A disease diagnosis method and system based on a knowledge graph

Abstract

The application discloses a disease diagnosis method and system based on a knowledge graph, comprising the following steps: obtaining medical data and completing desensitization processing to obtain a standardized evidence set; constructing a patient context subgraph, performing entity alignment and relationship fusion to obtain a diagnosis graph; performing phased candidate disease reasoning to obtain a candidate disease result; performing candidate disease pairing, hypergraph construction and injection sequence generation operations to obtain a hypergraph and a discriminative hyperedge injection sequence; performing Berge minimum cross-section set incremental enumeration on the hypergraph to obtain a minimum cross-section set; constructing a discriminative suggestion item, extracting a corresponding relationship associated evidence path to generate an evidence path set; performing formatting and sorting on the candidate disease result to obtain a diagnosis result, and performing formatting and association binding on the evidence path set to obtain a discriminative suggestion result. The application improves the accuracy of disease diagnosis through Berge minimum cross-section set incremental enumeration.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and in particular to a disease diagnosis method and system based on knowledge graphs. Background Technology

[0002] In recent years, intelligent assisted diagnostic technologies based on electronic medical records, laboratory test results, and consultation records have developed rapidly. Common solutions include rule-based expert systems, statistical learning-based classification models, and deep learning-based end-to-end inference models. Meanwhile, medical knowledge graphs are used to organize medical entities such as diseases, symptoms, signs, examinations, and treatments, and their relationships, to support reasoning, retrieval, and interpretable analysis. In existing technologies, one approach directly inputs medical record text or structured data into the model to output disease ranking results. This relies on the coverage of training data and the model's parameter fitting ability, and is prone to generating unstable candidate sets when encountering cross-institutional data distribution differences, missing evidence, or noisy information.

[0003] In the differential diagnosis stage, existing technologies generally employ fixed procedures or heuristic strategies to recommend examination and consultation items. For example, they select the next examination based on disease similarity or information gain, or output several differential points based on preset rules. Such methods often struggle to establish strict distinguishing constraints between candidate diseases, easily leading to redundant recommendations, uncontrollable costs and delays, and low efficiency in distinguishing specific candidate disease pairs. More importantly, existing solutions typically do not use a "minimum set of differential diagnoses" as a clear computational objective, lacking a mechanism for systematically enumerating and filtering the minimum set of examinations or evidence under the constraint of differentiating candidate disease pairs. This results in output differential recommendations that are difficult to guarantee in terms of minimization and completeness, leading to insufficient clinical feasibility.

[0004] Therefore, how to provide a disease diagnosis method and system based on knowledge graphs is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a disease diagnosis method based on a knowledge graph. This invention standardizes medical data into an evidence set based on a medical knowledge graph, constructs and merges a patient context subgraph with the medical knowledge graph to obtain a diagnostic graph, outputs candidate disease results through phased candidate disease reasoning, and constructs a hypergraph and differential hyperedge injection sequences based on candidate disease pairings. Berge's minimum cross-section set incremental enumeration is used to obtain a minimum cross-section set to generate minimum differential suggestion entries. Evidence paths are extracted from the diagnostic graph and associated with differential suggestions for output, reducing redundant checks and improving diagnostic accuracy and interpretability.

[0006] A disease diagnosis method and system based on knowledge graphs according to an embodiment of the present invention includes the following steps:

[0007] The medical records of the individuals to be diagnosed are obtained and anonymized. At the same time, structured and semantic normalization processes are performed to obtain a standardized set of evidence.

[0008] A patient context subgraph is constructed based on a standardized set of evidence, and entity alignment and relationship fusion are performed using a medical knowledge graph to obtain a diagnostic graph.

[0009] Based on the diagnostic map and standardized evidence set, a phased candidate disease inference is performed to obtain candidate disease results;

[0010] Receive candidate disease results and diagnostic maps, perform candidate disease pairing, hypergraph construction, and injection sequence generation operations to obtain the hypergraph and the identification hyperedge injection sequence;

[0011] Based on the discriminative superedge injection sequence, Berge minimum cross set incremental enumeration is performed on the hypergraph. Discriminative superedges are added to the hypergraph in sequence. After each discriminative superedge is added, the current cross set is updated and non-minimum cross sets are deleted to obtain the minimum cross set set.

[0012] Based on the minimum cross set, identification suggestion items are constructed one by one, and a correspondence with the identification superedge is established. Based on the diagnostic map, the evidence paths associated with the corresponding relationships are extracted to generate an evidence path set.

[0013] The candidate disease results are formatted and sorted to obtain the diagnostic results, and the evidence path set is formatted and associated to obtain the identification suggestion results.

[0014] Optionally, obtaining the standardized evidence set specifically includes:

[0015] Receive the medical data of the object to be diagnosed, perform field parsing and record segmentation operations on the medical data, obtain the medical record set and use it as the original medical dataset;

[0016] Receive the original medical visit dataset, perform desensitization processing, the desensitization processing includes identifying the identity information field, contact information field, and address field and generating an anonymous identifier, using the anonymous identifier to replace the content of the identity information field, setting the content of the contact information field and address field to empty, and writing the desensitized medical visit dataset into the desensitized medical visit dataset table.

[0017] The anonymized medical records dataset is subjected to structured processing, and semantic normalization processing is performed to obtain a standardized evidence set, which includes standard entity identifiers and evidence attributes.

[0018] Optionally, obtaining the diagnostic map specifically includes:

[0019] Receive a standardized set of evidence, write the standard entity identifier into the patient node set, write the evidence attributes into the node attributes, generate relation edges based on the evidence attributes and write them into the corresponding edge attributes to obtain the patient context subgraph;

[0020] It receives existing medical knowledge sources and performs extraction and standardization processes to obtain a medical knowledge graph that includes knowledge nodes;

[0021] Read the standard entity identifier of the patient context subgraph node, retrieve the knowledge node with the same standard entity identifier in the medical knowledge graph node set, and write the patient node and knowledge node into the alignment mapping set.

[0022] The patient node set and the knowledge node set are merged to obtain the diagnosis node set, and the patient relationship set and the knowledge relationship set are merged to obtain the diagnosis relationship set. Alignment association edges are established based on the alignment mapping set, and the node attributes and relationship attributes are merged and written to obtain the diagnosis graph.

