Device for Parkinson pathology subtype typing and storage medium

Abstract

The invention discloses a device for Parkinson's disease pathological subtype typing and a storage medium. The method comprises the following steps: extracting a magnetic susceptibility value of a key brain region based on acquired magnetic resonance image data, and constructing a standardized feature matrix; taking the standardized feature matrix as a graph node, and initializing a weighted adjacent matrix representing a causal relationship between the graph nodes; constructing a target optimization function based on each graph node and the weighted adjacency matrix; iteratively optimizing the weighted adjacency matrix according to the target optimization function to obtain an initial causal graph containing initial nodes and initial connection edges; performing post-processing on the initial causal graph to form a specific causal graph containing a target node and a causal relationship edge; and performing causal effect evaluation on the basis of the specific causal diagram to obtain a typing result of the Parkinson pathological subtype. By means of the scheme, objective and accurate subtype typing can be achieved, a pathological transmission mechanism is revealed, and a support is provided for early intervention and individualized diagnosis and treatment.

Description

TECHNICAL FIELD

[0001] The present application generally relates to the technical field of Parkinson's disease subtyping. More specifically, the present application relates to an apparatus for subtyping of Parkinson's disease pathology and a computer readable storage medium. BACKGROUND

[0002] Parkinson's Disease (PD) is the second most common neurodegenerative disease after Alzheimer's disease, and its core pathological manifestations are the misfolding and abnormal aggregation of alpha-synuclein, which spreads in the nervous system in a prion-like manner, ultimately leading to progressive and selective neuronal loss. The body-first / brain-first hypothesis reveals the heterogeneity of PD in terms of origin and transmission path, and there are significant differences in early intervention and treatment regimens for different subtypes, so accurate subtyping is crucial for individualized diagnosis and treatment.

[0003] The existing Parkinson's disease pathology subtyping technology has many key defects: traditional methods rely on clinical symptom observation, such as whether accompanied by rapid eye movement sleep behavior disorder ("RBD"), lack objective quantitative indicators, are easily influenced by physician experience, and have poor consistency; existing image analysis is mostly limited to correlation statistics of brain iron content, and cannot determine the causal transmission relationship between iron deposition in brain regions, making it difficult to distinguish the pathological origin differences between body-first and brain-first subtypes; it is impossible to achieve accurate subtyping in the prodromal stage (such as isolated RBD stage), delaying the opportunity for early intervention; physiological and causal relationships in healthy populations are not excluded, resulting in disease-related causal signals being masked, and limiting the accuracy of subtyping.

[0004] Therefore, there is an urgent need to provide a scheme for subtyping of Parkinson's disease pathology, which can achieve objective and accurate subtyping through causal modeling of brain iron deposition, reveal the pathological transmission mechanism, and provide support for early intervention and individualized diagnosis and treatment. SUMMARY

[0005] In order to at least solve one or more of the above-mentioned technical problems, the present application proposes a scheme for subtyping of Parkinson's disease pathology in multiple aspects.

[0006] In a first aspect, the application provides an apparatus for subtyping of Parkinson's pathology, comprising: a processor; and a memory having stored thereon computer instructions for subtyping of Parkinson's pathology, which when executed by the processor cause the following operations to be implemented: extracting susceptibility values of key brain regions based on collected magnetic resonance imaging data, and constructing a standardized feature matrix; initializing a weighted adjacency matrix representing causal relationships between graph nodes with the standardized feature matrix as the graph nodes; constructing an objective optimization function based on the graph nodes and the weighted adjacency matrix, wherein the objective optimization function comprises a fitting function related to data fitting, a sparse regularization term related to complexity of causal graph structure, and a loop-free constraint penalty term related to causal logic; iteratively optimizing the weighted adjacency matrix according to the objective optimization function to obtain an initial causal graph comprising initial nodes and initial connection edges; post-processing the initial causal graph to form a specific causal graph comprising target nodes and causal relationship edges; and performing causal effect evaluation based on the specific causal graph to obtain a subtyping result of Parkinson's pathology.

[0007] In some embodiments, wherein the magnetic resonance imaging data comprises structural image data for brain structure localization and specific sequence data for brain iron content quantification of a target population, wherein the target population comprises isolated RBD patients, RBD-companied Parkinson's patients, Parkinson's patients without RBD, and healthy controls.

[0008] In some embodiments, wherein before constructing the standardized feature matrix, the apparatus further performs the following operation: performing a pre-processing operation on the susceptibility values, wherein the pre-processing operation comprises one or more operations of outlier filtering, missing value imputation, or standardization.

[0009] In some embodiments, wherein the apparatus further constructs the objective optimization function by the following formula: wherein, represents the objective optimization function, represents the fitting function, ， represents the sparse regularization term, ， represents the loop-free constraint penalty term, , represents the graph node, represents data dimension, represents sample number, represents the weighted adjacency matrix, and represents a penalty coefficient.

[0010] In some embodiments, the device further performs the following operations to obtain the initial causal graph: iteratively updating the weighted adjacency matrix in the objective optimization function in continuous space using a first-order optimization method until the objective optimization function meets a convergence condition, to obtain the initial causal graph containing initial nodes and initial connection edges.

