first imaging examination of the first causal model, and the first imaging examination is an imaging examination that has been completed; The causal device further includes a decision analysis module for generating at least one confidence level corresponding to at least one second imaging examination according to the at least one causal feature, and the at least one second imaging examination is an imaging examination that has not been performed.

9. The causal device according to claim 1, wherein, Any one of the multiple variables is a first state variable, and one of the multiple first state variables includes an imaging examination, a medical condition, a medical history, a patient diagnosis, a treatment plan, or an overall health outcome.

10. The causal device according to claim 1, wherein, After determining a candidate imaging examination from at least one second imaging examination, the first causal model is updated to a second causal model, and the second causal model reflects at least one second state variable corresponding to the candidate imaging examination.

11. A causal method for a causal device, comprising: Identifying or exploiting causal relationships among multiple variables; and Extracting at least one causal feature of one of the multiple variables.

12. The causal method according to claim 11, wherein, Identifying causal relationships among the multiple variables includes identifying causal relationships between multiple input variables of the multiple variables and a reconstructed image of the multiple variables; After extracting at least one causal feature from each of the multiple input variables, a reconstructed image is generated based on the multiple causal features.

13. The causal method according to claim 11, wherein, One of the multiple input variables of the multiple variables corresponds to a positron emission tomography sinogram, a nuclear magnetic resonance sequence, patient demographic data, an imaging protocol, or scanner characteristics.

14. The causal method according to claim 11, further comprising: Preprocessing at least one first input variable data into at least one second input variable data, where the at least one second input variable data includes the multiple input variables, and the preprocessing includes attenuation correction, motion correction, registration, normalization, or standardization.

15. The causal method according to claim 11, wherein, Identifying causal relationships among the multiple variables includes using a continuous-time structural equation model framework to identify causal relationships between multiple input variables of the multiple variables and a reconstructed image of the multiple variables; After the multiple causal features of the multiple input variables are combined, a generative adversarial network is used to generate a reconstructed image based on the multiple causal features.

16. The causal method according to claim 11, further comprising: Estimating multiple probability density functions of the multiple variables; And Classifying the multiple variables into different clusters according to the multiple probability density functions to extract the at least one causal feature.

17. The causal method according to claim 11, wherein the at least one causal feature is extracted according to a causal feature learning algorithm, and the causal feature learning algorithm includes at least one parameter that changes over time, an error model that changes over time, a loss function related to time, or a regularization term related to time.

18. The causal method according to claim 11, wherein, Any one of the multiple variables is a first state variable; Utilizing the causal relationships among the multiple variables includes creating a first causal model using the causal relationships among the multiple first state variables; After extracting the at least one causal feature from a first imaging examination of the first causal model, generating at least one confidence level corresponding to at least one second imaging examination based on the at least one causal feature, where the first imaging examination is a completed imaging examination and the at least one second imaging examination is an imaging examination not yet performed.

19. The causal method according to claim 11, wherein, Any one of the multiple variables is a first state variable, and one of the multiple first state variables includes an imaging examination, a medical condition, a medical history, a patient diagnosis, a treatment plan, or an overall health outcome.

20. The causal method according to claim 11, wherein, After determining a candidate imaging examination from at least one second imaging examination, the first causal model is updated to a second causal model, and the second causal model reflects at least one second state variable corresponding to the candidate imaging examination.

Images