[0023] Optionally, obtaining the candidate disease results specifically includes:

[0024] Standard entity identifiers and evidence attributes are extracted from the standardized evidence set. Based on the standard entity identifiers, the corresponding nodes are retrieved and located in the diagnostic map to form an evidence node set. The evidence node set is associated with the evidence attributes and written into the evidence index table.

[0025] Using the set of evidence nodes in the evidence index table as the starting point of the traversal, a neighborhood traversal with a limited number of jumps is performed in the diagnostic graph according to the preset traversal step size. The disease nodes reached during the traversal are recorded and written into the candidate disease table. The candidate disease table is deduplicated to form a candidate disease set.

[0026] For each candidate disease in the candidate disease set, the relationship edges and path records between the candidate disease and the evidence node set are retrieved in the diagnostic graph. Based on the evidence attributes in the evidence index table, supporting evidence and negative evidence are weighted respectively. The association strength and conflict strength between the candidate disease and the standard entity identifier are calculated based on the relationship edge attributes and path records. The candidate disease score is calculated based on the supporting evidence weight, negative evidence weight, association strength and conflict strength. The candidate disease set is sorted according to the candidate disease score to obtain the candidate disease results.

[0027] Optionally, obtaining the hypergraph and the discriminative hyperedge injection sequence specifically includes:

[0028] Extract the candidate disease set and candidate disease score from the candidate disease results, and build a candidate disease index table in the diagnostic map;

[0029] Perform pairwise pairing operations on the candidate disease set in the candidate disease index table to generate a candidate disease pair set and write it into the candidate disease pair table. The candidate disease pair table records the first candidate disease identifier and the second candidate disease identifier contained in each candidate disease pair.

[0030] In the diagnostic graph, for each candidate disease pair in the candidate disease pair table, a differential diagnosis item retrieval operation is performed to extract the examination item identifier and the consultation item identifier to form a differential diagnosis item set. The differential diagnosis item set is written into the differential diagnosis superedge and a corresponding record between the differential diagnosis superedge and the candidate disease pair is established. The results are then summarized to form a differential diagnosis superedge set.

[0031] Based on the candidate disease pair table and the candidate disease index table, obtain the scores of the two candidate diseases for each candidate disease pair, calculate the confusion degree according to the score similarity and write it into the confusion degree table, sort the identification superedge set according to the confusion degree table to generate the identification superedge injection sequence, and write it into the injection sequence table.

[0032] The vertex set is obtained by deduplicating the identification items in the identification hyperedge set and summarizing them. The identification hyperedge set is written into the hyperedge set, and a hypergraph is constructed based on the vertex set and the hyperedge set.

[0033] Optionally, obtaining the set of minimum cross sections specifically includes:

[0034] Read the hypergraph and the discrimination hyperedge injection sequence, perform Berge minimum cross set incremental enumeration, assign a unique number to each discrimination item appearing in the hypergraph and generate a discrimination item encoding table, perform binary encoding on each discrimination hyperedge in the discrimination hyperedge injection sequence to generate a hyperedge encoding, select the first discrimination hyperedge in the discrimination hyperedge injection sequence, and construct a single-element cross set encoding for each position of the first discrimination hyperedge, summarize to obtain the initial cross set set and write it into the current cross set set table;

[0035] Select the superedges to be injected in sequence according to the superedge injection sequence and read the corresponding superedge codes. Write the superedges to be injected into the superedge set of the supergraph and update the supergraph. At the same time, perform a bitwise AND operation between each cross set code in the current cross set set table and the superedge code. Determine whether the operation result has a set bit. If the operation result has a set bit, write the cross set code into the hit set. If the operation result does not have a set bit, write the cross set code into the miss set.

[0036] Perform an expansion generation operation on the miss set, traverse all the set positions of the superedge code and extract the corresponding identification item number one by one, set the identification item number and write it into the cross set code in the miss set to obtain the expanded cross set code, write the expanded cross set code into the candidate cross set set and write it into the candidate index table, the candidate index table records the number of set positions of the expanded cross set code and records the identification superedge number marker corresponding to the expanded cross set code;

[0037] Merging the hit set and the candidate crosscut set sets yields a merged crosscut set set. A non-minimum deletion operation is then performed on the merged crosscut set set. The merged crosscut set set is sorted by the number of bits set and the crosscut set codes are traversed sequentially. The current crosscut set code and the retained crosscut set codes are compared one by one for inclusion determination. The inclusion determination includes performing a bitwise AND operation on the two crosscut set codes and determining whether the bitwise AND result is equal to the crosscut set code with the smaller number of bits set. If inclusion is found to be true, the crosscut set code with the larger number of bits set is deleted and the crosscut set code with the smaller number of bits set is retained. An updated crosscut set set is generated and written to the current crosscut set set table. The crosscut set code with the larger number of bits set is the non-minimum crosscut set.

[0038] Maintain a minimum size upper bound and perform size pruning. Write the minimum size upper bound into the upper bound record table. When generating the extended cross set code, read the upper bound record table and compare the number of bits set in the extended cross set code with the minimum size upper bound. If the number of bits set in the extended cross set code is greater than the minimum size upper bound, delete the extended cross set code and prohibit writing it into the candidate cross set set. After completing the writing to the current cross set set table, read the current cross set set table and update the upper bound record table with the minimum number of bits set. Determine the minimum cross set set using the cross set set in the current cross set set table.

[0039] Optionally, the generation of the evidence path set specifically includes:

[0040] Based on the minimum cross set, the identification item identifiers corresponding to the minimum cross set are extracted one by one. The identification item identifiers are sorted according to the number order in the identification item coding table to generate identification suggestion entries and assign identification suggestion entry identifiers. The identification suggestion entries are summarized to form an identification suggestion set.

[0041] For each identification suggestion item, the identification superedge set is searched one by one. The identification superedge identifier containing the identification item identifier is located. The identification suggestion item identifier, identification item identifier, and identification superedge identifier are written into the corresponding relationship table to generate the corresponding relationship record.

[0042] Based on the corresponding relationship records, locate the candidate disease pair identifiers, locate the first candidate disease identifier and the second candidate disease identifier, locate the differential item node in the diagnostic graph based on the differential item identifier, retrieve the relationship path from the first candidate disease node and the second candidate disease node to the differential item node in the diagnostic graph, and combine the node sequence of the relationship path with the relationship sequence as the evidence path set.