[0011] In some embodiments, the device further performs the following operations to form the specific causal graph: thresholding the initial causal graph to obtain a directed acyclic causal graph; and removing causal edges of the healthy control group from the directed acyclic causal graph to form the specific causal graph containing target nodes and causal relationship edges.

[0012] In some embodiments, the device further performs the following operations to obtain the directed acyclic causal graph: setting the causal edge weight below a preset value in the initial causal graph to zero; and deleting the edge with the minimum weight until all directed loops are eliminated, to obtain the directed acyclic causal graph.

[0013] In some embodiments, the typing result includes a body-priority subtype or a brain-priority subtype.

[0014] In some embodiments, the device further performs the following operations to obtain the typing result: selecting a target item and an intervention item from the specific causal graph; quantitatively evaluating a causal effect value of the intervention item on the target item; and obtaining the typing result based on the causal structure of the specific causal graph and the causal effect value.

[0015] In a second aspect, the present application provides a computer-readable storage medium having stored thereon computer program instructions for subtyping of Parkinson's pathology, which, when executed by one or more processors, cause the operations performed by the device described in the foregoing first aspect to be implemented.

[0016] By the scheme for subtyping of Parkinson's pathology as provided above, the embodiments of the present application achieve objective quantification of the basis for subtyping by extracting key brain region magnetization characteristics from collected magnetic resonance image data, constructing an objective optimization function containing a data fitting term, a sparse constraint term and a loop-free constraint term, iteratively optimizing to obtain a causal graph and obtaining a disease-specific causal structure through post-processing, and combining causal effect evaluation to achieve subtyping, reveal the pathological transmission mechanism of different subtypes, enable precise subtyping in the prodromal stage of the disease, effectively filter physiological interference of healthy people, and improve the accuracy and reliability of subtyping, thereby providing a scientific basis for early intervention and individualized diagnosis and treatment of Parkinson's disease. BRIEF DESCRIPTION OF DRAWINGS

[0017] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which a number of embodiments of the present application are shown by way of illustration, now merely by way of example, in which: Figure 1 is an exemplary structural block diagram illustrating an apparatus 100 for subtyping of Parkinson's pathology according to embodiments of the present application; Figure 2 is an exemplary flow block diagram illustrating operations 200 implemented by an apparatus for subtyping of Parkinson's pathology according to embodiments of the present application; Figure 3 is an exemplary flow block diagram illustrating an overall for subtyping of Parkinson's pathology according to embodiments of the present application; Figure 4 is an exemplary schematic diagram illustrating a specific causal graph according to embodiments of the present application; Figure 5 is an exemplary structural block diagram illustrating an electronic device 500 according to embodiments of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be apparently and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without any creative work fall within the protection scope of the present application.

[0019] It should be understood that the terms "comprise" and "include" used in the specification and claims of the present application indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0020] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. As used in the specification and claims of the present application, unless otherwise clearly indicated by the context, the singular forms "a", "an" and "the" are intended to include the plural forms as well. It should be further understood that the term "and / or" used in the specification of the present application means any combination of one or more of the associated listed items and all possible combinations thereof.

[0021] As used in the specification and claims, the term “if’ can be interpreted as meaning “when” or “upon” or “in response to a determination” or “in response to a detection” depending on the context. Similarly, the phrase “if it is determined” or “if [the described condition or event] is detected” can be interpreted as meaning “upon a determination” or “in response to a determination” or “upon a detection of [the described condition or event]” or “in response to a detection of [the described condition or event]” depending on the context.

[0022] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0023] Figure 1 is an exemplary structural block diagram showing an apparatus 100 for Parkinson's pathology subtyping according to an embodiment of the present application. As shown in Figure 1 The apparatus 100 can include a processor 110 and a memory 120, as shown in the figure. The processor 110 can include, for example, a general-purpose processor (“CPU”) or a dedicated graphics processor (“GPU”), and the memory 120 stores program instructions executable on the processor. In some embodiments, the memory 120 can include, but is not limited to, resistive random access memory RRAM (Resistive Random Access Memory), dynamic random access memory DRAM (Dynamic Random Access Memory), static random access memory SRAM (Static Random-Access Memory), enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory).

[0024] Further, the memory 120 can store program instructions for Parkinson's pathology subtyping, and when the program instructions are executed by the processor 110, the apparatus 100 implements the following operations: extracting the magnetic susceptibility values of the key brain regions based on the collected magnetic resonance image data, and constructing a standardized feature matrix; initializing a weighted adjacency matrix representing the causal relationship between the graph nodes with the standardized feature matrix as the graph nodes; constructing a target optimization function based on each graph node and the weighted adjacency matrix, wherein the target optimization function includes a fitting function related to data fitting, a sparse regularization term related to the complexity of the causal graph structure, and a loop-free constraint penalty term related to the causal logic; iteratively optimizing the weighted adjacency matrix according to the target optimization function to obtain an initial causal graph containing initial nodes and initial connection edges; post-processing the initial causal graph to form a specific causal graph containing target nodes and causal relationship edges; and performing causal effect evaluation based on the specific causal graph to obtain the subtyping result of Parkinson's pathology. The following will be described in detail in conjunction with Figure 2The operations implemented by the apparatus 100 of the embodiments of the present application are described in detail.

