A loinc stratification mapping method and system for clinical test items

Abstract

This invention discloses a LOINC hierarchical mapping method and system for clinical laboratory tests, belonging to the field of medical data standardization technology. Specifically, it includes: first, acquiring the local laboratory test names and codes; normalizing the names; extracting structured features such as the test object, sample type, and methodology to form a standardized description; prioritizing hierarchical and accurate matching based on the local mapping table, directly outputting the corresponding LOINC information upon a match; if a match fails, using multiple text similarity algorithms to recall candidate terms, then conducting field conflict detection, adjusting scores based on field matching results, and selecting Top-K candidates; determining the target term through a judgment module, completing automatic system verification or manual review; finally, writing back compliant and valid results to update the mapping table and resource library, continuously optimizing subsequent matching accuracy and efficiency.

Description

Technical Field

[0001] This invention relates to the field of medical data standardization technology, specifically to a LOINC hierarchical mapping method and system for clinical laboratory tests. Background Technology

[0002] Interoperability of clinical laboratory data relies on standardized terminology coding. Hospital-specific laboratory test names have long been influenced by factors such as historical system construction, manual data entry habits, equipment manufacturer naming conventions, and the use of mixed Chinese and English characters, abbreviations, and omissions. These names often differ from those in the standard terminology database, leading to significant obstacles in cross-system sharing, data analysis, and quality control statistics.

[0003] In current practices, common solutions include: first, relying on manually maintained local mapping tables for complete string matching; second, translating Chinese project names into English and then matching them with standard English terms; and third, directly using a general model to provide standard terminology codes based on open-ended project names. All of these solutions struggle to balance accuracy, stability, and engineering feasibility. Specifically, existing technologies suffer from several technical problems: local verification of non-standard project names, numerous abbreviations, and mixed use of Chinese and English leads to a high miss rate when relying solely on string equivalence comparisons. For the same detection object, key information such as sample type, methodology, and attributes is often missing or inconsistently expressed in different systems, making it difficult to stably distinguish semantic neighbors. Single similarity algorithms struggle to simultaneously cover Chinese word similarity, sparse matching of short texts, and semantic similarity across expression forms, leading to potential omissions or unstable ranking of candidate sets. Open-ended models, without candidate constraints, are prone to out-of-set output, sample type confusion, or methodological misjudgments, making them unsuitable for direct use in generating stable standard codes. Existing processes typically lack tiered processing and review triggering mechanisms for difficult and easy projects, resulting in high manual review costs and unauditable results. Summary of the Invention

[0004] The purpose of this invention is to provide a LOINC hierarchical mapping method and system for clinical laboratory tests, thereby solving the problems in the background art: The objective of this invention can be achieved through the following technical solutions: A LOINC hierarchical mapping method for clinical laboratory tests includes the following steps: S1. Obtain the name and / or code of the local inspection item to be mapped; S2. Normalize and extract structural features from the local test item names to be mapped, extract structural fields such as test object, sample type, methodology and time features, and form a standardized test item representation. S3. Perform hierarchical exact matching based on the local mapping table; when a unique and reliable result is found, directly output the corresponding LOINC code and name; S4. When S3 fails to find a match, at least two text similarity algorithms are used to recall candidates in the standard terminology database to obtain an initial candidate set. S5. Perform field conflict detection and penalty on the initial candidate set, adjust the candidate scores according to the matching or conflict between the extracted structural fields and the candidate field of each candidate, and generate the adjusted and sorted Top-K candidates. S6. Input the adjusted Top-K candidates into the candidate restriction determination module, determine the target standard terms from the range of Top-K candidates, and automatically pass or trigger manual review according to the rules; S7. Write back the results of manual confirmation or high-confidence system confirmation to the local mapping table, dictionary resources or rule base for subsequent priority hits.