[0043] Optionally, obtaining the diagnostic results and differential diagnosis recommendations specifically includes:

[0044] Based on the candidate disease results, extract the candidate disease set and candidate disease score, sort the candidate disease set from high to low according to the candidate disease score and generate a sorting rank for each candidate disease, and write the candidate disease identifier, candidate disease score and sorting rank into the diagnosis result table.

[0045] Based on the identification suggestion table, the corresponding relationship table and the evidence path set, perform the association binding operation, locate the corresponding relationship record in the corresponding relationship table according to the identification suggestion item identifier, locate the associated evidence path record in the evidence path set, and write the identification suggestion item identifier, identification item identifier, identification super-edge identifier and evidence path record into the identification suggestion output table.

[0046] Perform formatting operations on the diagnostic results table and differential diagnosis suggestion output table, and output the diagnostic results and differential diagnosis suggestion results.

[0047] A knowledge graph-based disease diagnosis system according to an embodiment of the present invention includes:

[0048] The evidence standardization module is used to acquire medical data and perform desensitization, structuring, and semantic normalization processing to generate a standardized evidence set.

[0049] The diagnostic graph fusion module is used to construct a patient context subgraph based on a standardized evidence set and write attributes, construct a medical knowledge graph and perform entity alignment and relation fusion to generate a diagnostic graph.

[0050] The candidate reasoning module is used to locate evidence nodes based on the diagnostic map and standardized evidence set, perform neighborhood traversal to obtain a set of candidate diseases, and generate candidate disease results.

[0051] The hypergraph construction module is used to perform pairing to generate candidate disease pairs based on candidate disease results, calculate confusion degree and sort to generate discriminative hyperedge injection sequences;

[0052] The Berge enumeration module is used to add discriminative superedges to the hypergraph sequentially according to the discriminative superedge injection sequence and perform Berge minimum cross set incremental enumeration to generate a minimum cross set set.

[0053] The evidence path module is used to generate identification suggestion entries based on the minimum cross set and establish the correspondence with the identification superedge, thereby generating the evidence path set.

[0054] The results output module is used to sort candidate disease results, perform association binding between identification suggestion items and evidence path sets, and output diagnostic results and identification suggestion results.

[0055] The beneficial effects of this invention are:

[0056] This invention achieves a unified output link for disease diagnosis and differential diagnosis through the collaborative design of evidence standardization based on medical knowledge graph, diagnostic graph fusion reasoning, and hypergraph minimum cross set enumeration. It can stably form a standardized evidence set that can be used for reasoning and construct a diagnostic graph even when there is noise, inconsistency in expression, and scattered evidence in the medical data. This improves the consistency of evidence and the reliability of candidate disease ranking in the candidate disease reasoning stage.

[0057] This invention explicitly models the differentiation requirements of candidate disease pairs as a differentiation hyperedge and constructs a hypergraph. It uses Berge's minimum cross set incremental enumeration to obtain the minimum cross set set, so that the differentiation suggestion items cover the differentiation constraints of candidate disease pairs in the form of a minimum set, reducing the input of repetitive and redundant examination items and consultation items, improving the efficiency of differential diagnosis and enhancing clinical feasibility.

[0058] This invention extracts the evidence path associated with the corresponding relationship on the diagnostic map and binds it to the identification suggestion result for output, so that both the diagnostic result and the identification suggestion have a traceable evidence link, improving the interpretability and credibility of doctors' review and adoption, thereby improving the overall accuracy of disease diagnosis and the quality of treatment decision-making. Attached Figure Description

[0059] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0060] Figure 1 This is a flowchart of a knowledge graph-based disease diagnosis method proposed in this invention;

[0061] Figure 2 This is a schematic diagram of a knowledge graph-based disease diagnosis system proposed in this invention.

[0062] Figure 3 This is a schematic diagram of Berge's minimum cross set incremental enumeration algorithm for a knowledge graph-based disease diagnosis method proposed in this invention. Detailed Implementation

[0063] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0064] refer to Figures 1-3 A disease diagnosis method based on knowledge graphs includes the following steps:

[0065] The medical data of the person to be diagnosed is obtained and desensitized. At the same time, structured and semantic normalization processing is performed to obtain a standardized evidence set, which includes standard entity identifiers and evidence attributes.

[0066] A patient context subgraph is constructed based on a standardized evidence set. The patient context subgraph uses standard entity identifiers as nodes and evidence attributes to form node attributes and edge attributes. Entity alignment and relationship fusion are performed in conjunction with a medical knowledge graph to obtain a diagnostic graph.

[0067] Based on the diagnostic map and standardized evidence set, a phased candidate disease inference is performed to obtain candidate disease results, which include a candidate disease set and a candidate disease score.

[0068] Receive candidate disease results and diagnostic maps, perform candidate disease pairing, hypergraph construction, and injection sequence generation operations to obtain the hypergraph and the identification hyperedge injection sequence;

[0069] Based on the discriminative superedge injection sequence, Berge minimum cross set incremental enumeration is performed on the hypergraph. Discriminative superedges are added to the hypergraph in sequence. After each discriminative superedge is added, the current cross set is updated and non-minimum cross sets are deleted to obtain the minimum cross set set.

[0070] Based on the minimum cross set, identification suggestion items are constructed one by one, and a correspondence with the identification superedge is established. Based on the diagnostic map, the evidence paths associated with the corresponding relationships are extracted to generate an evidence path set.

[0071] The candidate disease results are formatted and sorted to obtain the diagnostic results, and the evidence path set is formatted and associated to obtain the identification suggestion results. The diagnostic results include the disease identifiers and corresponding sorting information identified in the candidate disease results, and the identification suggestion results include identification suggestions for distinguishing candidate diseases and evidence path information corresponding to the identification suggestions.

[0072] In this embodiment, obtaining the standardized evidence set specifically includes:

[0073] Receive the medical data of the patient to be diagnosed, perform field parsing and record segmentation operations on the medical data to obtain a set of medical records as the original medical data set. The set of medical records includes chief complaint records, present medical history records, physical examination records, laboratory examination records, and imaging conclusion records.

[0074] Receive the original medical visit dataset, perform desensitization processing, the desensitization processing includes identifying the identity information field, contact information field, and address field and generating an anonymous identifier, using the anonymous identifier to replace the content of the identity information field, setting the content of the contact information field and address field to empty, and writing the desensitized medical visit dataset into the desensitized medical visit dataset table.

