Artificial intelligence-based patient screening method and system in clinical research

Abstract

The application provides an artificial intelligence-based patient screening method and system in clinical research, relates to the technical field of artificial intelligence, and comprises the following steps: obtaining a clinical research enrollment standard text and performing semantic analysis to extract a rule tuple, and converting the rule tuple into a standard schema graph; selecting a disease node in a medical knowledge graph to perform extension matching to obtain an adaptive patient subgraph; performing graph structure alignment to identify a matching substructure, a missing item and a conflict item; and finally calculating the adaptability of a patient to a research project and generating a matching result. The application improves the patient screening efficiency and accuracy, and reduces the clinical trial recruitment cost.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a patient screening method and system based on artificial intelligence in clinical research. Background Technology

[0002] Patient screening in clinical research is a crucial step in ensuring research quality and the reliability of results. Traditional clinical research patient screening typically relies on manual screening of patient medical records to determine the match between the records and the study's inclusion criteria. Researchers need to review each patient's medical record and compare it with the study's inclusion / exclusion criteria to determine if a patient meets the conditions for participating in a specific clinical study. With the advent of the era of big data in healthcare and the widespread application of electronic medical record systems, a data foundation has been provided for automated patient screening. Currently, some studies have begun to explore applying artificial intelligence technologies such as natural language processing and machine learning to the patient screening process in clinical trials. By constructing algorithmic models, they analyze the match between patient medical record data and clinical study inclusion criteria to improve screening efficiency.

[0003] However, traditional patient screening methods lack sufficient semantic understanding of inclusion criteria. Existing technologies typically employ keyword matching or simple rule templates, which struggle to accurately parse the complex semantic structure of inclusion criteria, particularly those involving multiple logical conditions, exclusion criteria, and time constraints, resulting in low accuracy of screening results.

[0004] Secondly, existing technologies lack deep integration of patient medical records and medical knowledge. Most screening systems simply compare patient data with inclusion criteria using direct matching, failing to utilize medical knowledge graphs for knowledge reasoning and thus unable to identify implicit medical connections, such as causal relationships or complication relationships, thereby missing potentially suitable subjects or incorrectly including unsuitable patients. Summary of the Invention

[0005] This invention provides a patient screening method and system based on artificial intelligence in clinical research, which can solve the problems in the prior art.

[0006] A first aspect of the present invention provides an artificial intelligence-based patient screening method in clinical research, comprising:

[0007] Obtain the inclusion criteria text of the clinical research project, perform semantic parsing on the inclusion criteria text, extract the inclusion rule tuple containing condition types, constraint logic and medical concepts, and transform the inclusion rule tuple into a standard pattern diagram containing necessary entity nodes and relational paths.

[0008] Starting from a pre-constructed medical knowledge graph, a disease node is randomly selected and expanded along the causal and co-disease association edges in the medical knowledge graph. When the relationship type of the expanded path matches the relationship path type in the standard pattern graph, the node is retained to obtain an adapted patient subgraph. The medical knowledge graph is constructed based on the acquired medical record data of the patients to be matched through entity mapping.

[0009] The adapted patient subgraph is topologically aligned with the standard pattern graph. Matching substructures in the adapted patient subgraph that are consistent with the structure of the standard pattern graph are identified. Missing conditions that exist in the standard pattern graph but are missing in the adapted patient subgraph are marked. Conflicting conditions that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph are also marked.

[0010] The coverage of the matching substructure, the missing conditions, and the conflicting conditions are normalized and weighted to calculate the patient's fit for the clinical research project, and the patient's clinical research matching result is generated based on the fit.

[0011] Semantic parsing is performed on the inclusion standard text to extract inclusion rule tuples containing condition types, constraint logic, and medical concepts. These inclusion rule tuples are then transformed into a standard schema diagram containing necessary entity nodes and relational paths, including:

[0012] The inclusion criteria text is parsed syntactically to identify conditional statements describing the inclusion conditions;

[0013] For each conditional statement, semantic role labeling identifies condition type elements representing the nature of conditional constraints, constraint logic elements representing the logical combination relationship between multiple conditions, and medical concept elements representing clinical medical concepts. The condition type elements, constraint logic elements, and medical concept elements are then structurally combined according to the semantic dependency relationship in the conditional statement to form an ingress rule tuple.

[0014] The medical concept elements in the grouping rule tuple are mapped to corresponding entity nodes. The relationship path type and connection method that need to be satisfied between the corresponding entity nodes are determined according to the constraint logic elements. The necessity level of the corresponding entity node and the relationship path type is marked according to the condition type elements.

[0015] The standard pattern diagram is constructed based on the corresponding entity node, the relationship path type, the connection method, and the necessity level.

[0016] Extending along the causal and comorbidity association edges in the medical knowledge graph, and retaining nodes when the relationship type of the extended path matches the relationship path type in the standard pattern graph, the resulting adapted patient subgraph includes:

[0017] Starting from the symptom node, the expansion process is layered according to topological distance. In each topological distance level, the expansion is carried out along the causal association edge and the co-disease association edge to the next level. The topological distance value from each expanded node to the symptom node and the sequence of relation types traversed in the expansion path are recorded.

[0018] Extract the relationship path types and their topological location information in the standard pattern graph, perform sequence matching between the relationship type sequence of the extended node and the relationship path type in the standard pattern graph, and align the topological distance value of the extended node with the topological location information of the corresponding relationship path in the standard pattern graph.

[0019] The node is retained when the sequence of relation types of the extended node matches the relation path type in the standard pattern graph and the topological distance value is consistent with the topological location information.

[0020] Based on the symptom nodes, the retained nodes, the relation type sequence, and the topological distance value, the adapted patient subgraph is constructed.

