CN-122025169-A - Judgment path generation and backtracking method for multi-source information of cardiology department

CN122025169ACN 122025169 ACN122025169 ACN 122025169ACN-122025169-A

Abstract

The invention relates to a method for generating and backtracking a judging path of multi-source information of a heart internal medicine, which comprises the steps of obtaining multi-source information of an object to be evaluated, recording source identification and time information, decomposing the multi-source information into evidence sentences, marking evidence levels and stability of the evidence sentences, taking one of hard evidence, soft evidence and elimination evidence, taking one of stable, variable and affected by intervention from the evidence levels, organizing the evidence sentences according to the order of priority of elimination evidence, priority of hard evidence and priority of time anchor, generating a judging path formed by a judging step sequence and step association evidence sentence identification, reversely tracking the judging path according to the identification and the fields when backtracking triggering conditions are met, and positioning evidence sentences causing bifurcation or convergence, wherein the backtracking triggering conditions comprise doctor requests, new evidence conflicts and curative effect anomalies. And the transparency and the clinical application value of the comprehensive judgment of the multi-source information of the cardiology department are improved.

Inventors

GUO LI
ZHANG FENGNA
YU HAIBO

Assignees

佳木斯大学

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. The method for generating and backtracking the judgment path of the multi-source information of the cardiology department is characterized by comprising the following steps: disassembling the multi-source information into evidence sentences, wherein the evidence sentences comprise judgment objects, assertions, conditions and time anchor fields; Marking the evidence level and the stability of the evidence statement, wherein the evidence level is one of hard evidence, soft evidence and evidence elimination, and the stability is one of stable, variable and affected by intervention; And when the retrospective trigger condition is met, reversely retrospecting the judging path according to the identification and the field to locate the evidence statement causing bifurcation or convergence, wherein the retrospective trigger condition comprises a doctor request, new evidence conflict and curative effect abnormality.
2. The method for generating and backtracking the judging path of the multi-source information of the cardiology department according to claim 1 is characterized in that the multi-source information of the same source identification is divided into time segments according to time information, original information is normalized into evidence sentences meeting preset grammar constraints in each time segment, the preset grammar constraints comprise object unique constraints, assertion polarity constraints and condition integrity constraints, the object unique constraints limit each evidence sentence to correspond to one judging object, assertion polarity constraint limits are asserted to be one of positive, negative or trend, and the condition integrity constraints limit the evidence sentences to contain establishment conditions or empty condition identifications.
3. The method for generating and backtracking the judging path of the multi-source information of the cardiology department according to claim 1, wherein when the judging path is generated by organizing evidence sentences, if the same judging object corresponds to a plurality of evidence sentences, stopping the judging step related to the judging object by excluding evidence, when the excluding evidence does not exist, selecting the evidence sentence closest to a time anchor from the hard evidence as a main evidence sentence of the judging step, when the hard evidence does not exist, selecting the evidence sentence with stable state from the soft evidence as the main evidence sentence, and the rest of the evidence sentences are associated to the judging step as auxiliary evidence sentence identifications.
4. The method for generating and tracing back a judging path of multi-source information in cardiology according to claim 1, wherein the tracing back includes reversing the judging steps of the judging path in reverse order and executing a key evidence statement judgment for each judging step, wherein the key evidence statement judgment includes removing the main evidence statement associated with the judging step and reapplying the priority rule to organize the evidence statement, and when the path state at the corresponding judging object is changed from convergence to bifurcation or from bifurcation to convergence after removal, determining the removed main evidence statement as a key evidence statement and outputting the source identification, the time anchor, the evidence level and the stability thereof.
5. The method for generating and tracing back a judgment path of multi-source information in cardiology according to claim 2, wherein the dividing of the time slices includes setting slice boundary rules including at least an event trigger boundary and an interval trigger boundary, the event trigger boundary is used for generating a new time slice when detecting that the polarity of the assertion is changed from positive to negative, from negative to positive or from trend to non-trend, and the interval trigger boundary is used for generating a new time slice when the time interval of the adjacent information exceeds a preset threshold.
6. The method for generating and backtracking the judging path of the intracardiac multi-source information according to claim 2, wherein when the object unique constraint is satisfied, a merging process is performed on a plurality of evidence sentences having the same judging object and being in the same time slice, the merging process includes retaining one main evidence sentence and generating reference identifications of the rest evidence sentences, the asserted polarities of the main evidence sentences are determined according to the order of positive preference trend and negative preference trend, and the source identification set of the merged evidence sentences is recorded in the main evidence sentences.
7. The method for generating and backtracking the judging path of the intracardiac multi-source information according to claim 2, wherein when the condition integrity constraint is satisfied, an empty condition identifier is generated when an evidence statement lacks a satisfaction condition, and further a condition state field is written for the empty condition identifier, wherein the condition state field comprises a default condition state and a condition to-be-completed state, and when the assertion polarity of the evidence statement is trend and the satisfaction condition is absent, the condition state field is set to the condition to-be-completed state.
8. The method for generating and backtracking a judging path of multi-source information in cardiology according to claim 4, wherein when the priority rule is reapplied to organize evidence sentences, candidate evidence sentences are selected from the rest evidence sentences according to the priority order of the evidence levels aiming at judging objects corresponding to the removed main evidence sentences, when a plurality of candidate evidence sentences exist, a substitute main evidence sentence is selected according to the priority order of the time anchor, and the source identification, the time anchor, the evidence level and the stability of the substitute main evidence sentence are output as the comparison information of key evidence sentences.
9. The method for generating and tracing back a judging path of multi-source information in cardiology according to claim 4, wherein when the removal of the main evidence statement does not cause the path state to change from convergent to divergent or from divergent to convergent, the auxiliary evidence statement associated with the judging step is continuously removed in sequence and the evidence statement is re-organized by the priority rule, and when the removal of the main evidence statement causes the path state to change at any time, the evidence statement corresponding to the removal is determined as a key evidence statement, and the source identifier, the time anchor, the evidence level and the stability of the evidence statement are output.
10. The method for generating and tracing back a judging path of multi-source information in cardiology according to claim 9, wherein the auxiliary evidence sentences are sequentially removed according to a removal sequence rule, the auxiliary evidence sentences with the evidence level being soft evidence and the stability being one of the volatile state and the interfered state are preferentially removed, then the auxiliary evidence sentences with the stability being the stable state are removed, when a plurality of auxiliary evidence sentences with the same level and the same stability exist, the auxiliary evidence sentences are removed from near to far according to time anchors, and a path state changer is determined to be a key evidence sentence.