[0075] The desensitized medical records dataset undergoes structured processing and semantic normalization to obtain a standardized evidence set. The structured processing includes medical entity recognition, medical entity attribute extraction, negation information recognition, time information recognition, and degree information recognition, forming evidence items and writing evidence attributes into them. The semantic normalization process includes performing candidate standard entity retrieval, similarity calculation, candidate entity selection, and standard entity identifier writing operations on medical entity text based on a preset medical terminology system. The standard entity identifier and evidence attributes are then written into the standardized evidence set. The standardized evidence set includes standard entity identifiers and evidence attributes. The standard entity identifier is a unique identifier obtained after semantic normalization of medical entities in the medical records data, used to point to the corresponding entity node in the medical terminology system, medical ontology, or medical knowledge graph. The evidence attributes are a set of structured attributes corresponding to the standard entity identifier, used to characterize the evidence features of the medical entity in the medical records data.

[0076] In this embodiment, obtaining the diagnostic map specifically includes:

[0077] Receive a standardized evidence set, write standard entity identifiers into the patient node set, write evidence attributes into node attributes, generate relational edges based on evidence attributes and write corresponding edge attributes. The generation of relational edges includes generating time sequence edges based on time information, generating negation edges based on negation information, generating degree edges based on degree information, and generating co-occurrence edges based on same visit information, thus obtaining a patient context subgraph.

[0078] The system receives existing medical knowledge sources and performs extraction and standardization processing to obtain a medical knowledge graph including knowledge nodes. The extraction and standardization processing includes performing sentence segmentation, terminology segmentation, entity recognition, and relation extraction on text knowledge sources, performing field reading, entity extraction, and relation extraction on structured knowledge sources, performing standardization mapping on the extraction results to obtain a set of triples, writing the head entity identifier and tail entity identifier into the knowledge node set, writing the relation type into the knowledge relation set and connecting the corresponding nodes, and writing the source identifier and confidence identifier.

[0079] Read the standard entity identifier of the patient context subgraph node, retrieve the knowledge node with the same standard entity identifier in the medical knowledge graph node set, and write the patient node and knowledge node into the alignment mapping set.

[0080] The patient node set and the knowledge node set are merged to obtain the diagnosis node set, and the patient relationship set and the knowledge relationship set are merged to obtain the diagnosis relationship set. Alignment association edges are established based on the alignment mapping set, and the node attributes and relationship attributes are merged and written to obtain the diagnosis graph.

[0081] In this embodiment, obtaining the candidate disease result specifically includes:

[0082] Standard entity identifiers and evidence attributes are extracted from the standardized evidence set. Based on the standard entity identifiers, the corresponding nodes are retrieved and located in the diagnostic map to form an evidence node set. The evidence node set is associated with the evidence attributes and written into the evidence index table.

[0083] Using the set of evidence nodes in the evidence index table as the starting point of the traversal, a neighborhood traversal with a limited number of jumps is performed in the diagnostic graph according to the preset traversal step size. The disease nodes reached during the traversal are recorded and written into the candidate disease table. The candidate disease table is deduplicated to form a candidate disease set.

[0084] For each candidate disease in the candidate disease set, the relationship edges and path records between the candidate disease and the evidence node set are retrieved in the diagnostic graph. Supporting evidence and negative evidence are weighted according to the evidence attributes in the evidence index table. The association strength and conflict strength between the candidate disease and the standard entity identifier are calculated based on the relationship edge attributes and path records. The candidate disease score is calculated based on the supporting evidence weight, negative evidence weight, association strength and conflict strength. The candidate disease set is sorted according to the candidate disease score to obtain the candidate disease results. The supporting evidence is the standard entity identifier corresponding to the information in the standardized evidence set whose evidence attributes contain the positive mark. The negative evidence is the standard entity identifier corresponding to the information in the standardized evidence set whose evidence attributes contain the negative mark.

[0085] In this embodiment, obtaining the hypergraph and the discrimination hyperedge injection sequence specifically includes:

[0086] Extract the candidate disease set and candidate disease score from the candidate disease results, and establish a candidate disease index table in the diagnostic map. The candidate disease index table records the correspondence between candidate disease identifiers and candidate disease scores.

[0087] Perform pairwise pairing operations on the candidate disease set in the candidate disease index table to generate a candidate disease pair set and write it into the candidate disease pair table. The candidate disease pair table records the first candidate disease identifier and the second candidate disease identifier contained in each candidate disease pair.

[0088] In the diagnostic graph, for each candidate disease pair in the candidate disease pair table, a differential diagnosis item retrieval operation is performed to extract the examination item identifier and the consultation item identifier to form a differential diagnosis item set. The differential diagnosis item set is written into the differential diagnosis superedge and a corresponding record between the differential diagnosis superedge and the candidate disease pair is established. The results are then summarized to form a differential diagnosis superedge set.

[0089] Based on the candidate disease pair table and the candidate disease index table, obtain the scores of the two candidate diseases for each candidate disease pair, calculate the confusion degree according to the score similarity and write it into the confusion degree table, sort the identification superedge set according to the confusion degree table to generate the identification superedge injection sequence, and write it into the injection sequence table.

[0090] The vertex set is obtained by deduplicating the identification items in the identification hyperedge set and summarizing them. The identification hyperedge set is written into the hyperedge set, and a hypergraph is constructed based on the vertex set and the hyperedge set.

[0091] This invention structures candidate disease results into an indexable candidate disease pairing space and retrieves them from the diagnostic graph to form a set of discriminative hyperedges. It generates a confusion degree based on the proximity of candidate disease scores and sorts them to obtain a sequence of discriminative hyperedge injections. Furthermore, it constructs a hypergraph by deduplicating and summarizing discriminative items, so that the discriminative constraints have a computable hypergraph representation and priority injection order. This improves the targeting and consistency of discriminative item retrieval and provides a more stable input for subsequent minimum cross set enumeration, thereby reducing invalid searches and redundant discriminatives.