[0021] Starting from the symptom node, the expansion process is layered according to topological distance. In each topological distance level, the expansion proceeds to the next level along causal and co-disease association edges. The topological distance value from each expanded node to the symptom node and the sequence of relation types traversed in the expansion path are recorded, including:

[0022] The symptom node is marked as the initial level, and the topological distance value of the initial level is set to zero;

[0023] Starting from the symptom node, traverse all outgoing edges of the symptom node, identify first-type adjacent nodes connected by causal association edges and second-type adjacent nodes connected by co-disease association edges, mark the first-type adjacent nodes and the second-type adjacent nodes as a first topological distance level, record a topological distance value of one for each node in the first topological distance level, record the relationship type sequence as causal association type for the first-type adjacent nodes, and record the relationship type sequence as co-disease association type for the second-type adjacent nodes;

[0024] Starting from each node in the first topological distance level, traverse all its outgoing edges, identify the lower-level adjacent nodes connected by causal or co-causal edges, mark the lower-level adjacent nodes as the second topological distance level, record the topological distance value of each lower-level adjacent node in the second topological distance level as the topological distance value of its parent node plus one, and record the relationship type sequence of the lower-level adjacent node as the concatenation result of the relationship type sequence of its parent node and the relationship type of the current extended edge;

[0025] Repeat the above expansion process until all nodes in all topological distance levels have been traversed, and obtain an expanded result set containing the topological distance values ​​and relation type sequences of all expanded nodes.

[0026] The adapted patient subgraph is topologically aligned with the standard pattern graph. Matching substructures in the adapted patient subgraph that are consistent with the structure of the standard pattern graph are identified. Missing conditions that exist in the standard pattern graph but are missing in the adapted patient subgraph are marked. Condition conflict conditions that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph are marked, including:

[0027] Using each standard entity node in the standard pattern diagram as an anchor point, a patient entity node that semantically matches the standard entity node is searched in the adapted patient subgraph. When a corresponding patient entity node is found,

[0028] Extract the first path type and standard topology connection structure of the standard entity node in the standard pattern graph; extract the second path type and patient topology connection structure of the patient entity node in the adapted patient subgraph;

[0029] When the first path type is completely consistent with the second path type and the standard topology connection structure is successfully matched with the patient topology connection structure, the patient entity node and the patient relationship path connected in the adapted patient subgraph are marked as a matching substructure;

[0030] Traverse the standard entity nodes. When a standard entity node does not find a semantically matching corresponding patient entity node in the adapted patient subgraph, mark the standard entity node and the standard relation path type connected in the standard pattern graph as a missing condition.

[0031] When the outgoing edge patient relationship path type of a certain patient entity node does not exist in the standard relationship path type set, the patient entity node and its outgoing edge patient relationship path type are marked as condition conflict items.

[0032] After normalizing and weighting the coverage of the matching substructure, the missing condition terms, and the conflicting condition terms, the fit of the patient to the clinical research project is calculated as follows:

[0033] Extract the topological adjacency matrix of the matching substructure and the standard adjacency matrix of the standard pattern graph, calculate the matrix similarity between the topological adjacency matrix and the standard adjacency matrix, and use the matrix similarity as the original term of the coverage degree of the matching substructure;

[0034] The connectivity of the standard entity node corresponding to each missing item in the missing condition is calculated in the standard pattern diagram. The connectivity is used as the missing influence factor of the missing item, and the missing influence factor is used as the original item of the missing condition.

[0035] The semantic distance between the patient entity node and the standard entity node corresponding to each conflict term in the conditional conflict terms is calculated. The semantic distance is used as the conflict intensity factor of the conflict term. The conflict intensity factors of all conditional conflict terms are summed to obtain the total conflict intensity factor. The total conflict intensity factor is used as the original term of the conditional conflict term.

[0036] After performing maximum and minimum value normalization on the original items of the coverage, the original items of the missing conditions, and the original items of the conflicting conditions, corresponding weight values ​​are assigned, and the fit of the patient to the clinical research project is calculated.

[0037] A second aspect of the present invention provides an artificial intelligence-based patient screening system for clinical research, comprising:

[0038] The first unit is used to obtain the inclusion criteria text of clinical research projects, perform semantic parsing on the inclusion criteria text, extract inclusion rule tuples containing condition types, constraint logic and medical concepts, and transform the inclusion rule tuples into a standard pattern diagram containing necessary entity nodes and relational paths.

[0039] The second unit is used to randomly select a symptom node in a pre-constructed medical knowledge graph as the starting point, and expand along the causal association edges and co-disease association edges in the medical knowledge graph. When the relationship type of the expansion path matches the relationship path type in the standard pattern graph, the node is retained to obtain an adapted patient subgraph. The medical knowledge graph is constructed based on the acquired medical record data of the patients to be matched through entity mapping.

[0040] The third unit is used to perform topological alignment between the adapted patient subgraph and the standard pattern graph, identify matching substructures in the adapted patient subgraph that are consistent with the structure of the standard pattern graph, mark missing condition items that exist in the standard pattern graph but are missing in the adapted patient subgraph, and mark condition conflict items that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph.

[0041] The fourth unit is used to normalize and weight the coverage of the matching substructure, the missing conditions and the conflicting conditions, calculate the fit of the patient to the clinical research project, and generate the patient's clinical research matching result based on the fit.

[0042] A third aspect of the embodiments of the present invention,

[0043] An electronic device is provided, comprising:

[0044] processor;

[0045] Memory used to store processor-executable instructions;

[0046] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0047] Fourth aspect of the present invention,

[0048] A computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0049] The beneficial effects of this application are as follows:

[0050] Patient suitability analysis based on medical knowledge graphs fully utilizes causal and comorbid relationships between diseases, enabling a comprehensive assessment of patients' clinical characteristics and avoiding the limitations of traditional methods that only consider isolated symptoms or diagnoses.

[0051] By identifying matching substructures, missing conditions, and conflicting conditions through topological alignment, the fit between patients and clinical research requirements is precisely quantified, providing a more granular assessment dimension compared to simple Boolean matching.

[0052] The overall approach enables the automatic conversion from unstructured medical text to structured knowledge, significantly reducing the human and time costs of patient screening in clinical research, while improving the objectivity and reproducibility of the screening process. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the process of a patient screening method based on artificial intelligence in a clinical study according to an embodiment of the present invention;

[0054] Figure 2 A flowchart illustrating the process of obtaining a subgraph that fits the patient. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0057] refer to Figure 1 and Figure 2 The patient screening method based on artificial intelligence in clinical research according to embodiments of the present invention includes:

[0058] Obtain the inclusion criteria text of the clinical research project, perform semantic parsing on the inclusion criteria text, extract the inclusion rule tuple containing condition types, constraint logic and medical concepts, and transform the inclusion rule tuple into a standard pattern diagram containing necessary entity nodes and relational paths.

