Target disease risk prediction model construction method, prediction method and related device
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INST OF MEDICAL INFORMATION CHINESE ACAD OF MEDICAL SCI
- Filing Date
- 2026-04-10
- Publication Date
- 2026-06-09
Smart Images
Figure CN122177464A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and in particular to a method for constructing a target disease risk prediction model, a prediction method, and related devices. Background Technology
[0002] With the rapid development of medical informatization and artificial intelligence technologies, using electronic medical record (EMR) data for disease risk prediction has become an important direction for clinical decision support.
[0003] Current disease-specific prediction models mostly adopt a general architecture, directly inputting high-dimensional full medical record features into deep learning models or large language models for training and prediction. The feature dimensions are usually as high as tens of thousands of dimensions, which contain a large number of noisy features that are unrelated to or weakly related to the specific disease, seriously interfering with the model learning process and resulting in low model prediction accuracy. Summary of the Invention
[0004] In view of the above problems, this application provides a method for constructing a risk prediction model for a target disease, a prediction method, and related devices, which allows the model learning process to focus more on the core influencing factors of the target disease, effectively improving the prediction accuracy and stability of the model. The specific solution is as follows:
[0005] The first aspect of this application provides a method for constructing a target disease risk prediction model, including:
[0006] A target disease knowledge base is constructed based on a multi-task enhanced retrieval framework. The target disease knowledge base includes: target disease, features, attributes, and evidence.
[0007] Multiple electronic medical records are preprocessed to obtain a set of patient characteristics;
[0008] The patient feature set is semantically aligned with the target disease knowledge base to generate an alignment report for the target disease.
[0009] Based on the alignment report, extract the key features of the target disease from the target disease knowledge base;
[0010] Training samples are constructed based on the key features, and the model is trained to obtain the risk prediction model for the target disease.
[0011] In one possible implementation, the construction of the target disease knowledge base based on the multi-task enhanced retrieval framework includes:
[0012] Based on the name of the target disease, generate multiple sets of structured query instructions;
[0013] The multi-set structured query commands are used to retrieve relevant literature on the target disease from multiple medical databases;
[0014] The structured data units corresponding to the relevant literature of the target disease are extracted using a large language model. The structured data units include: the target disease, features, attributes, and evidence.
[0015] Entity alignment and synonym merging are performed on the structured data units corresponding to the relevant literature of the target disease to generate a knowledge graph in the target disease knowledge base.
[0016] In one possible implementation, the step of semantically aligning the patient feature set with the target disease knowledge base to generate an alignment report for the target disease includes:
[0017] Multi-round semantic reasoning is performed using a large language model to sequentially align the patient feature set with the target disease knowledge base from multiple dimensions, generating an alignment report for the target disease. The multiple dimensions include: clinical manifestations, biomarkers, imaging features, and diagnostic criteria. The alignment report for the target disease includes at least: matching items, the confidence level of the matching items, and the reasoning basis.
[0018] In one possible implementation, based on the alignment report, key features of the target disease are extracted from the target disease knowledge base, including:
[0019] Based on the matching items in the alignment report, candidate features of the target disease are extracted from the target disease knowledge base, and the matching items correspond to the candidate features;
[0020] The importance parameters of the candidate features are analyzed using an interpretable statistical model.
[0021] If the importance parameter of the candidate feature meets the preset screening conditions, the candidate feature is determined as the key feature of the target disease.
[0022] In one possible implementation, the importance parameter of the candidate feature includes at least one of statistical significance and model contribution.
[0023] In one possible implementation, the analysis of the importance parameters of the candidate features using an interpretable statistical model includes:
[0024] Calculate the statistical significance index between the candidate features and the target disease;
[0025] Evaluate the contribution of the candidate features to the model for the target disease.
[0026] The second aspect of this application provides a method for predicting the risk of a specific disease, including:
[0027] Obtaining users' health information;
[0028] Extract key features of the target disease from the user's health information;
[0029] The key features are input into the risk prediction model of the target disease to obtain the risk prediction result output by the risk prediction model of the target disease. The risk prediction model of the target disease is constructed according to the target disease risk prediction model construction method of the first aspect or any implementation of the first aspect.
[0030] A third aspect of this application provides a computer program product, including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement the target disease risk prediction model construction method of the first aspect or any implementation thereof, or the target disease risk prediction method of the second aspect.