[0092] In this embodiment, obtaining the minimum cross-section set specifically includes:

[0093] Read the hypergraph and the discrimination hyperedge injection sequence, perform Berge minimum cross set incremental enumeration, assign a unique number to each discrimination item appearing in the hypergraph and generate a discrimination item encoding table, perform binary encoding on each discrimination hyperedge in the discrimination hyperedge injection sequence to generate a hyperedge encoding, select the first discrimination hyperedge in the discrimination hyperedge injection sequence, and construct a single-element cross set encoding for each position of the first discrimination hyperedge, summarize to obtain the initial cross set set and write it into the current cross set set table. The binary encoding uses the discrimination item number position as the position, and the cross set is represented by the same binary encoding and uses the discrimination item number position contained in the cross set as the position.

[0094] Select the superedges to be injected in sequence according to the superedge injection sequence and read the corresponding superedge codes. Write the superedges to be injected into the superedge set of the supergraph and update the supergraph. At the same time, perform a bitwise AND operation between each cross set code in the current cross set set table and the superedge code. Determine whether the operation result has a set bit. If the operation result has a set bit, write the cross set code into the hit set. If the operation result does not have a set bit, write the cross set code into the miss set.

[0095] Perform an expansion generation operation on the miss set, traverse all the set positions of the superedge code and extract the corresponding identification item number one by one, set the identification item number and write it into the cross set code in the miss set to obtain the expanded cross set code, write the expanded cross set code into the candidate cross set set and write it into the candidate index table, the candidate index table records the number of set positions of the expanded cross set code and records the identification superedge number marker corresponding to the expanded cross set code;

[0096] Merging the hit set and the candidate crosscut set sets yields a merged crosscut set set. A non-minimum deletion operation is then performed on the merged crosscut set set. The merged crosscut set set is sorted by the number of bits set and the crosscut set codes are traversed sequentially. The current crosscut set code and the retained crosscut set codes are compared one by one for inclusion determination. The inclusion determination includes performing a bitwise AND operation on the two crosscut set codes and determining whether the bitwise AND result is equal to the crosscut set code with the smaller number of bits set. If inclusion is found to be true, the crosscut set code with the larger number of bits set is deleted and the crosscut set code with the smaller number of bits set is retained. An updated crosscut set set is generated and written to the current crosscut set set table. The crosscut set code with the larger number of bits set is the non-minimum crosscut set.

[0097] Maintain a minimum size upper bound and perform size pruning operations. The minimum size upper bound is determined by the minimum number of bits set in the current cross set set table. Write the minimum size upper bound into the upper bound record table. When generating extended cross set codes, read the upper bound record table and compare the number of bits set in the extended cross set codes with the minimum size upper bound. If the number of bits set in the extended cross set codes is greater than the minimum size upper bound, delete the extended cross set codes and prohibit writing them into the candidate cross set set. After completing the writing to the current cross set set table, read the current cross set set table and update the upper bound record table with the minimum number of bits set. Determine the minimum cross set set using the cross set sets in the current cross set set table.

[0098] This invention expresses the discrimination constraints of candidate disease pairs in the form of a hypergraph and performs Berge minimum cross set incremental enumeration based on the discrimination hyperedge injection sequence. It uses binary encoding and bitwise operations to realize cross set hit determination, candidate expansion and inclusion removal while maintaining the minimumity of the cross set. Combined with the size pruning of the minimum size upper bound, it reduces the enumeration search space and computational overhead. Thus, while ensuring that the discrimination suggestions cover the discrimination requirements of candidate diseases, it outputs a smaller set of discrimination items and improves discrimination efficiency and real-time performance.

[0099] In this embodiment, the generation of the evidence path set specifically includes:

[0100] Based on the minimum cross set, the identification item identifiers corresponding to the minimum cross set are extracted one by one. The identification item identifiers are sorted according to the number order in the identification item coding table to generate identification suggestion entries and assign identification suggestion entry identifiers. The identification suggestion entries are summarized to form an identification suggestion set.

[0101] For each identification suggestion item, the identification superedge set is searched one by one. The identification superedge identifier containing the identification item identifier is located. The identification suggestion item identifier, identification item identifier, and identification superedge identifier are written into the corresponding relationship table to generate the corresponding relationship record.

[0102] Based on the corresponding relationship records, locate the candidate disease pair identifiers, locate the first candidate disease identifier and the second candidate disease identifier, locate the differential item node in the diagnostic graph based on the differential item identifier, retrieve the relationship path from the first candidate disease node and the second candidate disease node to the differential item node in the diagnostic graph, and combine the node sequence of the relationship path with the relationship sequence as the evidence path set.

[0103] In this embodiment, obtaining the diagnostic results and differential diagnosis recommendations specifically includes:

[0104] Based on the candidate disease results, extract the candidate disease set and candidate disease score, sort the candidate disease set from high to low according to the candidate disease score and generate a sorting rank for each candidate disease, and write the candidate disease identifier, candidate disease score and sorting rank into the diagnosis result table.

[0105] Based on the identification suggestion table, the corresponding relationship table and the evidence path set, perform the association binding operation, locate the corresponding relationship record in the corresponding relationship table according to the identification suggestion item identifier, locate the associated evidence path record in the evidence path set, and write the identification suggestion item identifier, identification item identifier, identification super-edge identifier and evidence path record into the identification suggestion output table.

[0106] Perform a formatting operation on the diagnostic result table and the differential diagnosis suggestion output table. The formatting operation includes generating the output field order for the diagnostic result table and the output field order for the differential diagnosis suggestion output table. The field order satisfies the following conditions: the field order of the diagnostic result table is candidate disease identifier, candidate disease score, and sorting position; the field order of the differential diagnosis suggestion output table is differential diagnosis suggestion item identifier, differential diagnosis item identifier, and evidence path record. Output the diagnostic result and differential diagnosis suggestion result.

[0107] A knowledge graph-based disease diagnosis system includes:

[0108] The evidence standardization module is used to acquire medical data and perform desensitization, structuring, and semantic normalization processing to generate a standardized evidence set.

[0109] The diagnostic graph fusion module is used to construct a patient context subgraph based on a standardized evidence set and write attributes, construct a medical knowledge graph and perform entity alignment and relation fusion to generate a diagnostic graph.

[0110] The candidate reasoning module is used to locate evidence nodes based on the diagnostic map and standardized evidence set, perform neighborhood traversal to obtain a set of candidate diseases, and generate candidate disease results.