[0059] Starting from a pre-constructed medical knowledge graph, a disease node is randomly selected and expanded along the causal and co-disease association edges in the medical knowledge graph. When the relationship type of the expanded path matches the relationship path type in the standard pattern graph, the node is retained to obtain an adapted patient subgraph. The medical knowledge graph is constructed based on the acquired medical record data of the patients to be matched through entity mapping.

[0060] The adapted patient subgraph is topologically aligned with the standard pattern graph. Matching substructures in the adapted patient subgraph that are consistent with the structure of the standard pattern graph are identified. Missing conditions that exist in the standard pattern graph but are missing in the adapted patient subgraph are marked. Conflicting conditions that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph are also marked.

[0061] The coverage of the matching substructure, the missing conditions, and the conflicting conditions are normalized and weighted to calculate the patient's fit for the clinical research project, and the patient's clinical research matching result is generated based on the fit.

[0062] In one optional implementation, semantic parsing is performed on the inclusion standard text to extract inclusion rule tuples containing condition types, constraint logic, and medical concepts. The inclusion rule tuples are then transformed into a standard schema diagram containing necessary entity nodes and relational paths, including:

[0063] The inclusion criteria text is parsed syntactically to identify conditional statements describing the inclusion conditions;

[0064] For each conditional statement, semantic role labeling identifies condition type elements representing the nature of conditional constraints, constraint logic elements representing the logical combination relationship between multiple conditions, and medical concept elements representing clinical medical concepts. The condition type elements, constraint logic elements, and medical concept elements are then structurally combined according to the semantic dependency relationship in the conditional statement to form an ingress rule tuple.

[0065] The medical concept elements in the grouping rule tuple are mapped to corresponding entity nodes. The relationship path type and connection method that need to be satisfied between the corresponding entity nodes are determined according to the constraint logic elements. The necessity level of the corresponding entity node and the relationship path type is marked according to the condition type elements.

[0066] The standard pattern diagram is constructed based on the corresponding entity node, the relationship path type, the connection method, and the necessity level.

[0067] Syntactic parsing is performed on the inclusion criteria text to identify conditional statements describing the inclusion conditions. Natural language processing (NLP) techniques are used to preprocess the inclusion criteria text, including word segmentation, part-of-speech tagging, and dependency parsing. By identifying conditional conjunctions (such as "if," "when," "if," etc.), relational words (such as "have," "suffer from," "diagnosed as," etc.), and conditional markers (such as "need," "must," "should," etc.), conditional statements describing the inclusion conditions are located. For example, for the text "The patient's age must be greater than 18 years and less than 65 years, and must be pathologically diagnosed with non-small cell lung cancer," two conditional statements can be identified: "The patient's age must be greater than 18 years and less than 65 years" and "must be pathologically diagnosed with non-small cell lung cancer."

[0068] For each conditional statement, semantic role labeling identifies condition type elements representing the nature of conditional constraints, constraint logic elements representing the logical combination relationship between multiple conditions, and medical concept elements representing clinical medical concepts. Condition type elements include "necessary condition," "exclusion condition," and "preferred condition," typically identified by words such as "must," "prohibited," and "preferred." Constraint logic elements include logical relationships such as "AND," "OR," and "NOT," typically represented by words such as "and," "or," and "not." Medical concept elements include medical professional terms such as disease names, symptoms, examinations, and drugs. A pre-trained biomedical language model and named entity recognition technology are used to identify medical concepts and map them to standard medical terminology databases (such as SNOMEDCT and ICD-10).

[0069] The condition type elements, constraint logic elements, and medical concept elements are structurally combined according to their semantic dependencies in the conditional statement to form an inclusion rule tuple. By analyzing the semantic dependency tree, the hierarchy and subordinate relationships between the elements are determined. For example, for the conditional statement "patient age must be greater than 18 years old and less than 65 years old", the condition type element can be extracted as "necessary condition", the constraint logic element as "AND", and the medical concept elements as "age > 18 years old" and "age < 65 years old", forming the inclusion rule tuple: (necessary condition, AND, [age > 18 years old, age < 65 years old]).

[0070] The medical concept elements in the inclusion rule tuple are mapped to corresponding entity nodes. Entity nodes include types such as "patient characteristics", "disease", "symptoms", "examination", and "treatment". Each entity node contains information such as entity type, entity value, and entity attribute. For example, "age > 18 years" is mapped to the entity node {type: "patient characteristics", attribute: "age", value: > 18 years"}; "non-small cell lung cancer" is mapped to the entity node {type: "disease", value: "non-small cell lung cancer", code: "C34.90"}.

[0071] The relationship path type and connection method that need to be satisfied between corresponding entity nodes are determined based on the constraint logic elements. Relationship path types include "have", "do not have", and "confirmed as", etc., and connection methods include "sequential connection", "parallel connection", and "mutually exclusive connection", etc. For example, for the constraint logic element "AND", a parallel connection method is used; for the constraint logic element "OR", a mutually exclusive connection method is used.

[0072] Label the corresponding entity nodes and relationship path types with necessity levels based on the condition type elements. Necessity levels include "must," "excluded," and "priority," each corresponding to different visual representations. For example, the necessity level corresponding to "necessary condition" is "must," represented by a solid line in the standard schema diagram; the necessity level corresponding to "excluded condition" is "excluded," represented by a dashed line in the standard schema diagram.

[0073] A standard pattern diagram is constructed based on the corresponding entity nodes, relationship path types, connection methods, and necessity levels. The standard pattern diagram adopts a directed graph structure, with entity nodes as nodes and relationship paths as edges. The styles of nodes and edges are determined according to the entity type and necessity level. In practical applications, graphical tools (such as D3.js, ECharts, etc.) can be used to visualize the standard pattern diagram.