[0005] As a further aspect of the present invention: In step S2, the specific method for performing normalization processing and structural feature extraction on the local verification item name to be mapped is as follows: Perform full-width / half-width conversion, case normalization, punctuation standardization, word segmentation, stop word filtering, synonym replacement, and abbreviation expansion on the input text; The standardized test item representation is formed by combining the vocabulary or rules to extract the detection object, sample type, methodology, time features, and attribute fields.

[0006] As a further aspect of the present invention: in step S3, the specific method for performing hierarchical precise matching based on the local mapping table is as follows: Access the local mapping table in the priority order of local project code matching, standardized name equality matching, normalized string matching, synonym matching, and abbreviation expansion matching; When a unique and reliable result is obtained, the corresponding LOINC code and name are output directly, and the hit stage is recorded.

[0007] As a further aspect of the present invention: In step S4, the specific method for using at least two text similarity algorithms to perform candidate recall in the standard terminology database is as follows: Using the standard Chinese name or its Chinese translation as the main search object, BM25 similarity, TF-IDF similarity, and semantic vector similarity were calculated to obtain an initial candidate set; The formula for calculating the basic candidate score is as follows: Score_base(c)=α·Sim_BM25(c)+β·Sim_TFIDF(c)+γ·Sim_SEM(c) Wherein, Score_base(c) is the basic candidate score, Sim_BM25(c), Sim_TFIDF(c) and Sim_SEM(c) represent the BM25 similarity, TF-IDF similarity and semantic vector similarity corresponding to candidate c, respectively; α, β and γ are normalized weight parameters, and α+β+γ=1.

[0008] As a further aspect of the present invention: In step S5, the specific method for performing field conflict detection and penalty on the initial candidate set is as follows: The extracted structure fields are compared with the LOINC field of each candidate, and an adjustment score is calculated when conflicts exist. The calculation formula is as follows: Score_adj(c)=Score_base(c)-λ1·P_sys(c)-λ2·P_method(c)-λ3·P_comp(c)-λ4·P_time(c) Where Score_adj(c) is the adjusted score, P_sys(c), P_method(c), P_comp(c), and P_time(c) represent the conflict penalty terms of the candidate in the dimensions of sample type, methodology, detection object, and time feature, respectively, and λ1 to λ4 are non-negative weight parameters; When there are inconsistencies in sample types, methodological conflicts, inconsistent detection targets, or significant discrepancies in time characteristics, the aforementioned penalty values ​​are applied to the corresponding candidates to form adjusted candidate scores and re-rank them.

[0009] As a further aspect of the present invention: In step S6, the specific method for automatically approving or triggering manual review according to the rules is as follows: If the target candidate matches the extracted field, the number of candidates does not exceed the preset threshold, and the score difference between Top-1 and Top-2 meets the preset requirement, then it is determined to pass automatically; When the number of candidates is greater than 6, the sample type conflict is not resolved, the methodological conflict is obvious, or the difference between the Top-1 and Top-2 scores is lower than the preset threshold, it is automatically marked as pending review and enters the manual review queue.

[0010] A LOINC hierarchical mapping system for clinical laboratory tests includes: The data acquisition module is used to obtain the name and / or code of the local test item to be mapped; The preprocessing module is used to perform normalization and structural feature extraction on the local test item names to be mapped, extracting structural fields such as test object, sample type, methodology and time features to form a standardized test item representation; The exact matching module is used to perform hierarchical exact matching based on the local mapping table; when a unique and reliable result is hit, the corresponding LOINC code and name are directly output. The candidate recall module is used to recall candidates in a standard terminology database using at least two text similarity algorithms when no exact match is found, so as to obtain an initial candidate set. The conflict detection module is used to perform field conflict detection and penalty on the initial candidate set, adjust the candidate scores according to the matching or conflict between the extracted structural fields and the fields of each candidate in the candidate set, and generate the adjusted and sorted Top-K candidates. The restricted judgment module is used to determine the target standard terms from the adjusted Top-K candidate range and automatically approve or trigger manual review according to the rules; The result write-back module is used to write back results that have been manually confirmed or passed by the system with high confidence to the local mapping table, dictionary resources or rule base for subsequent priority matching.