[0025] Figure 2 is an exemplary flow block diagram illustrating operations 200 implemented by an apparatus for subtyping Parkinson's pathology according to embodiments of the present application. As shown in Figure 2 As shown in the step S201, based on the collected magnetic resonance image data, the susceptibility values of the key brain regions are extracted, and a standardized feature matrix is constructed.

[0026] In some embodiments, the magnetic resonance image data includes structural image data for brain structure localization and specific sequence data for brain iron content quantification of a target population, where the target population includes isolated RBD patients (iRBD), RBD-companied Parkinson's patients (PD-RBD), Parkinson's patients without RBD (PD-nRBD), and healthy controls (HC). In some implementation scenarios, the structural image data for brain structure localization may, for example, be T1 / T2 weighted structural images, which can accurately locate the anatomical positions of the key brain regions. The specific sequence data for brain iron content quantification may, for example, be multi-echo gradient echo sequence data. After being processed by quantitative susceptibility mapping (QSM) technology, the sequence data can generate a whole brain susceptibility map, realizing accurate quantification of brain iron content.

[0027] The same model of magnetic resonance imaging system is used in the acquisition process, and uniform scanning parameters (such as magnetic field strength, echo number, scanning layer thickness, etc.) are set to avoid data deviation caused by equipment differences and inconsistent parameters, so as to ensure the consistency of the image data.

[0028] In some implementation scenarios, the key brain regions can be determined based on a pre-defined brain atlas, for example, a standardized brain atlas based on neuroanatomy. The key brain regions can include the pedunculopontine nucleus (PPN), the substantia nigra compacta (SNc), the entorhinal cortex (EC), the amygdala (Amyg), and the superior frontal gyrus (SFG). These brain regions are closely related to the pathological progression of Parkinson's disease, for example, PPN is a key origin brain region of the body-priority subtype, Amyg is a key origin brain region of the brain-priority subtype, and SNc, EC, and SFG are important target regions of pathological spread.

[0029] In some embodiments, before constructing the standardized feature matrix, the apparatus further performs the following operation: performing a pre-processing operation on the susceptibility values, where the pre-processing operation includes one or more of outlier filtering, missing value imputation, or standardization.

[0030] Specifically, first, QSM reconstruction is performed on the multi-echo gradient echo sequence data to generate a whole brain susceptibility map, and then the susceptibility map is spatially standardized, for example, registered to the MNI standard brain space. Next, the influence of inter-individual brain structural differences is eliminated through outlier filtering and / or missing value imputation, and the average susceptibility value of each key brain region in the standardized susceptibility map is extracted as the iron deposition quantification index of the brain region. The D-dimensional feature vector of each individual is composed of the susceptibility values of all key brain regions, and D is the number of key brain regions. The D-dimensional feature vectors of all individuals in the target population are collected to construct a standardized feature matrix , that is, the D-dimensional feature vector of the kth sample.

[0031] Next, at step S202, the standardized feature matrix is used as the graph node, and the weighted adjacency matrix representing the causal relationship between the graph nodes is initialized. It can be understood that each column of the standardized feature matrix corresponds to the susceptibility standardized value of a key brain region, so each column vector is a node of the causal graph, and the name of the node is the name of the corresponding key brain region, for example, R_PPN, L_SNc, etc. L and R are used to distinguish left and right brain regions. The value of the node is the susceptibility standardized value of the brain region in different samples, and the node is essentially an iron deposition quantification index of the key brain region, and the number of nodes is equal to the data dimension D.

[0032] The weighted adjacency matrix G is the mathematical carrier of the causal graph, and the dimension is DxD. The matrix element G[i][j] represents the causal correlation strength between the ith node (i.e. brain region i) and the jth node (i.e. brain region j). In some implementation scenarios, during initialization, a random initialization method can be used to assign the matrix element to a small random number in the interval [-0.1, 0.1], so as to ensure that there is only weak correlation between brain regions in the initial state, and to avoid that the initial value is too large to affect the subsequent optimization result.

[0033] Further, at step S203, a target optimization function is constructed based on the graph nodes and the weighted adjacency matrix, wherein the target optimization function includes a fitting function related to data fitting, a sparse regularization term related to the structural complexity of the causal graph, and a loop-free constraint penalty term related to the causal logic.

[0034] That is, the subsequent specific causal graph needs to satisfy the data fitting degree, the structural sparsity, and the directed acyclic property. Among them, the data fitting degree makes the causal relationship conform to the real susceptibility data; the structural sparsity only retains strong causal correlation and eliminates weak correlation edges to avoid overfitting; and the directed acyclic property conforms to the irreversibility of the causal logic. Therefore, the target optimization function can realize the above requirements through three function terms, and balance the weights of each term through a penalty coefficient.

[0035] In some embodiments, the device further constructs the target optimization function through the following formula: wherein, represents a target optimization function, represents a fitting function, , det represents a determinant operation of a matrix; represents a sparse regularization term, , is an element-wise defined L1 norm, i.e., the sum of absolute values of all elements of a matrix; represents the acyclic constraint penalty term, , tr represents a trace of a matrix, represents an exponent of a matrix A, represents a Hadamard product; represents the graph node, represents a data dimension, represents a sample number, represents a weighted adjacency matrix, and represents a penalty coefficient.