[0074] In a specific example, for the inclusion criterion text "Patients aged ≥18 years and ≤70 years, diagnosed by histopathology as stage IIIB or IV non-small cell lung cancer, with negative EGFR gene test, and no prior anti-tumor treatment", syntactic parsing is first performed to identify four conditional statements. Then, the condition type elements, constraint logic elements, and medical concept elements in each conditional statement are extracted to form an inclusion rule tuple. Finally, a standard schema diagram containing four entity nodes and corresponding relational paths is constructed to intuitively display the logical structure and constraints of the inclusion criteria.

[0075] The above method can transform the natural language description of the enrollment criteria text into a structured standard schema diagram, which is easier for computers to understand and process, thereby improving the efficiency and accuracy of clinical trial enrollment screening.

[0076] In one optional implementation, the medical knowledge graph is expanded along causal association edges and comorbidity association edges. Nodes are retained when the relationship type of the expanded path matches the relationship path type in the standard pattern graph, resulting in an adapted patient subgraph including:

[0077] Starting from the symptom node, the expansion process is layered according to topological distance. In each topological distance level, the expansion is carried out along the causal association edge and the co-disease association edge to the next level. The topological distance value from each expanded node to the symptom node and the sequence of relation types traversed in the expansion path are recorded.

[0078] Extract the relationship path types and their topological location information in the standard pattern graph, perform sequence matching between the relationship type sequence of the extended node and the relationship path type in the standard pattern graph, and align the topological distance value of the extended node with the topological location information of the corresponding relationship path in the standard pattern graph.

[0079] The node is retained when the sequence of relation types of the extended node matches the relation path type in the standard pattern graph and the topological distance value is consistent with the topological location information.

[0080] Based on the symptom nodes, the retained nodes, the relation type sequence, and the topological distance value, the adapted patient subgraph is constructed.

[0081] In this embodiment, a patient-fitting subgraph is constructed by expanding along causal and co-disease association edges in the medical knowledge graph. Starting from the symptom node, the graph is expanded hierarchically according to topological distance, recording the topological distance from the expanded node to the symptom node and the sequence of relation types traversed. Then, the relation type sequence of the expanded node is matched with the relation path type in the standard pattern graph, and the topological distance is aligned with the topological position information of the corresponding path in the standard pattern graph, thereby retaining the matching nodes to construct the patient-fitting subgraph.

[0082] The system identifies target disease nodes from a medical knowledge graph as starting points. For example, if the target disease is "hypertension," the hypertension node is used as the starting point for expansion. Simultaneously, a predefined standard pattern graph is loaded, which contains standard path patterns related to diseases, such as relationship path types and their topological location information, like "disease - [causal relationship] -> complications - [comorbidity relationship] -> medication."

[0083] The expansion process begins with hierarchical expansion based on topological distance. First, the topological distance of the symptom nodes is set to 0, representing layer 0. Then, expansion proceeds sequentially to layer 1, layer 2, and so on. During the expansion at each layer, all nodes in the current layer are traversed, and nodes in the next layer connected to these nodes via causal or co-pathological edges are queried. For example, starting from the hypertension node, expansion might extend to the "heart disease" node (causal association) and the "diabetes" node (co-pathological association). For each expanded node, its topological distance value (i.e., the current layer number plus 1) and the sequence of relation types traversed from the starting node to that node are recorded.

[0084] When recording the sequence of relation types, the actual type of the edge is recorded. For example, if hypertension leads to heart disease through a causal relationship edge, and then to kidney disease through a comorbidity relationship edge, the relation type sequence is ["causal relationship", "comorbidity relationship"]. Simultaneously, the topological distances of these extended nodes are recorded, with heart disease having a topological distance of 1 and kidney disease having a topological distance of 2.

[0085] After expansion, the relationship path types and their topological location information defined in the standard pattern graph are extracted. The standard pattern graph contains various disease association patterns; for example, the topological distance for "disease - [causal association] -> complications" is 1, and the topological distance for "disease - [causal association] -> complications - [comorbidity association] -> other diseases" is 2. This information will be used in the subsequent matching process.

[0086] For each extended node, a matching judgment is performed, comparing the sequence of relation types of the extended node with the relation path types in the standard pattern graph. For example, the relation type sequence of the extended node "nephropathy" is ["causal association", "comorbidity association"]. If the same relation path type exists in the standard pattern graph, the next matching step is performed. Secondly, the topological distance value of the extended node is compared with the topological position information of the corresponding relation path in the standard pattern graph. For example, the topological distance of "nephropathy" is 2. If the topological position of the corresponding path in the standard pattern graph is also 2, then the topological distance is considered a match.

[0087] When the relation type sequence of an extended node matches the relation path type in the standard pattern graph, and the topological distance value is consistent with the topological location information, the node is retained as an adapting node. For example, if both the relation type sequence and topological distance of a nephropathy node match the standard pattern graph, the nephropathy node is retained in the adapting patient subgraph.

[0088] Based on symptom nodes, all retained nodes, the sequence of relationship types between them, and topological distance values, an adapted patient subgraph is constructed. This subgraph contains nodes and relationships consistent with the standard pattern graph structure and can be used for subsequent patient similarity analysis or disease prediction.

[0089] In practical applications, this method can be used to analyze a patient's disease progression path. For example, for a newly diagnosed patient with hypertension, the constructed patient-fit subgraph can identify the potential risks of subsequent complications, such as heart disease or kidney disease. Doctors can then use this information to develop more targeted prevention and treatment plans, thereby improving medical outcomes and the patient's quality of life.

[0090] Furthermore, this method can also be used for disease association analysis in medical research. Researchers can discover novel association patterns between diseases by comparing aptamer subgraphs from different patients, providing new ideas and directions for medical research.