[0031] A fourth aspect of this application provides an electronic device, including at least one processor and a memory connected to the processor, wherein:
[0032] The memory is used to store computer programs;
[0033] The processor is used to execute the computer program so that the electronic device can implement the target disease risk prediction model construction method of the first aspect or any implementation of the first aspect or the target disease risk prediction method of the second aspect.
[0034] The fifth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to construct a target disease risk prediction model or a target disease risk prediction method according to the first aspect or any implementation thereof described above.
[0035] By employing the aforementioned technical solutions, the target disease risk prediction model construction method, prediction method, and related apparatus provided in this application construct a standardized target disease knowledge base containing the target disease, features, attributes, and evidence based on a multi-task enhanced retrieval framework. The patient feature set is semantically aligned with the target disease knowledge base to generate an alignment report. Key features are then selected based on the alignment report, and training samples are constructed and trained to obtain the target disease risk prediction model. By coupling literature knowledge, patient data, and the model training process in a closed loop, redundant features irrelevant to the target disease are eliminated, while key features verified by medical knowledge and actually related to the target disease are retained, thereby improving prediction accuracy, stability, and interpretability. Attached Figure Description
[0036] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.
[0037] Figure 1 A flowchart illustrating a method for constructing a target disease risk prediction model, provided in an embodiment of this application;
[0038] Figure 2 This is a partial flowchart illustrating a method for constructing a risk prediction model for a specific disease, as provided in an embodiment of this application.
[0039] Figure 3 This is a partial flowchart illustrating a method for constructing a risk prediction model for a specific disease, as provided in an embodiment of this application.
[0040] Figure 4 A flowchart illustrating another method for constructing a target disease risk prediction model provided in this application embodiment;
[0041] Figure 5 A flowchart illustrating a method for predicting the risk of a specific disease, provided in an embodiment of this application;
[0042] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0043] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.
[0044] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.
[0045] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.
[0046] This application provides a method for constructing a risk prediction model for a specific disease. The method for constructing a risk prediction model for a specific disease according to this application will be described in detail below with reference to the accompanying drawings.
[0047] Reference Figure 1 , Figure 1 A flowchart illustrating a method for constructing a target disease risk prediction model, as provided in this application embodiment, is shown below. Figure 1 As shown in the embodiment of this application, a method for constructing a risk prediction model for a specific disease may include steps 101 to 105, which are described in detail below.
[0048] 101: Construct a target disease knowledge base based on a multi-task enhanced retrieval framework. The target disease knowledge base includes: target disease, features, attributes, and evidence.
[0049] The multi-task enhanced retrieval framework extracts standardized, machine-readable knowledge related to specific diseases from multi-source medical literature to form a target disease knowledge base.
[0050] 102: Preprocess multiple electronic medical records to obtain a set of patient features.
[0051] With the patient's authorization, the patient's full electronic medical record data (including text records, image reports, and laboratory numerical indicators) is collected, cleaned, and de-identified to remove invalid and redundant data, and all valid information is organized into a structured set of patient characteristics.
[0052] 103: Semantically align the patient feature set with the target disease knowledge base to generate an alignment report for the target disease.
[0053] 104: Extract key features of the target disease from the target disease knowledge base based on the alignment report of the target disease.
[0054] 105: Construct training samples based on the key features of the target disease and train the model to obtain a risk prediction model for the target disease.
[0055] By constructing training samples based on the key features of the target disease, the traditional full EMR high-dimensional features are replaced, thereby reducing the feature dimensionality and noise interference from the source.
[0056] A lightweight and high-precision target disease risk prediction model is constructed using training samples corresponding to key features as input. The model parameters are optimized through methods such as cross-validation to ensure the model's prediction accuracy, stability, and interpretability. At the same time, the contribution of key features to the model's prediction results is verified, thereby achieving automatic prediction of target disease risk while ensuring the model's interpretability and stability.
[0057] The model here can be a basic interpretable statistical model, a classical machine learning model, or a deep learning pre-trained model; this embodiment does not impose any specific limitations.
[0058] This embodiment provides a method for constructing a target disease risk prediction model. It builds a standardized disease knowledge base containing diseases, features, attributes, and evidence based on a multi-task enhanced retrieval framework. The method then semantically aligns the patient feature set with this knowledge base to generate an alignment report. Based on the alignment report, key features are selected and training samples are constructed to form a risk prediction model for the target disease. By coupling literature knowledge, patient data, and the model training process in a closed loop, redundant features irrelevant to the target disease are eliminated, while core features verified by medical knowledge and actually related to the target disease are retained, thereby improving prediction accuracy, stability, and interpretability.