[0036] It can be understood that, the first part measures the fitting error by calculating the sum of squared residuals of real data and model predicted data; and the second part introduces the constraint of causal structure through the matrix determinant term to ensure the rationality of the causal relationship. The greater the value of , the better the fitting effect of the causal relationship with the real data.

[0037] By penalizing the number of non-zero elements in the weighted adjacency matrix G, only a small number of strong causal edges are retained in the optimization process, and weakly related edges are removed, thereby controlling the structural complexity of the causal graph, avoiding model overfitting caused by full connection, and conforming to the medical cognition that a small number of key brain regions dominate pathological progression.

[0038] In , when the graph corresponding to the weighted adjacency matrix G has a directed loop, the diagonal elements of the matrix will significantly increase, causing the value of to increase, thereby forcing the graph corresponding to the optimized G matrix to satisfy the directed acyclic graph (DAG) property through this penalty mechanism, conforming to the irreversibility of causal logic.

[0039] In some implementation scenarios, the above and can be determined by, for example, 5-fold cross-validation to determine the optimal value, is used to balance the fitting effect and sparsity, is used to balance the fitting effect and acyclicity. For example, if is too large, it will result in too few edges and loss of important causal relationships; if Overfitting leads to poor fitting effect, and deviation from the causal relationship and real data.

[0040] Based on the above constructed objective optimization function, at step S204, the weighted adjacency matrix is iteratively optimized according to the objective optimization function, to obtain an initial causal graph containing initial nodes and initial connection edges.

[0041] In some embodiments, the device further performs the following operations to obtain the initial causal graph: iteratively updating the weighted adjacency matrix in the objective optimization function in the continuous space by using a first-order optimization method until the objective optimization function satisfies a convergence condition, to obtain the initial causal graph containing the initial nodes and the initial connection edges. That is, taking the weighted adjacency matrix G as the optimization variable, the gradient of the objective optimization function is calculated , and the element values of the G matrix are updated in the negative direction of the gradient.

[0042] In each iteration process, the data fitting degree, the graph structure sparsity, and the directed acyclic property are considered simultaneously. When the difference between two adjacent iterations of the objective optimization function is less than a preset threshold (such as 1e-6), or the number of iterations reaches a preset upper limit (such as 1000 rounds), the iteration is stopped. At this time, the weighted adjacency matrix G obtained is the optimal matrix after optimization. The weighted adjacency matrix G after optimization is converted into a graphical structure, that is, the initial causal graph. The columns and rows of the matrix correspond to the nodes (corresponding to the key brain regions) of the causal graph, respectively, the positions of the matrix elements G[i][j] that are not zero correspond to the initial connection edges from the node i to the node j in the causal graph, the numerical value of the element is the weight of the edge (corresponding to the causal effect strength), and the sign of the element indicates the influence direction, for example, the positive weight promotes the action, and the negative weight inhibits the action.

[0043] After obtaining the above initial causal graph, at step S205, the initial causal graph is post-processed to form a specificity causal graph containing target nodes and causal relationship edges. In some embodiments, the device further performs the following operations to form the specificity causal graph: performing thresholding processing on the initial causal graph to obtain a directed acyclic causal graph; and eliminating the causal edges of the healthy control group from the directed acyclic causal graph to form the specificity causal graph containing the target nodes and the causal relationship edges.

[0044] In some embodiments, the device further performs the following operations to obtain the directed acyclic causal graph: setting the causal edge weight below a preset value in the initial causal graph to zero; and deleting the edge with the minimum weight until all directed loops are eliminated, to obtain the directed acyclic causal graph.

[0045] Specifically, in some implementation scenarios, the aforementioned preset value can be determined based on several (e.g., 3) times of standard deviation of the sample data, and the standard deviation is the standard deviation of all non-zero elements of the weighted adjacency matrix G. The edge weights in the initial causal graph whose absolute values are lower than the threshold are set to zero, that is, the edges with weak causal correlation are removed, and only the edges with strong causal correlation are retained, thereby further strengthening the sparsity of the causal graph and improving the reliability of the causal relationship.

[0046] Then, it is checked whether the causal graph has a directed loop, for example, A→B→C→A. In implementation scenarios, a loop detection algorithm (such as a depth-first search DFS) in graph theory is used to traverse the causal graph, and if a directed loop is found, the edge with the minimum absolute weight in the loop is deleted. The operation is repeated until all directed loops in the causal graph are eliminated, and a directed acyclic causal graph is obtained. This step ensures that the causal graph conforms to the causal logic and avoids the circular contradiction of the causal relationship.

[0047] Further, the directed acyclic causal graphs of the healthy control group (HC) and the Parkinson's patient cohort (including iRBD, PD-RBD, and PD-nRBD) (c∈{iRBD,PD-RBD,PD-nRBD}). Using a graph theory cut operation, the causal edges that coincide in the causal graph of the patient cohort and the causal graph of the healthy control group are cut, that is , where \ represents the cut operation, and the edges in with non-zero weights at the same positions as those in are deleted. That is, the physiological and chemical causal relationships common to the healthy population are removed, and only the specific causal relationships related to the disease are retained, thereby avoiding the interference of physiological factors on the disease classification.