[0091] In one optional implementation, starting from the symptom node, the expansion process is layered according to topological distance. In each topological distance layer, the expansion proceeds to the next lower layer along causal association edges and co-disease association edges. The topological distance value from each expanded node to the symptom node and the sequence of relation types traversed in the expansion path are recorded, including:

[0092] The symptom node is marked as the initial level, and the topological distance value of the initial level is set to zero;

[0093] Starting from the symptom node, traverse all outgoing edges of the symptom node, identify first-type adjacent nodes connected by causal association edges and second-type adjacent nodes connected by co-disease association edges, mark the first-type adjacent nodes and the second-type adjacent nodes as a first topological distance level, record a topological distance value of one for each node in the first topological distance level, record the relationship type sequence as causal association type for the first-type adjacent nodes, and record the relationship type sequence as co-disease association type for the second-type adjacent nodes;

[0094] Starting from each node in the first topological distance level, traverse all its outgoing edges, identify the lower-level adjacent nodes connected by causal or co-causal edges, mark the lower-level adjacent nodes as the second topological distance level, record the topological distance value of each lower-level adjacent node in the second topological distance level as the topological distance value of its parent node plus one, and record the relationship type sequence of the lower-level adjacent node as the concatenation result of the relationship type sequence of its parent node and the relationship type of the current extended edge;

[0095] Repeat the above expansion process until all nodes in all topological distance levels have been traversed, and obtain an expanded result set containing the topological distance values ​​and relation type sequences of all expanded nodes.

[0096] Choose a symptom node as the starting point for expansion. This node can be any disease, symptom, or syndrome already defined in the medical knowledge graph. For example, choose "Type 2 diabetes" as the initial symptom node for expansion.

[0097] Mark the symptom node as the initial level and set its topological distance value to zero. Create a data structure to store information for each node, including node identifier, node type, topological distance value, and relation type sequence. For the initial symptom node, its topological distance value is zero, and its relation type sequence is empty.

[0098] Starting from the initial symptom node, traverse all outgoing edges from that node. In the medical knowledge graph, these outgoing edges mainly include causal relationship edges and comorbidity relationship edges. Causal relationship edges represent a direct cause-effect relationship between diseases, such as "hypertension leads to coronary heart disease"; comorbidity relationship edges represent two diseases that frequently occur simultaneously but have no clear causal relationship, such as "diabetes mellitus comorbid with retinopathy".

[0099] During the traversal, the set of nodes connected by causal association edges is identified and classified as first-type adjacent nodes; at the same time, the set of nodes connected by co-disease association edges is identified and classified as second-type adjacent nodes. For example, starting with "type 2 diabetes", it may be found that "kidney disease" is a first-type adjacent node connected by causal association edges, while "hypertension" is a second-type adjacent node connected by co-disease association edges.

[0100] The first and second types of adjacent nodes are merged and jointly labeled as the first topological distance level. Each node in this level is assigned a topological distance value of one, and a sequence of relationship types is recorded based on its association type with the initial node. Specifically, for the first type of adjacent nodes, the relationship type sequence is recorded as "causal association"; for the second type of adjacent nodes, the relationship type sequence is recorded as "common disease association".

[0101] Establish a queue or stack structure to enqueue or push all nodes in the first topological distance level sequentially for subsequent processing. Simultaneously, use a set structure to record visited nodes to avoid redundant expansion and potential loops.

[0102] Process each node in the first topological distance level sequentially, traversing all its outgoing edges. For each node, identify its lower-level adjacent nodes connected by causal or co-disease-related edges. For example, starting from the "kidney disease" node, it may be found that "uremia" is a lower-level node connected by a causal edge.

[0103] For each identified lower-level neighbor node, first check if it has been visited. If not, mark it as a second topological distance level and record its topological distance value as the topological distance value of its parent node plus one. In this example, the topological distance value of "uremia" is one plus one, which is two, compared to the topological distance value of "kidney disease".

[0104] Simultaneously, a sequence of relation types is recorded for each lower-level adjacent node. This sequence is formed by concatenating the relation type sequence of the parent node with the relation type of the currently extended edge. For example, if the relation type sequence of "kidney disease" is "causal association", and there is a causal association edge between "kidney disease" and "uremia", then the relation type sequence of "uremia" is "causal association-causal association".

[0105] Add all newly identified second-level topological distance nodes to a queue or stack for later processing. Simultaneously, add these nodes to the set of visited nodes to prevent duplicate expansion.

[0106] Repeat the expansion process described above, processing nodes in the queue or stack until the queue or stack is empty, or the preset maximum topological distance limit is reached. For each node processed, remove it from the queue or stack and identify and process its lower-level adjacent nodes as described above.

[0107] In this way, extended sets of higher topological distance levels, such as the third and fourth, are formed sequentially. Each node in each level has a defined topological distance value and a record of the relationship type sequence path from the initial symptom node to that node.

[0108] During the expansion process, additional filtering conditions can be set according to actual application needs. For example, the maximum topological distance value can be limited to focus only on nodes whose topological distance from the initial symptom node does not exceed a certain threshold; or filtering can be performed based on the sequence of relation types to focus on analyzing paths with specific relation patterns.

[0109] Ultimately, a complete extended result set is obtained, containing all nodes reachable from the initial symptom node, along with the topological distance value and relation type sequence for each node. This information can be used for subsequent medical association analysis, such as disease transmission path prediction, potential complication identification, or drug side effect risk assessment.

[0110] In one optional implementation, the adapted patient subgraph is topologically aligned with the standard pattern graph; matching substructures in the adapted patient subgraph that are structurally consistent with the standard pattern graph are identified; missing condition items that exist in the standard pattern graph but are missing in the adapted patient subgraph are marked; and condition conflict items that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph are marked, including:

[0111] Using each standard entity node in the standard pattern diagram as an anchor point, a patient entity node that semantically matches the standard entity node is searched in the adapted patient subgraph. When a corresponding patient entity node is found,

[0112] Extract the first path type and standard topology connection structure of the standard entity node in the standard pattern graph; extract the second path type and patient topology connection structure of the patient entity node in the adapted patient subgraph;

[0113] When the first path type is completely consistent with the second path type and the standard topology connection structure is successfully matched with the patient topology connection structure, the patient entity node and the patient relationship path connected in the adapted patient subgraph are marked as a matching substructure;

[0114] Traverse the standard entity nodes. When a standard entity node does not find a semantically matching corresponding patient entity node in the adapted patient subgraph, mark the standard entity node and the standard relation path type connected in the standard pattern graph as a missing condition.

[0115] When the outgoing edge patient relationship path type of a certain patient entity node does not exist in the standard relationship path type set, the patient entity node and its outgoing edge patient relationship path type are marked as condition conflict items.