Description

Judgment path generation and backtracking method for multi-source information of cardiology department Technical Field The invention relates to the technical field of medical information processing, in particular to a method for generating and backtracking a judging path of multisource information of a department of cardiology. Background In the prior art, as shown in the technical route of the publication No. CN116434976A, a multi-source knowledge graph and a deep learning model are fused, the key goal of the scheme is to construct a unified knowledge graph through cross-language multi-source biomedical data, and predict potential association between drugs and diseases by using a graph attention network and a multi-layer perceptron, and the method has the advantages of association mining and prediction capability at a macroscopic level, but the method is essentially used for offline inference and result prediction scenes, rather than generating a judgment path for clinical real-time decision. For highly dynamic multisource information in a cardiology scene, such as continuous electrocardiographic monitoring, short-time fluctuation of laboratory indexes and symptom change, the knowledge graph method generally solidifies the information into a static entity and relation representation, and the influence of the evidence intensity change of the same patient at different time points on the judgment path trend is difficult to reflect due to the lack of fine modeling capability of information time sequence, time segment division and state evolution process. Meanwhile, the method takes vector representation and model weight as main decision basis, the prediction result is often expressed as probability value or classification label, which is particularly the electrocardiographic record, which test result or which symptom description can not be clearly explained, plays a key role in the judging process, and the interpretability and the doctor trust degree can be obviously weakened in a clinical decision support system. In addition, the knowledge graph method has higher dependence on data quality and integrity, when multi-source information is missing, condition description is incomplete or trend judgment can only be formed, the scheme is usually processed through vector complementation or model generalization, but the condition completeness and the condition to-be-complemented state are not explicitly distinguished at a logic level, and the influence of uncertain information is easily amplified in the process of judging a path generation, so that potential misjudgment risks are introduced. From the view of the metadata tracing method in the medical management system proposed by CN117076515A, the scheme is mainly focused on the data management and metadata-level tracing problem, determines the source attribution of metadata through attribute matrix, feature matrix and approximation calculation, is suitable for solving the problems of data integration and metadata consistency, but does not directly serve for the generation of clinical judgment paths. The trace-source object of the method is metadata, rather than evidence statement based on clinical semantics, and the comparison dimension is concentrated on static attribute sets of patient attributes, diagnosis scheme attributes, examination attributes, medicine attributes and consumable attributes, so that the description of evidence assertion polarity, evidence level and clinical judgment key elements of stability is lacking. In a multi-source information scene of the department of cardiology, the core of the judging path is not simple data similarity or attribute matching, but the causal influence relationship of different evidences on the judging result at different time nodes is required to be clarified, and the existing metadata tracing method emphasizes approximation screening and mode rejection more and cannot reflect the dynamic change process of the judging path from convergence to bifurcation or from bifurcation to convergence. In addition, although the tracing concept is introduced, the tracing result still stays at the level of selecting the most similar metadata, the tracing capability of judging the change of the path level is lacking, and whether the addition or removal of which evidence causes the change of the path state when the judgment changes at a certain time cannot be answered. Disclosure of Invention The invention aims to provide a method for generating and backtracking a judging path of multisource information of a cardiology department, so that part of defects and shortcomings pointed out in the background art are overcome. The invention solves the technical problems by adopting the following technical scheme that the method for generating and backtracking the judging path of the multi-source information of the cardiology comprises the steps of acquiring the multi-source information of an object to be evaluated and recording source identific