[0048] In the causal graph after cutting the causal edges of the healthy control group, a subgraph in the range of the pre-defined key brain regions (PPN, SNc, EC, Amyg, and SFG) is extracted. The subgraph is a specific causal graph that contains the target nodes (key brain regions) and the disease-specific causal relationship edges. The purpose of the subgraph extraction is to focus on the core pathological related brain regions, exclude the interference of irrelevant brain regions, and simplify the complexity of subsequent causal effect evaluation and classification analysis.

[0049] Finally, at step S206, causal effect evaluation is performed based on the specific causal graph, and a classification result of the Parkinson's pathological subtype is obtained. In some embodiments, the aforementioned classification result includes a body-preferred subtype or a brain-preferred subtype. In some embodiments, the apparatus further performs the following operations to obtain the classification result: selecting a target item and an intervention item from the specific causal graph; quantitatively evaluating the causal effect value of the intervention item on the target item; and obtaining the classification result based on the causal structure and the causal effect value of the specific causal graph.

[0050] In some implementation scenarios, the intervention item and the target item are selected according to the distribution of the directed edges of the specific causal graph. Specifically, if there is a directed edge of brain region A→brain region B in the causal graph, brain region A is selected as the intervention item and brain region B is selected as the target item, that is, the causal effect of the iron deposition change of brain region A on the iron deposition change of brain region B is evaluated. For example, if there is a directed edge of R_PPN→L_SNc, the intervention item is R_PPN (right side of the peduncular nucleus) and the target item is L_SNc (left side of the substantia nigra compacta).

[0051] In other implementation scenarios, the causal effect strength of the intervention item on the target item can be quantified by, for example, average treatment effect (ATE). The greater the absolute value of ATE, the stronger the causal effect of the intervention item on the target item.

[0052] In yet other implementation scenarios, to ensure the correctness and reliability of the causal relationship, two types of refutation tests can be performed to verify the causal relationship. One is a random confounder refutation test, that is, a randomly generated confounder (a random variable unrelated to real data) is added to the causal model, and the average treatment effect ATE' is recalculated. If the original causal relationship is correct, after adding the random confounder, ATE' has no significant change compared with the original ATE, that is, the p value is greater than 0.05, indicating that the model is not disturbed by irrelevant confounders.

[0053] The other is a placebo intervention refutation test, which replaces the intervention item with a placebo variable (such as a randomly generated irrelevant brain region feature) that has no causal relationship with the target item, and recalculates ATE''. If the original causal relationship is correct, the absolute value of ATE'' should be significantly close to 0, indicating that the model is only sensitive to the real causal relationship.

[0054] Based on the structural features and causal effect value features of the specific causal graph, the typing result can be determined. Specifically, if the specific causal graph has a causal starting point of a brainstem nucleus (for example, PPN), and the absolute value of the causal effect value of the starting point meets a preset high effect threshold (for example, |ATE|≥1.0), and the influence range covers at least two other key brain regions (for example, SNc, EC), it is determined as the body priority subtype. This subtype corresponds to isolated RBD patients and RBD associated Parkinson patients, whose pathological diffusion mode is from bottom to top and relatively symmetric, corresponding to the body priority subtype. In PD-RBD, the number of causal relationships is more than that in iRBD, further embodying the progression of pathological progression.

[0055] If the specific causal graph takes the edge area brain region (e.g., Amyg) as the causal starting point, and the absolute value of the causal effect value of the starting point meets the preset high effect threshold (e.g., |ATE|≥1.0), and the influence range covers at least two other key brain regions (e.g., SFG, EC), it is determined as a brain priority subtype. This subtype corresponds to Parkinson's patients without RBD, and the pathological diffusion mode is from top to bottom and asymmetric, corresponding to the brain priority subtype.

[0056] As described above, the embodiments of the present application extract the magnetic susceptibility values of the key brain regions by collecting magnetic resonance image data and construct a standardized feature matrix. After initializing the weighted adjacency matrix, a target optimization function containing data fitting, sparsity constraint and loop-free constraint is constructed. After iterative optimization and post-processing, a disease-specific causal graph is formed. Combined with causal effect evaluation, the pathological subtype of Parkinson's disease is classified. The core realizes the objective quantification of the classification basis, breaks through the limitations of traditional subjective evaluation, and reveals the pathological transmission mechanism of different subtypes.

[0057] Further, the embodiments of the present application further improve the accuracy, reliability and specificity of the classification by collecting specific image data and target groups, preprocessing the magnetic susceptibility values, optimizing the target function construction and iteration method, refining the causal graph post-processing process, strengthening the causal relationship verification and clarifying the classification rules, effectively filtering physiological interference of healthy people, adapting to the whole cycle classification needs from the prodromal stage to the confirmed stage, and providing scientific support for individualized diagnosis and treatment.