[0116] Obtain the standard schema diagram and the adapted patient subgraph. The standard schema diagram represents the standard diagnosis and treatment process in medical guidelines, and contains multiple standard entity nodes and standard relationship paths. The adapted patient subgraph represents the medical data of a specific patient, and contains multiple patient entity nodes and patient relationship paths.

[0117] During topology alignment, each standard entity node in the standard pattern graph serves as an anchor point, and semantically matching patient entity nodes are searched in the adapted patient subgraph. Semantic matching can be achieved through entity type and attribute similarity calculation. For example, if the standard entity node "diabetes" exists in the standard pattern graph, similar disease nodes are searched in the patient subgraph, and matching is performed by disease code or name.

[0118] For each patient entity node that finds a semantic match, extract the first path type and standard topological connection structure of the standard entity node in the standard pattern graph. The first path type refers to the set of relation types that can be reached from this node, such as "cause" and "treat" relation types; the standard topological connection structure is the actual way these relations are connected, including the relation direction and the target node of the connection.

[0119] Simultaneously, starting from the patient entity node, we extract its second path type and patient topological connection structure in the adapted patient subgraph. The second path type refers to the set of actual relation types existing in the patient data; the patient topological connection structure is the specific connection details of these relations.

[0120] Next, we compare whether the first path type and the second path type are completely identical, and check whether the standard topology connection structure and the patient's topology connection structure match successfully. Complete identity means that the set of relation types is exactly the same; a successful match requires that, in addition to the same type, the type of the target node and the direction of the relation are also consistent.

[0121] When the path types are completely identical and the topological connection structure matches successfully, the patient entity node and its patient relationship path connected in the adapted patient subgraph are marked as a matching substructure. For example, if the "diabetes" node in the standard schema graph is connected to the "insulin" node through the "treatment" relationship, and the same structure exists in the patient subgraph, then this part of the patient subgraph is marked as a matching substructure.

[0122] After traversing all standard entity nodes, standard entity nodes that could not find a semantically matching node in the adapted patient subgraph are processed. These standard entity nodes and the standard relation path types connecting them in the standard schema graph are marked as missing condition items. Missing condition items indicate that the patient data lacks certain elements that should be in the standard treatment plan, such as the lack of a certain necessary examination or medication record.

[0123] Examine the outgoing patient relationship path types of each patient entity node in the patient subgraph. When a patient relationship path type is found to be absent from the standard relationship path type set, mark the patient entity node and its outgoing patient relationship path types as conditional conflicts. Conditional conflicts represent medical behaviors that exist in the actual patient treatment process but are not specified in the standard protocol, which may be personalized treatment plans or potentially non-standard treatments.

[0124] This topological alignment method can clearly identify the consistency, missing items, and conflicts between a patient's treatment plan and standard medical guidelines, providing important references for subsequent clinical decisions. For example, for a hypertensive patient, the system may find that the patient's medication plan matches the standard treatment guidelines, but lacks records of regular blood pressure monitoring (missing item), and there are some medications used outside the standard plan (conflict item).

[0125] Furthermore, for multi-pathway treatment plans for complex diseases, recursive analysis of path matching at different depths can ensure accurate alignment even for complex network structures. This graph structure alignment method is particularly suitable for medical scenarios requiring precise comparison of patient treatment pathways with standard treatment guidelines, such as chronic disease management and standardized cancer treatment.

[0126] In one optional implementation, the coverage of the matching substructure, the missing condition terms, and the conflicting condition terms are normalized and weighted to calculate the patient's fit for the clinical research project, including:

[0127] Extract the topological adjacency matrix of the matching substructure and the standard adjacency matrix of the standard pattern graph, calculate the matrix similarity between the topological adjacency matrix and the standard adjacency matrix, and use the matrix similarity as the original term of the coverage degree of the matching substructure;

[0128] The connectivity of the standard entity node corresponding to each missing item in the missing condition is calculated in the standard pattern diagram. The connectivity is used as the missing influence factor of the missing item, and the missing influence factor is used as the original item of the missing condition.

[0129] The semantic distance between the patient entity node and the standard entity node corresponding to each conflict term in the conditional conflict terms is calculated. The semantic distance is used as the conflict intensity factor of the conflict term. The conflict intensity factors of all conditional conflict terms are summed to obtain the total conflict intensity factor. The total conflict intensity factor is used as the original term of the conditional conflict term.

[0130] After performing maximum and minimum value normalization on the original items of the coverage, the original items of the missing conditions, and the original items of the conflicting conditions, corresponding weight values ​​are assigned, and the fit of the patient to the clinical research project is calculated.

[0131] A standard schema diagram is constructed based on the standard screening criteria for clinical research projects. This diagram uses entity nodes as the core and describes the conditional requirements of clinical research projects through connections. During the patient-clinical research project matching analysis based on the standard schema diagram, the coverage of matching substructures, missing conditions, and conflicting conditions need to be normalized and weighted to calculate the patient's fit with the clinical research project.

[0132] A patient knowledge graph is constructed from patient data, containing entity nodes and their relationships, including patient health information, disease status, and medication history. Subsequently, the patient knowledge graph is matched with a standard schema graph to obtain matching substructures, missing conditions, and conflicting conditions.

[0133] When calculating patient fit for clinical research projects, the topological adjacency matrix of the matching substructure and the standard adjacency matrix of the standard pattern graph are extracted. The adjacency matrix is ​​a two-dimensional array representing the connectivity relationships between nodes in the graph. If there is a connection between node i and node j, the corresponding element in the matrix has a value of 1; otherwise, it has a value of 0. For weighted graphs, the matrix element values ​​can be the weights of the connectivity relationships. By calculating the similarity between these two adjacency matrices, the original term for the coverage of the matching substructure is obtained.

[0134] Matrix similarity can be calculated using the cosine similarity method. Two adjacency matrices are flattened into vectors A and B respectively. The calculation formula is: Similarity = Dot product of vectors A and B divided by the product of the magnitudes of vectors A and B. This similarity value ranges from 0 to 1; a higher value indicates a greater similarity between the matching substructure and the standard pattern diagram.