[0011] The beneficial effects of this invention are: This invention preserves key technical semantics such as sample type, detection object, and methodology during the standardization process, reducing semantic neighbor confusion. By combining multi-algorithm candidate recall with field penalties, it improves candidate set coverage and ranking stability. Through candidate constraint determination, it suppresses out-of-set output and unexplainable errors caused by open generation.

[0012] This invention reduces the overall workload of manual review by prioritizing high-frequency projects and streamlining the review of difficult projects. Through a result write-back mechanism, the coverage of the local mapping table can be continuously expanded in subsequent batch processing, improving the long-term usability of the system. Attached Figure Description

[0013] The invention will now be further described with reference to the accompanying drawings.

[0014] Figure 1 This is a flowchart illustrating a LOINC hierarchical mapping method for clinical laboratory tests according to the present invention.

[0015] Figure 2 This is a schematic diagram of the LOINC hierarchical mapping system for clinical testing items according to the present invention.

[0016] Figure 3 This is a schematic diagram illustrating the candidate field comparison and consistency verification of the LOINC hierarchical mapping method for clinical laboratory items according to the present invention. Detailed Implementation

[0017] 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.

[0018] Please see Figure 1 As shown, this invention provides a LOINC hierarchical mapping method for clinical laboratory tests, comprising the following steps: S1. Obtain the name and / or code of the local inspection item to be mapped; S2. Normalize and extract structural features from the local test item names to be mapped, extract structural fields such as test object, sample type, methodology and time features, and form a standardized test item representation. S3. Perform hierarchical exact matching based on the local mapping table; when a unique and reliable result is found, directly output the corresponding LOINC code and name; S4. When S3 fails to find a match, at least two text similarity algorithms are used to recall candidates in the standard terminology database to obtain an initial candidate set. S5. Perform field conflict detection and penalty on the initial candidate set, adjust the candidate scores according to the matching or conflict between the extracted structural fields and the candidate field of each candidate, and generate the adjusted and sorted Top-K candidates. S6. Input the adjusted Top-K candidates into the candidate restriction determination module, determine the target standard terms from the range of Top-K candidates, and automatically pass or trigger manual review according to the rules; S7. Write back the results of manual confirmation or high-confidence system confirmation to the local mapping table, dictionary resources or rule base for subsequent priority hits.

[0019] In one embodiment of the present invention, the specific method for performing normalization processing and structural feature extraction on the local verification item name to be mapped in step S2 is as follows: Perform full-width / half-width conversion, case normalization, punctuation standardization, word segmentation, stop word filtering, synonym replacement, and abbreviation expansion on the input text; The standardized test item representation is formed by combining the vocabulary or rules to extract the detection object, sample type, methodology, time features, and attribute fields.

[0020] In one embodiment of the present invention, the specific method for performing hierarchical precise matching based on the local mapping table in step S3 is as follows: Access the local mapping table in the priority order of local project code matching, standardized name equality matching, normalized string matching, synonym matching, and abbreviation expansion matching; When a unique and reliable result is obtained, the corresponding LOINC code and name are output directly, and the hit stage is recorded.

[0021] In one embodiment of the present invention, the specific method for using at least two text similarity algorithms to perform candidate recall in the standard terminology database in step S4 is as follows: Using the standard Chinese name or its Chinese translation as the main search object, BM25 similarity, TF-IDF similarity, and semantic vector similarity were calculated to obtain an initial candidate set; The formula for calculating the basic candidate score is as follows: Score_base(c)=α·Sim_BM25(c)+β·Sim_TFIDF(c)+γ·Sim_SEM(c) Wherein, Score_base(c) is the basic candidate score, Sim_BM25(c), Sim_TFIDF(c) and Sim_SEM(c) represent the BM25 similarity, TF-IDF similarity and semantic vector similarity corresponding to candidate c, respectively; α, β and γ are normalized weight parameters, and α+β+γ=1.