[0059] In one possible implementation, please refer to Figure 2 One implementation of step 101 in the above embodiments includes the following steps 1011-1014:
[0060] 1011: Generate multiple sets of structured query commands based on the name of the target disease.
[0061] The system retrieves the name of the target disease to be studied (e.g., neuromyelitis optica spectrum disorder NMOSD) and automatically breaks it down into multiple sets of structured query commands that are relevant to clinical practice. Each set of structured commands includes at least: diagnostic criteria, core clinical manifestations, imaging features, biomarkers, treatment options, and prognostic / risk factors, replacing traditional single keywords.
[0062] 1012: Use multiple sets of structured query commands to retrieve relevant literature on a target disease from multiple medical databases.
[0063] Based on the above structured instructions, relevant literature was retrieved in parallel from multiple medical databases, including PubMed (National Library of Medicine Biomedical Literature Retrieval System), Cochrane (Cochrane Systematic Reviews Database), and professional society guideline databases.
[0064] To improve the efficiency of subsequent data processing, search results can be sorted according to the quality of evidence, prioritizing the retention of systematic reviews, meta-analyses, randomized controlled trials, and authoritative guidelines, while downplaying low-quality case reports and reducing invalid information.
[0065] 1013: Utilize large language models to extract structured data units corresponding to relevant literature on the target disease.
[0066] A large language model (LLM) pre-trained on a medical corpus is used to perform a detailed reading of the sorted full text or abstract of the literature, and to extract standardized information containing multiple fields. For example, the standardized information includes: target disease entity, medical dimension to which the feature belongs, specific description of the feature, quantitative or semi-quantitative expression of the feature, applicable population or stage of the target disease, literature source identifier and evidence level.
[0067] Among them, the large language model does not need to be trained from scratch. It can be lightweightly tuned by pre-defined prompt words to improve the processing performance of the large language model.
[0068] The large language model transforms the extracted standardized information into structured data units that include “target disease – features – attributes – evidence”, thus realizing the transformation from unstructured text to structured data.
[0069] Among them, the target disease specifically refers to the name of the target disease;
[0070] Features are specific medical dimensions directly related to the target disease, which are the specific manifestations or related indicators of the target disease in clinical, examination, and pathological aspects, covering categories such as clinical manifestations, imaging features, biomarkers, laboratory test results, and prognostic indicators;
[0071] Attributes are specific quantitative or qualitative descriptions or scope definitions of the characteristics of a target disease. They are used to clarify key information such as the specific state, value, incidence, and applicable conditions of the characteristics. They are a concrete supplement to the characteristics, giving the characteristics specific attribute values that can be measured and matched.
[0072] Evidence is the medical basis that supports the relationship between "target disease - characteristics - attributes". It includes information such as the literature source, level of evidence, research type, and literature identifier of the relationship. It is the core source basis for verifying the medical rationality and authority of the relationship.
[0073] Taking neuromyelitis optica spectrum targeted disease (NMOSD) as an example, the following table shows an example of a standardized structured data unit:
[0074]
[0075] 1014: Perform entity alignment and synonym merging on the structured data units corresponding to the relevant literature of the target disease to generate a knowledge graph in the target disease knowledge base.
[0076] All structured data units of “target disease – characteristics – attributes – evidence” extracted from all relevant literature will be imported into a preset disease data pool. The target disease name (e.g., NMOSD) will be used as the unique root node, and all structured data units will be associated with the corresponding target disease root node. At the same time, the original literature source information of each structured data unit will be retained.
[0077] Then, entity alignment is performed to unify the naming and types of entities at the target disease and feature levels. The core of entity alignment is to uniformly map different representations of the same entity in the data pool to a pre-defined standardized entity name and type in the knowledge base, eliminating naming ambiguity and ensuring the uniqueness of entities in the atlas. These entities include target disease entities and feature entities. For example, if expressions such as "neuromyelitis optica spectrum disorder" and "NMOSD (Neuromyelitis optica spectrum disorder)" appear in the data pool, they are automatically mapped to the pre-defined standard name "Neuromyelitis optica spectrum disorder (NMOSD)," unifying the target disease root node. For entities at the feature level, alignment is performed uniformly according to medical dimensions + standard names. For example, "spinal cord MRI findings" and "spinal cord MRI features" are both mapped to the standard feature name "imaging features - spinal cord MRI."