[0111] The hypergraph construction module is used to perform pairing to generate candidate disease pairs based on candidate disease results, calculate confusion degree and sort to generate discriminative hyperedge injection sequences;

[0112] The Berge enumeration module is used to add discriminative superedges to the hypergraph sequentially according to the discriminative superedge injection sequence and perform Berge minimum cross set incremental enumeration to generate a minimum cross set set.

[0113] The evidence path module is used to generate identification suggestion entries based on the minimum cross set and establish the correspondence with the identification superedge, thereby generating the evidence path set.

[0114] The results output module is used to sort candidate disease results, perform association binding between identification suggestion items and evidence path sets, and output diagnostic results and identification suggestion results.

[0115] Example 1: To verify the feasibility of this invention in practice, it was applied to the auxiliary diagnosis and differential diagnosis of common internal medicine diseases in outpatient and emergency departments. In this scenario, doctors face complex sources of patient data, inconsistent descriptions, scattered evidence, and high levels of confusion among candidate diseases. Common problems include unstable ranking of candidate diseases, redundant recommendations of differential examinations, and a lack of traceable evidence paths for recommended items, leading to increased time consumption in clinical decision-making and difficulty in verifying interpretations. This example selects anonymized patient data from real business processes as input samples. The data covers various record formats such as chief complaint, present medical history, physical examination, laboratory tests, and imaging conclusions, and simultaneously includes structured fields and free text. The data contains common noise such as synonyms, abbreviations, negative descriptions, and time-span descriptions, which can realistically reflect the core pain points of difficulty in evidence standardization, difficulty in distinguishing candidate diseases, and missing interpretation chains.

[0116] During application, after the system receives patient data, it first performs desensitization and structured parsing. Symptom descriptions from the chief complaint and present medical history, physical examination findings, laboratory test results and indicators, and key findings from imaging conclusions are extracted as evidence items. Semantic normalization is then performed on these evidence items to obtain a standardized evidence set. The standard entity identifiers in the standardized evidence set directly point to entity nodes in the medical terminology system and medical knowledge graph. Evidence attributes record negation markers, time information, degree information, numerical information, unit information, source information, etc., ensuring that the same medical concept can be uniformly aligned under different expressions. Subsequently, the system constructs a patient context subgraph using the standard entity identifiers as nodes, and writes the evidence attributes into the node attributes and edge attributes. This is then aligned with the existing medical knowledge graph extracted from medical knowledge sources, and relationships are fused to obtain a diagnostic graph. The diagnostic graph simultaneously contains patient evidence relationships and medical knowledge relationships, supporting path retrieval and evidence tracing in the subsequent reasoning stage.

[0117] In the diagnostic reasoning phase, the system locates the set of evidence nodes corresponding to the standardized evidence set in the diagnostic graph. It recalls disease nodes through neighborhood traversal with a limited number of hops to form a candidate disease set. Based on evidence attributes, it assigns weights, treating affirmative evidence as support items and negative evidence as conflict items. Combining the relational edge attributes and path information in the diagnostic graph, it calculates candidate disease scores to obtain candidate disease results. To address the issues of "difficulty in distinguishing candidate diseases and redundancy in differential diagnosis," the system pairs the top-ranked candidate diseases in the results to form candidate disease pair sets. For each candidate disease pair, it retrieves differential diagnosis items that can distinguish between them in the diagnostic graph, constructs a set of differential diagnosis hyperedges, and calculates the confusion degree based on the similarity of candidate disease scores to form a differential diagnosis hyperedge injection sequence. Then, using differential diagnosis items as vertices and differential diagnosis hyperedges as hyperedges, it constructs a hypergraph and executes Berge under the drive of the differential diagnosis hyperedge injection sequence. The minimum cross set is incrementally enumerated. During the incremental enumeration process, the system sequentially adds discriminative superedges to the hypergraph. After each addition, the current cross set is updated, and non-minimum cross sets are deleted using inclusion criteria to ensure that the cross set always maintains its minimumity. At the same time, a minimum size upper bound is introduced for size pruning to suppress combinatorial explosion. The final minimum cross set naturally corresponds to the "minimum discriminative item set". Based on this, the system constructs discriminative suggestion items and establishes the correspondence between discriminative suggestion items and discriminative superedges. Then, it returns to the diagnostic graph to retrieve the relationship path from candidate disease nodes to discriminative item nodes and generates an evidence path set.

[0118] During the output phase, the system generates diagnostic results and writes them into a diagnostic result table. The diagnostic result table includes disease identifier, disease name, ranking, and score fields, and is output according to the ranking. At the same time, it generates differential diagnosis suggestions and writes them into a differential diagnosis suggestion output table. The differential diagnosis suggestion output table includes differential diagnosis suggestion item identifier, differential diagnosis item identifier, differential diagnosis item name, suggestion type, differentiation target, and evidence path fields. The suggestion type indicates whether it is a consultation item or an examination item, the differentiation target indicates the associated candidate disease pair, and the evidence path records the relationship path sequence from candidate disease nodes to differential diagnosis item nodes in the diagnostic graph. The system outputs diagnostic results and differential diagnosis suggestions to support the display of diagnostic ranking and the traceability of differential diagnosis basis. The diagnostic result table is used to present the preliminary diagnostic ranking, and the differential diagnosis suggestion output table is used to present differential diagnosis suggestions and their basis paths.

[0119] To verify the beneficial effects, this embodiment uses the same batch of desensitized medical samples for comparative evaluation. The comparison method selects four representative existing technical routes: expert systems based on rule bases, structured feature classification models based on gradient boosting trees, path scoring retrieval models based on knowledge graphs, and a scheme that does not perform minimum cross-set enumeration but uses information gain greedily to select identification items. The specific comparison data is shown in Table 1:

[0120] Table 1: Comparison of Diagnostic Indicators among Different Auxiliary Diagnostic Protocols

[0121] Top-1 accuracy 57.3% 68.1% 70.3% 71.1% 74.2% Top-3 accuracy 74.5% 85.4% 86.6% 87.2% 89.2% Average number of identification items 3.4 3.0 2.8 3.1 2.3 Redundancy ratio 22.1% 19.4% 15.8% 18.6% 6.9% Average number of interaction rounds 3.1 2.6 2.4 2.8 1.9 Average diagnosis time (minutes) 6.9 5.8 5.5 6.4 4.7 Explanation of traceable pass rate 58.7% 41.2% 79.3% 81.0% 93.5% Cost Index 0.95 0.88 0.83 0.92 0.68