[0022] In one embodiment of the present invention, the specific method for performing field conflict detection and penalty on the initial candidate set in step S5 is as follows: The extracted structure fields are compared with the LOINC field of each candidate, and an adjustment score is calculated when conflicts exist. The calculation formula is as follows: Score_adj(c)=Score_base(c)-λ1·P_sys(c)-λ2·P_method(c)-λ3·P_comp(c)-λ4·P_time(c) Where Score_adj(c) is the adjusted score, P_sys(c), P_method(c), P_comp(c), and P_time(c) represent the conflict penalty terms of the candidate in the dimensions of sample type, methodology, detection object, and time feature, respectively, and λ1 to λ4 are non-negative weight parameters; When there are inconsistencies in sample types, methodological conflicts, inconsistent detection targets, or significant discrepancies in time characteristics, the aforementioned penalty values ​​are applied to the corresponding candidates to form adjusted candidate scores and re-rank them.

[0023] In one embodiment of the present invention, the specific method for automatically approving or triggering manual review according to the rules in step S6 is as follows: If the target candidate matches the extracted field, the number of candidates does not exceed the preset threshold, and the score difference between Top-1 and Top-2 meets the preset requirement, then it is determined to pass automatically; When the number of candidates is greater than 6, the sample type conflict is not resolved, the methodological conflict is obvious, or the difference between the Top-1 and Top-2 scores is lower than the preset threshold, it is automatically marked as pending review and enters the manual review queue.

[0024] Please see Figure 2 As shown, the present invention also provides a LOINC hierarchical mapping system for clinical laboratory tests, comprising: The data acquisition module is used to obtain the name and / or code of the local test item to be mapped; The preprocessing module is used to perform normalization and structural feature extraction on the local test item names to be mapped, extracting structural fields such as test object, sample type, methodology and time features to form a standardized test item representation; The exact matching module is used to perform hierarchical exact matching based on the local mapping table; when a unique and reliable result is hit, the corresponding LOINC code and name are directly output. The candidate recall module is used to recall candidates in a standard terminology database using at least two text similarity algorithms when no exact match is found, so as to obtain an initial candidate set. The conflict detection module is used to perform field conflict detection and penalty on the initial candidate set, adjust the candidate scores according to the matching or conflict between the extracted structural fields and the fields of each candidate in the candidate set, and generate the adjusted and sorted Top-K candidates. The restricted judgment module is used to determine the target standard terms from the adjusted Top-K candidate range and automatically approve or trigger manual review according to the rules; The result write-back module is used to write back results that have been manually confirmed or passed by the system with high confidence to the local mapping table, dictionary resources or rule base for subsequent priority matching.

[0025] Specific embodiment 1 includes: Step 1, Data Acquisition. The system retrieves the local test item names and optional local item codes from LIS, HIS, EMR, or the data platform. Duplicate items are preferably merged, and the character encoding format is standardized.

[0026] Step two, preprocessing and structural feature extraction. The project name undergoes full-width / half-width conversion, case normalization, punctuation standardization, word segmentation, stop word filtering, synonym replacement, and abbreviation expansion. Subsequently, based on a vocabulary or rules, the detection object, sample type, methodology, time features, and attribute fields are extracted to form a standardized representation of the test items.

[0027] Step 3, Layered Precise Matching. The local mapping table is accessed preferentially in the following order: "Local Project Code Matching - Standardized Name Equivalence Matching - Normalized String Matching - Synonym Matching - Abbreviation Extension Matching". When a unique and reliable result is obtained, the corresponding LOINC code and name are directly output, and the hit stage is recorded.

[0028] Step four, multi-algorithm candidate recall. For items that were not found, the standard Chinese name or its Chinese translation is preferred as the primary search target. BM25 similarity, TF-IDF similarity, and semantic vector similarity are calculated to obtain an initial candidate set. Compared to "translate first, then search," this method can better preserve the structural features of the locally tested items in their Chinese expressions.