[0135] For the calculation of original items with missing conditions, it is necessary to calculate the connectivity of the standard entity node corresponding to each missing item in the standard schema graph. Connectivity represents the number of connections that node has with other nodes, reflecting the importance of that node in the graph. The higher the connectivity, the more important the node is in the standard schema graph, and the greater the negative impact of missing that node on fitness.

[0136] For example, in a clinical research project, if the standard entity node "HbA1c value" is connected to multiple nodes such as "diabetes diagnosis," "glycemic control effect," and "medication regimen" in the standard schema diagram, its connectivity is high. If this indicator is missing in the patient data, it will significantly affect its suitability.

[0137] For each conflicting condition, the semantic distance between the patient entity node and the standard entity node is calculated. This semantic distance can be calculated using an entity type hierarchy or a predefined semantic network. For example, the semantic distance between "type 2 diabetes" and "diabetes" is relatively small, while the semantic distance between "type 2 diabetes" and "hypertension" is relatively large.

[0138] Semantic distance can be calculated using medical ontology libraries, such as SNOMED CT or the ICD-10 coding system, by calculating the path distance of concepts within the ontology. A larger semantic distance indicates a higher degree of conflict. The original terms of the conditionally conflicting terms are obtained by summing the conflict intensity factors of all conflicting terms.

[0139] After obtaining the original terms for coverage, missing conditions, and conflicting conditions, these three terms are normalized to their maximum and minimum values. The normalization formula is: Normalized value = (Original value - Minimum value) / (Maximum value - Minimum value), which maps each value to the interval [0,1], making indicators of different dimensions comparable.

[0140] Finally, weights are assigned to the three normalized indicators, and a weighted sum is calculated to obtain the final fit of the patient to the clinical research project. The weighting can be adjusted according to the specific requirements of the clinical research project. Generally, the weight of coverage is higher, such as 0.6; the weight of missing conditions is moderate, such as 0.3; and the weight of conflicting conditions is lower, such as 0.1.

[0141] In practical applications, the weights of each indicator can be dynamically adjusted according to the characteristics of different types of clinical studies. For example, for early clinical trials with extremely high safety requirements, the weight of conflicting conditions can be increased; for real-world studies with lenient inclusion criteria, the weight of missing conditions can be decreased.

[0142] This computational method quantifies the match between patients and clinical research projects, providing researchers with objective evidence for selecting suitable subjects and improving the enrollment efficiency and quality of clinical trials. Furthermore, the method's flexibility makes it applicable to clinical research projects of different types and sizes.

[0143] The patient screening system based on artificial intelligence in clinical research according to embodiments of the present invention includes:

[0144] The first unit is used to obtain the inclusion criteria text of clinical research projects, perform semantic parsing on the inclusion criteria text, extract inclusion rule tuples containing condition types, constraint logic and medical concepts, and transform the inclusion rule tuples into a standard pattern diagram containing necessary entity nodes and relational paths.

[0145] The second unit is used to randomly select a symptom node in a pre-constructed medical knowledge graph as the starting point, and expand along the causal association edges and co-disease association edges in the medical knowledge graph. When the relationship type of the expansion path matches the relationship path type in the standard pattern graph, the node is retained to obtain an adapted patient subgraph. The medical knowledge graph is constructed based on the acquired medical record data of the patients to be matched through entity mapping.

[0146] The third unit is used to perform topological alignment between the adapted patient subgraph and the standard pattern graph, identify matching substructures in the adapted patient subgraph that are consistent with the structure of the standard pattern graph, mark missing condition items that exist in the standard pattern graph but are missing in the adapted patient subgraph, and mark condition conflict items that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph.

[0147] The fourth unit is used to normalize and weight the coverage of the matching substructure, the missing conditions and the conflicting conditions, calculate the fit of the patient to the clinical research project, and generate the patient's clinical research matching result based on the fit.

[0148] A third aspect of the present invention provides an electronic device, comprising:

[0149] processor;

[0150] Memory used to store processor-executable instructions;

[0151] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0152] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0153] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0154] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A patient screening method based on artificial intelligence in clinical research, characterized in that, include: Obtain the inclusion criteria text of the clinical research project, perform semantic parsing on the inclusion criteria text, extract the inclusion rule tuple containing condition types, constraint logic and medical concepts, and transform the inclusion rule tuple into a standard pattern diagram containing necessary entity nodes and relational paths. Starting from a pre-constructed medical knowledge graph, a disease node is randomly selected and expanded along the causal and co-disease association edges in the medical knowledge graph. When the relationship type of the expanded path matches the relationship path type in the standard pattern graph, the node is retained to obtain an adapted patient subgraph. The medical knowledge graph is constructed based on the acquired medical record data of the patients to be matched through entity mapping. The adapted patient subgraph is topologically aligned with the standard pattern graph. Matching substructures in the adapted patient subgraph that are consistent with the structure of the standard pattern graph are identified. Missing conditions that exist in the standard pattern graph but are missing in the adapted patient subgraph are marked. Conflicting conditions that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph are also marked. The coverage of the matching substructure, the missing conditions, and the conflicting conditions are normalized and weighted to calculate the patient's fit for the clinical research project. Based on the fit, the patient's clinical research matching results are generated, including: Extract the topological adjacency matrix of the matching substructure and the standard adjacency matrix of the standard pattern graph, calculate the matrix similarity between the topological adjacency matrix and the standard adjacency matrix, and use the matrix similarity as the original term of the coverage degree of the matching substructure; The connectivity of the standard entity node corresponding to each missing item in the missing condition is calculated in the standard pattern diagram. The connectivity is used as the missing influence factor of the missing item, and the missing influence factor is used as the original item of the missing condition. The semantic distance between the patient entity node and the standard entity node corresponding to each conflict term in the conditional conflict terms is calculated. The semantic distance is used as the conflict intensity factor of the conflict term. The conflict intensity factors of all conditional conflict terms are summed to obtain the total conflict intensity factor. The total conflict intensity factor is used as the original term of the conditional conflict term. After performing maximum and minimum value normalization on the original items of the coverage, the original items of the missing conditions, and the original items of the conflicting conditions, corresponding weight values ​​are assigned, and the fit of the patient to the clinical research project is calculated.