[0078] Synonym merging is the process of fusing and deduplicating attributes and evidence related to features. Based on entity alignment, synonym merging fuses, deduplicates, and supplements duplicate or similar attributes and evidence under the same standard feature entity, preserving core information without losing key differences.
[0079] Specifically, attribute-level synonym merging includes:
[0080] For completely duplicate attributes, remove duplicates directly: for example, if two structured data units are both labeled "AQP4 antibody positive specificity 98%", keep only one record and delete the duplicate;
[0081] Similar and complementary attributes are integrated: For example, if one structured data unit is labeled "Spinal MRI shows long-segment myelitis with an incidence of 90%", and another structured data unit is labeled "Spinal MRI shows no abnormalities in 10% of early-stage serologically positive patients", the two can be integrated into "Spinal MRI shows long-segment myelitis with an incidence of 90%; 10% of early-stage AQP4 antibody-positive patients show no abnormalities", thus forming a complete attribute description;
[0082] For attribute labels with slight numerical differences, coexistence is allowed: for example, different literature labels the AQP4 antibody sensitivity as "90%" and "88%", without forcing uniformity, the attribute is labeled as "sensitivity 88%-90% (different study results)" to preserve the authenticity of the data.
[0083] Synonymous merging of evidence includes:
[0084] Deduplication of duplicate evidence from the same document: If multiple structured data units cite the same document, only one source information is retained;
[0085] Different pieces of evidence for the same characteristic are classified and integrated: sorted by evidence level (authoritative guidelines > meta-analysis > systematic review > randomized controlled trial). For example, the "2023 NMOSD diagnosis and treatment guidelines" and "meta-analysis" are both regarded as evidence of "AQP4 antibody positivity". After sorting by level, they are associated with the corresponding characteristics, while retaining the literature identification, study type and publication time of all evidence.
[0086] After completing all entity alignment and synonym merging, the standardized "target disease – feature – attribute – evidence" relationship is automatically mapped to a pre-defined disease knowledge graph framework, generating a visual, machine-readable target disease knowledge graph. Specifically, the standardized target disease name serves as the root node (first-level node) of the graph; standardized feature entities (with type labels) serve as second-level nodes, connected to the target disease root node through the "contains feature" relationship; the fused attribute description serves as a third-level node, connected to the corresponding feature second-level node through the "attribute is" relationship; and the integrated evidence information (with hierarchical sorting) serves as a fourth-level node, connected to the corresponding attribute third-level node through the "evidence support" relationship. The final generated knowledge graph features standardized naming for all nodes, pre-defined fixed relationships for all relationships, and each node / relationship can be traced back to the original literature. The graph also supports dynamic updates; newly added structured data units extracted from literature can be automatically aligned and merged according to the above steps and added to the corresponding nodes of the graph.
[0087] In one possible implementation, step 103 in the above embodiments includes: performing multi-round semantic reasoning using a large language model, sequentially semantically aligning the patient feature set with the target disease knowledge base from multiple dimensions, and generating an alignment report for the target disease.
[0088] For example, multiple dimensions include: clinical presentation, biomarkers, imaging features, and diagnostic criteria.
[0089] The large language model accesses the features, attributes, and evidence information in the target disease knowledge base, and performs multiple rounds of semantic alignment reasoning according to the multiple dimensions, dimension weights, and confidence scoring rules in the preset prompt template; when there is evidence corresponding to exceptional cases, the semantic alignment results are corrected and the corresponding basis is recorded.
[0090] Alignment reports for target diseases should include at least: matches, confidence levels of matches, and reasoning. Alignment reports for target diseases may also include: non-matches (or if-matches) and their confidence levels and reasoning, and uncertainties and their confidence levels and reasoning.
[0091] LLM's chain semantic reasoning follows the logic of single-dimensional feature matching, preliminary score determination, exception correction, cross-dimensional comprehensive verification, and output of the confidence of the matching item. After completing single-feature matching, the results of all dimensions are integrated to form an overall alignment report. Taking neuromyelitis optica spectrum disorder (NMOSD) as an example, the following is the specific implementation process of the subdivision.