[0058] Figure 3 is an exemplary flow chart showing the whole for the pathological subtype classification of Parkinson's disease according to the embodiments of the present application. As shown in Figure 3 In step S301, magnetic resonance image data is collected. For example, T1 / T2 weighted structural images and multi-echo gradient echo sequence data of iRBD, PD-RBD, PD-nRBD and HC. In step S302, the magnetic susceptibility values of the key brain regions are extracted, and in step S303, the magnetic susceptibility values are preprocessed, such as outlier filtering, missing value interpolation or standardization, to construct a standardized feature matrix. As described previously, the key brain regions can include PPN, SNc, EC, Amyg and SFG. The magnetic susceptibility values of the key brain regions can be obtained by using QSM on the sequence data.

[0059] Then, in step S304, the target optimization function is constructed. As known from the foregoing, first, the standardized feature matrix can be taken as the graph node, and the weighted adjacency matrix representing the causal relationship between the graph nodes is initialized. By considering the data fitting degree, structural sparsity and directed acyclic nature, a target optimization function containing a fitting function related to data fitting , a sparsity regular term related to the complexity of the causal graph structure and a loop-free constraint penalty term related to the causal logic Objective optimization function For more details, please refer to the above. Figure 2 The description of the subject matter will not be repeated here.

[0060] Based on the constructed objective optimization function, in step S305, the weighted adjacency matrix is ​​iteratively optimized. Specifically, a first-order optimization method, such as gradient descent, can be used to iteratively update the weighted adjacency matrix in the objective optimization function in continuous space. In step S306, it is determined whether the objective optimization function satisfies the convergence condition. If it does not converge, return to step S305; if it converges, the initial causal graph is obtained. Further, in step S307, the initial causal graph is thresholded, and in step S308, directed cycles are detected and deleted. Specifically, the weights of causal edges in the initial causal graph below a preset value are reset to zero. The causal graph is traversed by, for example, Depth-First Search (DFS). If a directed cycle is found, the edge with the smallest absolute weight on the cycle is deleted until all directed cycles are eliminated, resulting in a directed acyclic causal graph.

[0061] Further, in step S309, causal edges representing the healthy control group are removed from the directed acyclic causal graph. In step S310, a specific causal graph containing the target node and causal relationship edges is formed. Based on the obtained specific causal graph, in step S311, causal effect evaluation is performed to obtain the typing results of the body-preferred subtype or brain-preferred subtype in step S312. For more details, please refer to the above. Figure 2 The descriptions made will not be repeated here.

[0062] Figure 4 This is an exemplary schematic diagram illustrating a specific causal graph according to an embodiment of this application. For example... Figure 4 The diagram illustrates, from left to right, disease-specific causal maps of brain iron deposition corresponding to iRBD, PD-RBD, and PD-nRBD, respectively. Each node represents a key brain region distinguishing between the left and right hemispheres, and the edges represent the directed causal relationships between these regions. The style of the edges—solid / dashed lines and their thickness—represents the positive / negative direction and effect strength of the causal effect, respectively. Specifically, solid arrows represent positive causal effects, i.e., increased iron deposition in one brain region promotes increased iron deposition in the resulting brain region; dashed arrows represent negative causal effects, i.e., increased iron deposition in one brain region inhibits increased iron deposition in the resulting brain region. Thin solid lines correspond to 0 ≤ |ATE| < 1.0; medium solid lines correspond to 1.0 ≤ |ATE| < 2.0; and thick solid lines correspond to |ATE| > 2.0.

[0063] As can be seen from the figure, the iRBD subgraph takes the right pedal nucleus (R_PPN) as the core node, and presents multi-directional causal conduction, such as L_SNc→R_PPN (solid line, positive value), R_PPN→R_SNc (solid line, positive value), R_PPN→L_EC (solid line, positive value), R_SFG→L_SFG (solid line, positive value), L_EC→L_SFG (solid line, positive value). That is, taking the brain stem nucleus (pedal nucleus) as the starting point, the causal effect is mostly promoting, involving the initial pathological conduction of the substantia nigra compacta, the entorhinal cortex, and the superior frontal gyrus, and the whole embodies the brain iron deposition causal network characteristics of the pre-Parkinson disease of the body priority subtype.

[0064] The PD-RBD subgraph nodes include left / right pedal nuclei (L_PPN, R_PPN), left / right substantia nigra compacta (L_SNc, R_SNc), left / right amygdala (L_Amyg, R_Amyg), left / right entorhinal cortex (L_EC, R_EC), and left / right superior frontal gyrus (L_SFG, R_SFG). The causal network is significantly expanded, involving multi-directional interaction of the pedal nucleus, the substantia nigra compacta, the amygdala, the entorhinal cortex, and the superior frontal gyrus, such as L_SNc→R_PPN (solid line, positive value), R_PPN→R_SNc (solid line, positive value), L_Amyg→L_EC (dashed line, negative value), R_EC→L_SFG (solid line, positive value), etc. The causal conduction is more complex, with positive and negative effect directions, involving an expanded range of brain regions, and the whole embodies the pathological progression of the clinical stage of Parkinson's disease of the body priority subtype.