[0029] Step 5, Field Conflict Detection and Penalty. The structural fields extracted in Step 2 are compared with the LOINC field of each candidate. When there are inconsistencies in sample types, methodological conflicts, inconsistencies in detection objects, or significant discrepancies in time characteristics, a penalty value is applied to the corresponding candidate to form an adjusted candidate score and re-rank them.

[0030] Step Six: Candidate Restriction Determination. The structural fields of the Top-K candidates, the extracted fields of the local verification items, and their original names are input into the candidate restriction determination module to determine the target terms from the Top-K candidate range. Preferably, the determination module outputs the target code, target name, field consistency flag, and review flag.

[0031] Step 7: Consistency Verification and Review. If the target candidate matches the extracted field, the number of candidates does not exceed the preset threshold, and the score difference between Top-1 and Top-2 meets the preset requirements, it automatically passes; otherwise, it enters the manual review queue. Preferably, when the number of candidates is greater than 6, sample type conflicts are not eliminated, methodological conflicts are obvious, or the score difference between Top-1 and Top-2 is lower than the preset threshold, it is automatically marked as pending review.

[0032] Step 8: Feedback and Updates. Write back the results of manual verification or system high-confidence approval to the local mapping table, synonym resources, or rule base to improve the priority hit rate for subsequent similar projects.

[0033] Specific embodiment 2 includes, for example, "serum ferritin".

[0034] The system first resolves "serum ferritin" as: Detection subject = ferritin, Sample type = serum, Methodology = unspecified. If the local mapping table is not matched, multiple candidate terms related to ferritin are recalled from the standard terminology library.

[0035] like Figure 3As shown, candidate A is consistent with the local testing project in terms of both the detection target and sample type; candidate B has the same detection target, but its sample type is plasma; candidate C has a sample type of serum, but its detection target is transferrin. After field conflict detection and penalty, candidate A has the highest adjustment score, and then enters the candidate restriction judgment and passes the consistency check. Finally, the standard terminology code corresponding to candidate A is output.

[0036] This example demonstrates that the present invention does not rely solely on word similarity, but rather incorporates key structural fields such as sample type and detection object into candidate ranking and final determination, thereby reducing the risk of misselection of semantic nearest neighbors.

[0037] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A LOINC hierarchical mapping method for clinical laboratory tests, characterized in that, Includes the following steps: S1. Obtain the name and / or code of the local inspection item to be mapped; S2. Normalize and extract structural features from the local test item names to be mapped, extract structural fields such as test object, sample type, methodology and time features, and form a standardized test item representation. S3. Perform hierarchical exact matching based on the local mapping table; when a unique and reliable result is found, directly output the corresponding LOINC code and name; S4. When S3 fails to find a match, at least two text similarity algorithms are used to recall candidates in the standard terminology database to obtain an initial candidate set. S5. Perform field conflict detection and penalty on the initial candidate set, adjust the candidate scores according to the matching or conflict between the extracted structural fields and the candidate field of each candidate, and generate the adjusted and sorted Top-K candidates. S6. Input the adjusted Top-K candidates into the candidate restriction determination module, determine the target standard terms from the range of Top-K candidates, and automatically pass or trigger manual review according to the rules; S7. Write back the results of manual confirmation or high-confidence system confirmation to the local mapping table, dictionary resources or rule base for subsequent priority hits.

2. The LOINC hierarchical mapping method for clinical laboratory tests according to claim 1, characterized in that, In step S2, the specific method for performing normalization processing and structural feature extraction on the local test item names to be mapped is as follows: Perform full-width / half-width conversion, case normalization, punctuation standardization, word segmentation, stop word filtering, synonym replacement, and abbreviation expansion on the input text; The standardized test item representation is formed by combining the vocabulary or rules to extract the detection object, sample type, methodology, time features, and attribute fields.