[0092] The first round of reasoning involves matching clinical presentation dimensions:
[0093] Matching target: The clinical manifestation characteristics of patients are matched with the clinical manifestation characteristics of NMOSD in the target disease knowledge base;
[0094] LLM extracts clinical features of the patient: optic neuritis, no limb weakness symptoms;
[0095] LLM retrieved clinical features of NMOSD from the target disease knowledge base: optic neuritis / long segmental myelitis (incidence 95%, level of evidence: authoritative guideline), and limb weakness as a common accompanying symptom (incidence 70%, level of evidence: meta-analysis);
[0096] The matching term is: optic neuritis;
[0097] Reasoning basis: Authoritative guidelines;
[0098] Based on the matching results above and the pre-set reliability scoring rules, the confidence level of the matching item for this dimension is determined.
[0099] The second round of reasoning involves matching based on biomarkers.
[0100] Matching targets: Patient laboratory biomarkers and diagnostic / characteristic biomarkers for NMOSD in the target disease knowledge base;
[0101] LLM extracted the patient's biomarker characteristics: serum AQP4 antibody positive;
[0102] LLM retrieved NMOSD biomarker characteristics from the target disease knowledge base: AQP4 antibody positivity (specificity 98%, sensitivity 90%, evidence level: Meta-analysis) and MOG antibody positivity were used as differential diagnostic features;
[0103] The matching condition is: AQP4 antibody positive;
[0104] Reasoning basis: Meta-analysis;
[0105] Based on the matching results above and the pre-set reliability scoring rules, the confidence level of the matching item for this dimension is determined.
[0106] The third round of reasoning involves matching image feature dimensions:
[0107] Matching target: Patient imaging features and typical NMOSD imaging features in the target disease knowledge base;
[0108] LLM extracted the patient's imaging features: spinal MRI showed no abnormalities;
[0109] LLM retrieved the imaging features of NMOSD from the target disease knowledge base: typical spinal MRI findings are long-segmental myelitis (≥3 vertebral segments, incidence 90%, evidence level: systematic review); exception: approximately 10% of AQP4 antibody-positive patients may have no abnormalities on early spinal MRI of the target disease (dynamic follow-up required, evidence level: randomized controlled trial).
[0110] No match found.
[0111] The fourth round of reasoning involves matching diagnostic criteria dimensions:
[0112] Matching target: Patient comprehensive characteristics and NMOSD diagnostic criteria in the target disease knowledge base;
[0113] LLM integrates the first three dimensions of the patient's core characteristics: AQP4 antibody positive + optic neuritis + no abnormalities in the early stage of spinal MRI;
[0114] LLM retrieved the diagnostic criteria for NMOSD from the target disease knowledge base: According to the IPND 2021 criteria, patients who are positive for AQP4 antibodies can be diagnosed with NMOSD if they meet one core clinical feature, without needing to meet all imaging features (Level of evidence: Clinical guidelines).
[0115] Matching criteria: AQP4 antibody positive + optic neuritis;
[0116] Basis for reasoning: NMOSD IPND 2021 diagnostic criteria;
[0117] Based on the matching results above and the pre-set reliability scoring rules, the confidence level of the matching item for this dimension is determined.
[0118] In this embodiment, the result of the previous dimension in semantic reasoning can trigger the association verification of the next dimension, forming a logical closed loop. Each matching item corresponds to a confidence level and reasoning basis, rather than a simple "match / non-match", providing a traceable basis for subsequent key feature screening.
[0119] In one possible implementation, please refer to Figure 3 One implementation of step 104 in the above embodiments includes the following steps 1041-1043:
[0120] 1041: Based on the matching items in the alignment report of the target disease, extract candidate features of the target disease from the target disease knowledge base, with the matching items corresponding to the candidate features.
[0121] Key feature screening uses the full results of the aligned report as the sole patient-side data input, while combining the literature features of the target disease knowledge base to form a dual input basis of "literature knowledge + patient matching results".
[0122] The alignment report includes matching items between patient characteristics and target disease characteristics, along with the confidence level and reasoning basis for these matching items. This clarifies that candidate characteristics for the target disease actually exist in the patient data and have clinical relevance, defining the range of candidate characteristics for the target disease and eliminating features from the alignment report that are completely mismatched or have extremely low confidence. For example, if the target disease knowledge base includes "sleep apnea" as a candidate characteristic for cognitive decline in the elderly, but the alignment report shows that 95% of the subjects have no clinical record of sleep apnea (complete mismatch, confidence score of 0), then this feature will be directly removed from the candidate features to avoid meaningless data analysis.