2. The method according to claim 1, characterized in that, Semantic parsing is performed on the inclusion standard text to extract inclusion rule tuples containing condition types, constraint logic, and medical concepts. These inclusion rule tuples are then transformed into a standard schema diagram containing necessary entity nodes and relational paths, including: The inclusion criteria text is parsed syntactically to identify conditional statements describing the inclusion conditions; For each conditional statement, semantic role labeling identifies condition type elements representing the nature of conditional constraints, constraint logic elements representing the logical combination relationship between multiple conditions, and medical concept elements representing clinical medical concepts. The condition type elements, constraint logic elements, and medical concept elements are then structurally combined according to the semantic dependency relationship in the conditional statement to form an ingress rule tuple. The medical concept elements in the grouping rule tuple are mapped to corresponding entity nodes. The relationship path type and connection method that need to be satisfied between the corresponding entity nodes are determined according to the constraint logic elements. The necessity level of the corresponding entity node and the relationship path type is marked according to the condition type elements. The standard pattern diagram is constructed based on the corresponding entity node, the relationship path type, the connection method, and the necessity level.

3. The method according to claim 1, characterized in that, Extending along the causal and comorbidity association edges in the medical knowledge graph, nodes are retained when the relationship type of the extended path matches the relationship path type in the standard pattern graph, resulting in an adapted patient subgraph including: Starting from the symptom node, the expansion process is layered according to topological distance. In each topological distance level, the expansion is carried out along the causal association edge and the co-disease association edge to the next level. The topological distance value from each expanded node to the symptom node and the sequence of relation types traversed in the expansion path are recorded. Extract the relationship path types and their topological location information in the standard pattern graph; perform sequence matching between the relationship type sequence of the extended nodes and the relationship path types in the standard pattern graph; and perform distance alignment between the topological distance value of the extended nodes and the topological location information of the corresponding relationship path in the standard pattern graph. The node is retained when the sequence of relation types of the extended node matches the relation path type in the standard pattern graph and the topological distance value is consistent with the topological location information. Based on the symptom nodes, the retained nodes, the relation type sequence, and the topological distance value, the adapted patient subgraph is constructed.

4. The method according to claim 3, characterized in that, Starting from the symptom node, the expansion process is layered according to topological distance. In each topological distance level, the expansion proceeds to the next level along causal and co-disease association edges. The topological distance value from each expanded node to the symptom node and the sequence of relation types traversed in the expansion path are recorded, including: The symptom node is marked as the initial level, and the topological distance value of the initial level is set to zero; Starting from the symptom node, traverse all outgoing edges of the symptom node, identify first-type adjacent nodes connected by causal association edges and second-type adjacent nodes connected by co-disease association edges, mark the first-type adjacent nodes and the second-type adjacent nodes as a first topological distance level, record a topological distance value of one for each node in the first topological distance level, record the relationship type sequence as causal association type for the first-type adjacent nodes, and record the relationship type sequence as co-disease association type for the second-type adjacent nodes; Starting from each node in the first topological distance level, traverse all its outgoing edges, identify the lower-level adjacent nodes connected by causal or co-causal edges, mark the lower-level adjacent nodes as the second topological distance level, record the topological distance value of each lower-level adjacent node in the second topological distance level as the topological distance value of its parent node plus one, and record the relationship type sequence of the lower-level adjacent node as the concatenation result of the relationship type sequence of its parent node and the relationship type of the current extended edge; Repeat the above expansion process until all nodes in all topological distance levels have been traversed, and obtain an expanded result set containing the topological distance values ​​and relation type sequences of all expanded nodes.

5. The method according to claim 1, characterized in that, The adapted patient subgraph is topologically aligned with the standard pattern graph. Matching substructures in the adapted patient subgraph that are consistent with the structure of the standard pattern graph are identified. Missing conditions that exist in the standard pattern graph but are missing in the adapted patient subgraph are marked. Condition conflict conditions that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph are marked, including: Using each standard entity node in the standard pattern diagram as an anchor point, a patient entity node that semantically matches the standard entity node is searched in the adapted patient subgraph. When a corresponding patient entity node is found, Extract the first path type and standard topology connection structure of the standard entity node in the standard pattern graph; extract the second path type and patient topology connection structure of the patient entity node in the adapted patient subgraph; When the first path type is completely consistent with the second path type and the standard topology connection structure is successfully matched with the patient topology connection structure, the patient entity node and the patient relationship path connected in the adapted patient subgraph are marked as a matching substructure; Traverse the standard entity nodes. When a standard entity node does not find a semantically matching corresponding patient entity node in the adapted patient subgraph, mark the standard entity node and the standard relation path type connected in the standard pattern graph as a missing condition. When the outgoing edge patient relationship path type of a certain patient entity node does not exist in the standard relationship path type set, the patient entity node and its outgoing edge patient relationship path type are marked as condition conflict items.

6. A patient screening system based on artificial intelligence in clinical research, used to implement the method as described in any one of claims 1-5, characterized in that, include: The first unit is used to obtain the inclusion criteria text of clinical research projects, perform semantic parsing on the inclusion criteria text, extract inclusion rule tuples containing condition types, constraint logic and medical concepts, and transform the inclusion rule tuples into a standard pattern diagram containing necessary entity nodes and relational paths. The second unit is used to randomly select a symptom node in a pre-constructed medical knowledge graph as the starting point, and expand along the causal association edges and co-disease association edges in the medical knowledge graph. When the relationship type of the expansion path matches the relationship path type in the standard pattern graph, the node is retained to obtain an adapted patient subgraph. The medical knowledge graph is constructed based on the acquired medical record data of the patients to be matched through entity mapping. The third unit is used to perform topological alignment between the adapted patient subgraph and the standard pattern graph, identify matching substructures in the adapted patient subgraph that are consistent with the structure of the standard pattern graph, mark missing condition items that exist in the standard pattern graph but are missing in the adapted patient subgraph, and mark condition conflict items that exist in the adapted patient subgraph but exceed the requirements of the standard pattern graph. The fourth unit is used to normalize and weight the coverage of the matching substructure, the missing conditions and the conflicting conditions, calculate the fit of the patient to the clinical research project, and generate the patient's clinical research matching result based on the fit.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images