3. The LOINC hierarchical mapping method for clinical laboratory tests according to claim 1, characterized in that, In S3, the specific method for performing hierarchical precise matching based on the local mapping table is as follows: Access the local mapping table in the priority order of local project code matching, standardized name equality matching, normalized string matching, synonym matching, and abbreviation expansion matching; When a unique and reliable result is obtained, the corresponding LOINC code and name are output directly, and the hit stage is recorded.

4. The LOINC hierarchical mapping method for clinical laboratory tests according to claim 1, characterized in that, In step S4, the specific method for using at least two text similarity algorithms to perform candidate recall in the standard terminology database is as follows: Using the standard Chinese name or its Chinese translation as the main search object, BM25 similarity, TF-IDF similarity, and semantic vector similarity were calculated to obtain an initial candidate set; The formula for calculating the basic candidate score is as follows: Score_base(c)=α·Sim_BM25(c)+β·Sim_TFIDF(c)+γ·Sim_SEM(c) Wherein, Score_base(c) is the basic candidate score, Sim_BM25(c), Sim_TFIDF(c) and Sim_SEM(c) represent the BM25 similarity, TF-IDF similarity and semantic vector similarity corresponding to candidate c, respectively; α, β and γ are normalized weight parameters, and α+β+γ=1.

5. The LOINC hierarchical mapping method for clinical laboratory tests according to claim 4, characterized in that, In step S5, the specific method for performing field conflict detection and penalty on the initial candidate set is as follows: The extracted structure fields are compared with the LOINC field of each candidate, and an adjustment score is calculated when conflicts exist. The calculation formula is as follows: Score_adj(c)=Score_base(c)-λ1·P_sys(c)-λ2·P_method(c)-λ3·P_comp(c)-λ4·P_time(c) Where Score_adj(c) is the adjusted score, P_sys(c), P_method(c), P_comp(c), and P_time(c) represent the conflict penalty terms of the candidate in the dimensions of sample type, methodology, detection object, and time feature, respectively, and λ1 to λ4 are non-negative weight parameters; When there are inconsistencies in sample types, methodological conflicts, inconsistent detection targets, or significant discrepancies in time characteristics, the aforementioned penalty values ​​are applied to the corresponding candidates to form adjusted candidate scores and re-rank them.

6. The LOINC hierarchical mapping method for clinical laboratory tests according to claim 1, characterized in that, In S6, the specific methods for automatically approving or triggering manual review based on the rules are as follows: If the target candidate matches the extracted field, the number of candidates does not exceed the preset threshold, and the score difference between Top-1 and Top-2 meets the preset requirement, then it is determined to pass automatically; When the number of candidates is greater than 6, the sample type conflict is not resolved, the methodological conflict is obvious, or the difference between the Top-1 and Top-2 scores is lower than the preset threshold, it is automatically marked as pending review and enters the manual review queue.

7. A LOINC hierarchical mapping system for clinical laboratory tests, implemented in any one of claims 1-6, characterized in that, include: The data acquisition module is used to obtain the name and / or code of the local test item to be mapped; The preprocessing module is used to perform normalization and structural feature extraction on the local test item names to be mapped, extracting structural fields such as test object, sample type, methodology and time features to form a standardized test item representation; The exact matching module is used to perform hierarchical exact matching based on the local mapping table; when a unique and reliable result is hit, the corresponding LOINC code and name are directly output. The candidate recall module is used to recall candidates in a standard terminology database using at least two text similarity algorithms when no exact match is found, so as to obtain an initial candidate set. The conflict detection module is used to perform field conflict detection and penalty on the initial candidate set, adjust the candidate scores according to the matching or conflict between the extracted structural fields and the fields of each candidate in the candidate set, and generate the adjusted and sorted Top-K candidates. The restricted judgment module is used to determine the target standard terms from the adjusted Top-K candidate range and automatically approve or trigger manual review according to the rules; The result write-back module is used to write back results that have been manually confirmed or passed by the system with high confidence to the local mapping table, dictionary resources or rule base for subsequent priority matching.

Images