[0122] As shown in Table 1, the present invention achieves superior diagnostic accuracy, with a Top-1 accuracy rate of 74.2% and a Top-3 accuracy rate of 89.2%, both higher than rule-based expert systems (57.3%, 74.5%), structured classification models (68.1%, 85.4%), knowledge graph path scoring (70.3%, 86.6%), and information gain greedy identification (71.1%, 87.2%). Regarding identification efficiency, the present invention's scheme has an average of 2.3 identification items, a redundancy rate of 6.9%, an average of 1.9 interaction rounds, and an average diagnosis time of 4.7 minutes, all superior to the comparative schemes. In terms of interpretability and cost control, the present invention's scheme achieves a 93.5% traceability pass rate and a cost index of 0.68, demonstrating that the minimum identification suggestion based on the minimum cross-section and the evidence path binding output can effectively reduce redundant identification and improve result traceability.

[0123] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A disease diagnosis method based on knowledge graphs, characterized in that, The steps include the following: The medical records of the individuals to be diagnosed are obtained and anonymized. At the same time, structured and semantic normalization processes are performed to obtain a standardized set of evidence. A patient context subgraph is constructed based on a standardized set of evidence, and entity alignment and relationship fusion are performed using a medical knowledge graph to obtain a diagnostic graph. Based on the diagnostic map and standardized evidence set, a phased candidate disease inference is performed to obtain candidate disease results; Receive candidate disease results and diagnostic maps, perform candidate disease pairing, hypergraph construction, and injection sequence generation operations to obtain the hypergraph and the identification hyperedge injection sequence; Based on the discriminative superedge injection sequence, Berge minimum cross set incremental enumeration is performed on the hypergraph. Discriminative superedges are added to the hypergraph in sequence. After each discriminative superedge is added, the current cross set is updated and non-minimum cross sets are deleted to obtain the minimum cross set set. Based on the minimum cross set, identification suggestion items are constructed one by one, and a correspondence with the identification superedge is established. Based on the diagnostic map, the evidence paths associated with the corresponding relationships are extracted to generate an evidence path set. The candidate disease results are formatted and sorted to obtain the diagnostic results, and the evidence path set is formatted and associated to obtain the identification suggestion results.

2. The disease diagnosis method based on knowledge graphs according to claim 1, characterized in that, The standardization of the evidence set is obtained specifically through: Receive the medical data of the object to be diagnosed, perform field parsing and record segmentation operations on the medical data, obtain the medical record set and use it as the original medical dataset; Receive the original medical visit dataset, perform desensitization processing, the desensitization processing includes identifying the identity information field, contact information field, and address field and generating an anonymous identifier, using the anonymous identifier to replace the content of the identity information field, setting the content of the contact information field and address field to empty, and writing the desensitized medical visit dataset into the desensitized medical visit dataset table. The anonymized medical records dataset is subjected to structured processing, and semantic normalization processing is performed to obtain a standardized evidence set, which includes standard entity identifiers and evidence attributes.

3. The disease diagnosis method based on knowledge graphs according to claim 1, characterized in that, The diagnostic chart is obtained specifically through: Receive a standardized set of evidence, write the standard entity identifier into the patient node set, write the evidence attributes into the node attributes, generate relation edges based on the evidence attributes and write them into the corresponding edge attributes to obtain the patient context subgraph; It receives existing medical knowledge sources and performs extraction and standardization processes to obtain a medical knowledge graph that includes knowledge nodes; Read the standard entity identifier of the patient context subgraph node, retrieve the knowledge node with the same standard entity identifier in the medical knowledge graph node set, and write the patient node and knowledge node into the alignment mapping set. The patient node set and the knowledge node set are merged to obtain the diagnosis node set, and the patient relationship set and the knowledge relationship set are merged to obtain the diagnosis relationship set. Alignment association edges are established based on the alignment mapping set, and the node attributes and relationship attributes are merged and written to obtain the diagnosis graph.

4. The disease diagnosis method based on knowledge graph according to claim 1, characterized in that, The specific methods for obtaining the candidate disease results include: Standard entity identifiers and evidence attributes are extracted from the standardized evidence set. Based on the standard entity identifiers, the corresponding nodes are retrieved and located in the diagnostic map to form an evidence node set. The evidence node set is associated with the evidence attributes and written into the evidence index table. Using the set of evidence nodes in the evidence index table as the starting point of the traversal, a neighborhood traversal with a limited number of jumps is performed in the diagnostic graph according to the preset traversal step size. The disease nodes reached during the traversal are recorded and written into the candidate disease table. The candidate disease table is deduplicated to form a candidate disease set. For each candidate disease in the candidate disease set, the relationship edges and path records between the candidate disease and the evidence node set are retrieved in the diagnostic graph. Based on the evidence attributes in the evidence index table, supporting evidence and negative evidence are weighted respectively. The association strength and conflict strength between the candidate disease and the standard entity identifier are calculated based on the relationship edge attributes and path records. The candidate disease score is calculated based on the supporting evidence weight, negative evidence weight, association strength and conflict strength. The candidate disease set is sorted according to the candidate disease score to obtain the candidate disease results.

5. The disease diagnosis method based on knowledge graph according to claim 1, characterized in that, The acquisition of the hypergraph and the discriminative hyperedge injection sequence specifically includes: Extract the candidate disease set and candidate disease score from the candidate disease results, and build a candidate disease index table in the diagnostic map; Perform pairwise pairing operations on the candidate disease set in the candidate disease index table to generate a candidate disease pair set and write it into the candidate disease pair table. The candidate disease pair table records the first candidate disease identifier and the second candidate disease identifier contained in each candidate disease pair. In the diagnostic graph, for each candidate disease pair in the candidate disease pair table, a differential diagnosis item retrieval operation is performed to extract the examination item identifier and the consultation item identifier to form a differential diagnosis item set. The differential diagnosis item set is written into the differential diagnosis superedge and a corresponding record between the differential diagnosis superedge and the candidate disease pair is established. The results are then summarized to form a differential diagnosis superedge set. Based on the candidate disease pair table and the candidate disease index table, obtain the scores of the two candidate diseases for each candidate disease pair, calculate the confusion degree according to the score similarity and write it into the confusion degree table, sort the identification superedge set according to the confusion degree table to generate the identification superedge injection sequence, and write it into the injection sequence table. The vertex set is obtained by deduplicating the identification items in the identification hyperedge set and summarizing them. The identification hyperedge set is written into the hyperedge set, and a hypergraph is constructed based on the vertex set and the hyperedge set.