[0123] 1042: Use interpretable statistical models to analyze the importance parameters of candidate features.
[0124] In one possible implementation, the importance parameter of the candidate feature includes at least one of the statistical significance index (P-value) and the model contribution.
[0125] Among them, the statistical significance index is the statistical significance index between the candidate feature and the target disease, and the model contribution is the model contribution of the candidate feature to the target disease.
[0126] 1043: If the importance parameter of the candidate feature meets the preset screening conditions, the candidate feature is determined as the key feature of the target disease.
[0127] For example, a preset screening condition is met when the statistical significance index of a candidate feature is greater than a first threshold.
[0128] For example, a preset screening condition is met when the model contribution of a candidate feature is greater than a second threshold.
[0129] For example, importance parameters include statistical significance and model contribution. A preset screening condition is met when the statistical significance of a candidate feature is greater than a first threshold and the model contribution of the candidate feature is greater than a second threshold.
[0130] Taking cognitive decline in the elderly as a target disease as an example, the method for determining its key characteristics is as follows.
[0131] The following are the identified candidate features of cognitive function in the elderly:
[0132] Sleep characteristics: nighttime sleep duration, sleep disorders, sleep satisfaction, sleep apnea;
[0133] Depressive characteristics: depression scale score, depressed mood, loss of interest, suicidal ideation;
[0134] Exercise characteristics: weekly exercise duration, exercise frequency, exercise intensity, and daily activity level;
[0135] Basic target disease characteristics: hypertension, diabetes, cerebral infarction, atrial fibrillation.
[0136] Candidate features are mapped to the dataset corresponding to the target disease. Statistical significance indices are calculated for categorical and continuous features in the candidate features, and the model contribution is calculated in conjunction with the interpretable model to determine key features.
[0137] For categorical characteristics (such as sleep disorders: yes / no; diabetes: yes / no), the chi-square test was used to analyze the differences in characteristic distribution between the cognitive decline group and the normal group to obtain statistical significance indicators; for continuous characteristics (such as nighttime sleep duration and depression scale scores), the independent samples t-test was used to analyze the differences in characteristic means between the two groups to obtain statistical significance indicators.
[0138] An interpretable model can be a logistic regression model. For example, a logistic regression model can be constructed with "age-related cognitive decline (yes / no)" as the dependent variable and the feature values corresponding to the above candidate features as independent variables. After the model is trained, the model contribution of each feature is output.
[0139] Please see Figure 4 The method for constructing a target disease risk prediction model provided in this application includes the following steps 401-405:
[0140] 401: Construct a target disease knowledge base based on a multi-task enhanced retrieval framework. The target disease knowledge base includes: target disease, features, attributes, and evidence.
[0141] 402: Preprocess multiple electronic medical records to obtain a set of patient characteristics.
[0142] 403: Semantically align the patient feature set with the target disease knowledge base to generate an alignment report for the target disease.
[0143] 404: Based on the alignment report of the target disease, extract candidate features of the target disease from the target disease knowledge base, and determine the key features of the target disease based on the statistical significance and / or model contribution of the candidate features.
[0144] 405: Construct training samples based on the key features of the target disease and train the model to obtain a risk prediction model for the target disease.
[0145] For details on the implementation of each of the above steps, please refer to the above embodiments, which will not be repeated here.
[0146] This application also provides a method for predicting the risk of a specific disease. Please refer to [link / reference]. Figure 5 The flowchart shown illustrates a method for predicting the risk of a specific disease. This method may include steps 501-503:
[0147] 501: Obtaining user's health information.
[0148] 502: Extract key features of the target disease from the user's health information.
[0149] In the target disease risk prediction model construction method provided in the above embodiments, the target disease corresponds to a set of key features, and based on this, the key features of the target disease are extracted from the user's health information.
[0150] 503: Input the key features into the risk prediction model of the target disease to obtain the risk prediction results output by the risk prediction model of the target disease.
[0151] The risk prediction model for the target disease is constructed according to any of the target disease risk prediction model construction methods provided in the embodiments of this application.
[0152] The risk prediction results include at least one of the following: target disease risk score, risk level, and explanation of the contribution of key features.
[0153] The target disease risk prediction method provided in this embodiment has high prediction accuracy and interpretability because the key features input into the target disease risk prediction model are redundant features that are irrelevant to the target disease, while retaining the core features that have been verified by medical knowledge and are actually related to the target disease. This greatly improves the overall reliability and clinical application value of target disease risk prediction.