[0065] The PD-nRBD subgraph nodes include left / right amygdala (L_Amyg, R_Amyg), left / right entorhinal cortex (L_EC, R_EC), left / right substantia nigra compacta (L_SNc, R_SNc), left / right pedal nucleus (L_PPN, R_PPN), and left / right superior frontal gyrus (L_SFG, R_SFG). It takes the right amygdala (R_Amyg) as the core node, and presents a top-down causal diffusion, such as R_Amyg→L_Amyg (solid line, positive value), R_Amyg→R_SNc (thick solid line, |ATE|>2.0), R_Amyg→L_EC (medium solid line, 1.0≤|ATE|<2.0), R_PPN→L_PPN (dashed line, negative value), etc. That is, taking the edge area brain region (amygdala) as the causal starting point, the causal effect intensity is high (there are thick edges with |ATE|>2.0), the diffusion mode involves cross-brain region conduction from the edge area to the substantia nigra compacta, the entorhinal cortex, and the pedal nucleus, and the whole embodies the pathological characteristics of the brain priority subtype.

[0066] As can be seen, the specific causal graph obtained based on the scheme of the embodiments of the present application can directly obtain the pathological propagation mechanism of different subtypes after causal effect evaluation, improving the accuracy, reliability, and specificity of the typing, and providing scientific support for individualized diagnosis and treatment.

[0067] Figure 5 An exemplary structural block diagram of an electronic device 500 according to an embodiment of this application is shown. It will be understood that the device implementing the solution of this application may be a single device (e.g., a computing device) or a multifunctional device including various peripheral devices.

[0068] like Figure 5 As shown, the electronic device of this application may include a central processing unit (“CPU”) 511, which may be a general-purpose CPU, a dedicated CPU, or other information processing and program execution unit. Furthermore, the electronic device 500 may also include a mass storage device 512 and a read-only memory (“ROM”) 513. The mass storage device 512 may be configured to store various types of data, including various magnetic resonance imaging data, normalized feature matrices, weighted adjacency matrices, objective optimization functions, causal graphs, classification results, algorithm data, intermediate results, and various programs required to run the electronic device 500. The ROM 513 may be configured to store power-on self-test (POST) data for the electronic device 500, initialization of various functional modules in the system, drivers for the system's basic input / output, and data and instructions required to boot the operating system.

[0069] Optionally, the electronic device 500 may also include other hardware platforms or components, such as the tensor processing unit (“TPU”) 514, graphics processing unit (“GPU”) 515, field-programmable gate array (“FPGA”) 516, and machine learning unit (“MLU”) 517 shown. It is understood that although various hardware platforms or components are shown in the electronic device 500, they are merely exemplary and not limiting, and those skilled in the art can add or remove appropriate hardware as needed. For example, the electronic device 500 may include only a CPU, associated storage devices, and interface devices to implement the operations performed by the device for Parkinson's pathology subtyping of this application.

[0070] In some embodiments, to facilitate the transmission and interaction of data with external networks, the electronic device 500 of the present application further comprises a communication interface 518, so that it can be connected to a local area network / wireless local area network (“LAN / WLAN”) 505 through the communication interface 518, and then connected to a local server 506 or connected to the Internet (“Internet”) 507 through the LAN / WLAN. Alternatively or additionally, the electronic device 500 of the present application can also be directly connected to the Internet or a cellular network based on wireless communication technology through the communication interface 518, such as based on the third generation (“3G”), fourth generation (“4G”), or fifth generation (“5G”) wireless communication technology. In some application scenarios, the electronic device 500 of the present application can also access the servers 508 and databases 509 of external networks as needed in order to obtain various known algorithms, data, and modules, and can remotely store various data, such as various types of data or instructions for presenting, for example, magnetic resonance image data, standardized feature matrices, weighted adjacency matrices, objective optimization functions, causal graphs, typing results, etc.

[0071] The peripheral devices of the electronic device 500 can include a display device 502, an input device 503, and a data transmission interface 504. In one embodiment, the display device 502 can for example include one or more speakers and / or one or more visual displays configured for voice prompting and / or image / video display of the operations performed by the device for subtyping of Parkinson's pathology of the present application. The input device 503 can include, for example, a keyboard, a mouse, a microphone, a gesture capture camera, and other input buttons or controls configured for receiving input of audio data and / or user instructions. The data transmission interface 504 can include, for example, a serial interface, a parallel interface, or a universal serial bus interface (“USB”), a small computer system interface (“SCSI”), a serial ATA, a FireWire, a PCI Express, and a high-definition multimedia interface (“HDMI”), etc., configured for data transmission and interaction with other devices or systems. According to the scheme of the present application, the data transmission interface 504 can receive magnetic resonance image data acquired by a magnetic resonance device, and transmit to the electronic device 500 magnetic resonance image data or various other types of data or results.

[0072] The above-mentioned CPU 511, mass storage 512, ROM 513, TPU 514, GPU 515, FPGA 516, MLU 517, and communication interface 518 of the electronic device 500 of the present application can be connected to each other through a bus 519, and realize data interaction with peripheral devices through the bus. In one embodiment, through the bus 519, the CPU 511 can control other hardware components and their peripheral devices in the electronic device 500.

[0073] The above combinations Figure 5 An electronic device that can be used to perform the present application is described. It needs to be understood that the device structure or architecture herein is merely exemplary, and implementations and entities of the present application are not limited thereby, but can be changed without departing from the spirit of the present application.