6. The disease diagnosis method based on knowledge graph according to claim 1, characterized in that, The specific steps to obtain the minimum cross set include: Read the hypergraph and the discrimination hyperedge injection sequence, perform Berge minimum cross set incremental enumeration, assign a unique number to each discrimination item appearing in the hypergraph and generate a discrimination item encoding table, perform binary encoding on each discrimination hyperedge in the discrimination hyperedge injection sequence to generate a hyperedge encoding, select the first discrimination hyperedge in the discrimination hyperedge injection sequence, and construct a single-element cross set encoding for each position of the first discrimination hyperedge, summarize to obtain the initial cross set set and write it into the current cross set set table; Select the superedges to be injected in sequence according to the superedge injection sequence and read the corresponding superedge codes. Write the superedges to be injected into the superedge set of the supergraph and update the supergraph. At the same time, perform a bitwise AND operation between each cross set code in the current cross set set table and the superedge code. Determine whether the operation result has a set bit. If the operation result has a set bit, write the cross set code into the hit set. If the operation result does not have a set bit, write the cross set code into the miss set. Perform an expansion generation operation on the miss set, traverse all the set positions of the superedge code and extract the corresponding identification item number one by one, set the identification item number and write it into the cross set code in the miss set to obtain the expanded cross set code, write the expanded cross set code into the candidate cross set set and write it into the candidate index table, the candidate index table records the number of set positions of the expanded cross set code and records the identification superedge number marker corresponding to the expanded cross set code; Merging the hit set and the candidate crosscut set sets yields a merged crosscut set set. A non-minimum deletion operation is then performed on the merged crosscut set set. The merged crosscut set set is sorted by the number of bits set and the crosscut set codes are traversed sequentially. The current crosscut set code and the retained crosscut set codes are compared one by one for inclusion determination. The inclusion determination includes performing a bitwise AND operation on the two crosscut set codes and determining whether the bitwise AND result is equal to the crosscut set code with the smaller number of bits set. If inclusion is found to be true, the crosscut set code with the larger number of bits set is deleted and the crosscut set code with the smaller number of bits set is retained. An updated crosscut set set is generated and written to the current crosscut set set table. The crosscut set code with the larger number of bits set is the non-minimum crosscut set. Maintain a minimum size upper bound and perform size pruning. Write the minimum size upper bound into the upper bound record table. When generating the extended cross set code, read the upper bound record table and compare the number of bits set in the extended cross set code with the minimum size upper bound. If the number of bits set in the extended cross set code is greater than the minimum size upper bound, delete the extended cross set code and prohibit writing it into the candidate cross set set. After completing the writing to the current cross set set table, read the current cross set set table and update the upper bound record table with the minimum number of bits set. Determine the minimum cross set set using the cross set set in the current cross set set table.

7. The disease diagnosis method based on knowledge graph according to claim 1, characterized in that, The generation of the evidence path set specifically includes: Based on the minimum cross set, the identification item identifiers corresponding to the minimum cross set are extracted one by one. The identification item identifiers are sorted according to the number order in the identification item coding table to generate identification suggestion entries and assign identification suggestion entry identifiers. The identification suggestion entries are summarized to form an identification suggestion set. For each identification suggestion item, the identification superedge set is searched one by one. The identification superedge identifier containing the identification item identifier is located. The identification suggestion item identifier, identification item identifier, and identification superedge identifier are written into the corresponding relationship table to generate the corresponding relationship record. Based on the corresponding relationship records, locate the candidate disease pair identifiers, locate the first candidate disease identifier and the second candidate disease identifier, locate the differential item node in the diagnostic graph based on the differential item identifier, retrieve the relationship path from the first candidate disease node and the second candidate disease node to the differential item node in the diagnostic graph, and combine the node sequence of the relationship path with the relationship sequence as the evidence path set.

8. The disease diagnosis method based on knowledge graph according to claim 1, characterized in that, The specific methods for obtaining the diagnostic results and differential diagnosis recommendations include: Based on the candidate disease results, extract the candidate disease set and candidate disease score, sort the candidate disease set from high to low according to the candidate disease score and generate a sorting rank for each candidate disease, and write the candidate disease identifier, candidate disease score and sorting rank into the diagnosis result table. Based on the identification suggestion table, the corresponding relationship table and the evidence path set, perform the association binding operation, locate the corresponding relationship record in the corresponding relationship table according to the identification suggestion item identifier, locate the associated evidence path record in the evidence path set, and write the identification suggestion item identifier, identification item identifier, identification super-edge identifier and evidence path record into the identification suggestion output table. Perform formatting operations on the diagnostic results table and differential diagnosis suggestion output table, and output the diagnostic results and differential diagnosis suggestion results.

9. A knowledge graph-based disease diagnosis system, comprising executing the knowledge graph-based disease diagnosis method according to any one of claims 1 to 8, characterized in that, include: The evidence standardization module is used to acquire medical data and perform desensitization, structuring, and semantic normalization processing to generate a standardized evidence set. The diagnostic graph fusion module is used to construct a patient context subgraph based on a standardized evidence set and write attributes, construct a medical knowledge graph and perform entity alignment and relation fusion to generate a diagnostic graph. The candidate reasoning module is used to locate evidence nodes based on the diagnostic map and standardized evidence set, perform neighborhood traversal to obtain a set of candidate diseases, and generate candidate disease results. The hypergraph construction module is used to perform pairing to generate candidate disease pairs based on candidate disease results, calculate confusion degree and sort to generate discriminative hyperedge injection sequences; The Berge enumeration module is used to add discriminative superedges to the hypergraph sequentially according to the discriminative superedge injection sequence and perform Berge minimum cross set incremental enumeration to generate a minimum cross set set. The evidence path module is used to generate identification suggestion entries based on the minimum cross set and establish the correspondence with the identification superedge, thereby generating the evidence path set. The results output module is used to sort candidate disease results, perform association binding between identification suggestion items and evidence path sets, and output diagnostic results and identification suggestion results.

Images