[0154] This application also provides an electronic device in its embodiments. (See reference...) Figure 6 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this application. The electronic device in the embodiments of this application may include, but is not limited to, fixed terminals such as smartphones, laptops, tablets, desktop computers, etc., and may also include network-side devices such as servers. Figure 6 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.
[0155] like Figure 6 As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. When the electronic device is powered on, the RAM 603 also stores various programs and data required for the operation of the electronic device. The processing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.
[0156] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 608 including, for example, memory cards, hard drives, etc.; and communication devices 609. Communication device 609 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.
[0157] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the target disease risk prediction model construction methods or target disease risk prediction methods provided in this application.
[0158] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the target disease risk prediction model construction methods or target disease risk prediction methods provided in this application.
[0159] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.
[0160] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.
[0161] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.
[0162] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).
Claims
1. A method for constructing a target disease risk prediction model, characterized in that, include: A target disease knowledge base is constructed based on a multi-task enhanced retrieval framework. The target disease knowledge base includes: target disease, features, attributes, and evidence. Multiple electronic medical records are preprocessed to obtain a set of patient characteristics; The patient feature set is semantically aligned with the target disease knowledge base to generate an alignment report for the target disease. Based on the alignment report, extract the key features of the target disease from the target disease knowledge base; Training samples are constructed based on the key features, and the model is trained to obtain the risk prediction model for the target disease.
2. The method for constructing a target disease risk prediction model according to claim 1, characterized in that, The construction of the target disease knowledge base based on the multi-task enhanced retrieval framework includes: Based on the name of the target disease, generate multiple sets of structured query instructions; The multi-set structured query commands are used to retrieve relevant literature on the target disease from multiple medical databases; The structured data units corresponding to the relevant literature of the target disease are extracted using a large language model. The structured data units include: the target disease, features, attributes, and evidence. Entity alignment and synonym merging are performed on the structured data units corresponding to the relevant literature of the target disease to generate a knowledge graph in the target disease knowledge base.
3. The method for constructing a target disease risk prediction model according to claim 1, characterized in that, The step of semantically aligning the patient feature set with the target disease knowledge base to generate an alignment report for the target disease includes: Multi-round semantic reasoning is performed using a large language model to sequentially align the patient feature set with the target disease knowledge base from multiple dimensions, generating an alignment report for the target disease. The multiple dimensions include: clinical manifestations, biomarkers, imaging features, and diagnostic criteria. The alignment report for the target disease includes at least: matching items, confidence levels of matching items, and reasoning basis.
4. The method for constructing a target disease risk prediction model according to claim 1, characterized in that, The step of extracting key features of the target disease from the target disease knowledge base based on the alignment report includes: Based on the matching items in the alignment report, candidate features of the target disease are extracted from the target disease knowledge base, and the matching items correspond to the candidate features; The importance parameters of the candidate features are analyzed using an interpretable statistical model. If the importance parameter of the candidate feature meets the preset screening conditions, the candidate feature is determined as the key feature of the target disease.
5. The method for constructing a target disease risk prediction model according to claim 4, characterized in that, The importance parameters of the candidate features include at least one of the statistical significance index and the model contribution.
6. The method for constructing a target disease risk prediction model according to claim 4, characterized in that, The analysis of the importance parameters of the candidate features using an interpretable statistical model includes: Calculate the statistical significance index between the candidate features and the target disease; Evaluate the contribution of the candidate features to the model for the target disease.
7. A method for predicting the risk of a specific disease, characterized in that, include: Obtaining users' health information; Extract key features of the target disease from the user's health information; The key features are input into the risk prediction model of the target disease to obtain the risk prediction result output by the risk prediction model of the target disease, wherein the risk prediction model of the target disease is constructed by the target disease risk prediction model construction method according to any one of claims 1 to 6.
8. A computer program product, characterized in that, It includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the target disease risk prediction model construction method as described in any one of claims 1 to 6 or the target disease risk prediction method as described in claim 7.
9. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program so that the electronic device can implement the target disease risk prediction model construction method as described in any one of claims 1 to 6 or the target disease risk prediction method as described in claim 7.
10. A computer storage medium, characterized in that, The storage medium carries one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the target disease risk prediction model construction method as described in any one of claims 1 to 6 or the target disease risk prediction method as described in claim 7.