[0074] According to the above description in conjunction with the drawings, those skilled in the art can also understand that the embodiments of the present application can also be implemented by a software program. Therefore, the present application also provides a computer readable storage medium, which stores computer readable instructions for subtyping of Parkinson's pathology, and the computer readable instructions can be executed by one or more processors to implement the present application in conjunction with the above description. Figure 1 Operations performed by the apparatus for subtyping of Parkinson's pathology.

[0075] It should be noted that although the operations of the method of the present application are described in a particular order in the drawings, this does not require or imply that the operations must be performed in that particular order, or that all of the illustrated operations must be performed to achieve the desired results. On the contrary, the steps depicted in the flowchart can be changed in execution order. Additionally or alternatively, some steps can be omitted, a plurality of steps can be combined into one step, and / or one step can be divided into a plurality of steps.

[0076] It should be understood that when the terms "first", "second", "third", and "fourth" and the like are used in the claims, the specification and the drawings of the present application, these terms are used only to distinguish different objects, and are not used to describe a particular order. The terms "include" and "contain" used in the specification and claims of the present application indicate the presence of the described features, whole, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, whole, steps, operations, elements, components and / or sets thereof.

[0077] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. As used in the specification and claims of the present application, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include the plural forms. It should be further understood that the term "and / or" used in the specification and claims of the present application means any combination of one or more of the associated listed items and all possible combinations thereof, and includes these combinations.

[0078] While several embodiments of the application have been shown and described herein, it will be obvious to those skilled in the art that many changes, modifications, and substitutions can be made to the embodiments without departing from the spirit and scope of the application. It is to be understood that various alternatives to the embodiments of the application described herein can be employed in practicing the application. The appended claims are intended to cover all such alternatives as would be included within the spirit and scope of the application.

Claims

1. A device for subtyping Parkinson's disease pathology, comprising: processor; as well as A memory storing computer instructions for subtyping Parkinson's pathology subtypes, which, when executed by a processor, cause the following operations to be performed: Magnetization values ​​of key brain regions were extracted based on the acquired magnetic resonance imaging data, and a standardized feature matrix was constructed. Using the standardized feature matrix as graph nodes, initialize a weighted adjacency matrix representing causal relationships between graph nodes; A target optimization function is constructed based on the graph nodes and the weighted adjacency matrix, wherein the target optimization function includes a fitting function related to data fitting, a sparse regularization term related to the structural complexity of the causal graph, and an acyclic constraint penalty term related to causal logic. The weighted adjacency matrix is ​​iteratively optimized according to the objective optimization function to obtain an initial causal graph containing initial nodes and initial connecting edges; The initial causal graph is post-processed to form a specific causal graph containing the target node and causal relationship edges; Based on the specific causal map, the causal effect was evaluated to obtain the classification results of Parkinson's pathological subtypes.

2. The apparatus of claim 1, wherein the magnetic resonance imaging data includes structural imaging data for brain structure localization and specific sequence data for quantifying brain iron content in the target population, wherein the target population includes patients with isolated RBD, Parkinson's patients with RBD, Parkinson's patients without RBD, and healthy controls.

3. The apparatus of claim 1, wherein before constructing the standardized feature matrix, the apparatus further performs the following operations: The magnetic susceptibility value is preprocessed, wherein the preprocessing operation includes one or more of the following: outlier filtering, missing value imputation, or standardization.

4. The apparatus according to claim 1, wherein the apparatus further constructs the objective optimization function by the following formula: in, Denotes the objective optimization function. Denotes the fitting function, , This represents the sparse regularization term. , This represents the acyclic constraint penalty term. , Represents the graph node, Indicates data dimension, Indicates the number of samples. Denotes the weighted adjacency matrix, and This represents the penalty coefficient.

5. The apparatus of claim 1, wherein the apparatus further comprises the following operations to obtain the initial causal graph: The weighted adjacency matrix in the objective optimization function is iteratively updated in a continuous space using a first-order optimization method until the objective optimization function satisfies the convergence condition, thereby obtaining the initial causal graph containing the initial nodes and initial connecting edges.

6. The apparatus of claim 1, wherein the apparatus further comprises the following operations to form the specific causal graph: The initial causal graph is thresholded to obtain a directed acyclic causal graph; The causal edges of the healthy control group are removed from the directed acyclic causal graph to form the specific causal graph containing the target node and causal relationship edges.

7. The apparatus of claim 6, wherein the apparatus further comprises the following operation to obtain the directed acyclic causal graph: Reset the causal edge weights in the initial causal graph that are below the preset value to zero; Delete the edge with the smallest weight until all directed cycles are eliminated, and obtain the directed acyclic causal graph.

8. The apparatus of claim 1, wherein the typing result includes a body-preferred subtype or a brain-preferred subtype.

9. The apparatus of claim 8, wherein the apparatus further comprises the following operation to obtain the classification result: Select target items and intervention items from the specific causal graph; Quantitatively assess the causal effect of the intervention on the target item; The typing results are obtained based on the causal structure of the specific causal graph and the causal effect value.

10. A computer-readable storage medium having stored thereon computer program instructions for subtyping Parkinson's pathology, the computer program instructions, when executed by one or more processors, causing the operation performed by the apparatus according to any one of claims 1-9 to be implemented.

Images