An auxiliary diagnosis method, device and equipment based on a joint auxiliary diagnosis model

Abstract

The application relates to an auxiliary diagnosis method, device and equipment based on a joint auxiliary diagnosis model. The method comprises the following steps: obtaining auxiliary diagnosis guidance information according to target diagnosis and treatment data, calling a joint auxiliary diagnosis model, the model comprising at least a plurality of normal auxiliary diagnosis branches and a plurality of specialized auxiliary diagnosis branches; determining a current dominant branch in the plurality of normal auxiliary diagnosis branches and / or the plurality of specialized auxiliary diagnosis branches, analyzing the target diagnosis and treatment data to generate a current dominant auxiliary diagnosis result, if the current dominant branch is unstable, obtaining a shielded auxiliary diagnosis branch associated with the current dominant branch; determining a candidate takeover branch, and determining whether the candidate takeover branch can take over the current dominant branch to perform subsequent auxiliary diagnosis. The timely intervention and selection of the specialized auxiliary diagnosis branch are realized, the problems that the specialized auxiliary diagnosis model and the actual auxiliary diagnosis process are insufficient in convergence, the intervention time is not easy to determine, and the auxiliary diagnosis path is difficult to adapt in the prior art, so that the timeliness, continuity and accuracy of the rare disease auxiliary diagnosis are affected.

Description

Technical Field

[0001] This application relates to the field of auxiliary diagnostic technology, and in particular to an auxiliary diagnostic method, device and equipment based on a combined auxiliary diagnostic model. Background Technology

[0002] In existing technologies, assisted diagnosis typically involves training a disease identification model based on historical samples, and then using the trained model to extract features and perform matching analysis on the corresponding diagnosis-related data of the target object to output assisted diagnostic results. This type of technology is widely used in assisted diagnosis of common diseases, but in assisted diagnosis of rare diseases, it is often affected by factors such as scarce samples, numerous disease types, large phenotypic differences, and unclear early characteristics, leading to a decrease in the accuracy of assisted diagnosis. At the same time, the diagnosis-related data of some rare diseases overlaps significantly with those of common diseases, making rare diseases vulnerable to the influence of higher prior probabilities and more abundant sample support from common diseases within a unified candidate disease space. To improve the accuracy of assisted diagnosis of rare diseases, some specialized assisted diagnostic models have been proposed for specific rare diseases or categories. However, these models are usually based on a narrow candidate range, a highly specialized input data organization method, and a highly directional discrimination logic.

[0003] However, in actual auxiliary diagnosis, target subjects typically do not directly enter the corresponding specialized auxiliary diagnostic model with a clear indication of rare disease. Instead, they are often processed according to the conventional auxiliary diagnostic process first, and only gradually show a need for further auxiliary diagnosis of rare diseases as relevant diagnostic data is continuously supplemented and relevant characteristics gradually emerge. Because existing specialized auxiliary diagnostic models usually assume that the target subject already has a strong indication of rare disease, has completed pre-screening, or has entered a specific auxiliary diagnostic context, they are not easily and effectively connected with the real, continuous, and dynamically changing auxiliary diagnostic process. Furthermore, different specialized auxiliary diagnostic models are usually only applicable to specific diseases or categories. When multiple specialized auxiliary diagnostic capabilities exist, it is easy to encounter problems such as difficulty in determining the timing of auxiliary diagnostic model intervention and difficulty in adapting auxiliary diagnostic paths, thus affecting the timeliness, continuity, and accuracy of rare disease auxiliary diagnosis. Summary of the Invention

[0004] Based on this, it is necessary to address the aforementioned technical problems by providing an auxiliary diagnostic method, device, and computer equipment that can integrate multiple auxiliary diagnostic branches and incorporate continuous dynamic changes to achieve timely intervention and orderly selection of specialized auxiliary diagnostic branches in the auxiliary diagnostic process. This would at least solve the problems in related technologies such as insufficient connection between specialized auxiliary diagnostic models and actual auxiliary diagnostic processes, difficulty in determining the timing of intervention, and difficulty in adapting auxiliary diagnostic paths, which affect the timeliness, continuity, and accuracy of auxiliary diagnosis for rare diseases.

[0005] On the one hand, an auxiliary diagnostic method based on a combined auxiliary diagnostic model is provided, the method comprising: Based on the target diagnosis and treatment data, auxiliary diagnosis guidance information is obtained, and a pre-constructed joint auxiliary diagnosis model is invoked. The joint auxiliary diagnosis model includes at least multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches, wherein a branch occlusion relationship is predetermined between the multiple normal auxiliary diagnosis branches and / or the multiple specialized auxiliary diagnosis branches. Based on the auxiliary diagnosis guidance information, determine the current dominant branch among the multiple normal auxiliary diagnosis branches and / or the specialized auxiliary diagnosis branches; The target diagnosis and treatment data is analyzed by the current dominant branch to generate the current dominant auxiliary diagnosis result, and the degree of closure of the interpretation of the target diagnosis and treatment data by the current dominant branch is determined. In response to the interpretation closure degree characterizing the instability of the current dominant branch interpretation, one or more occluded auxiliary diagnostic branches associated with the current dominant branch are obtained based on the branch occlusion relationship, and at least one candidate takeover branch is determined from the one or more occluded auxiliary diagnostic branches. The candidate takeover branch is used to parse the target diagnosis and treatment data to generate candidate auxiliary diagnosis results, and it is determined whether the candidate takeover branch meets the branch takeover conditions of the current dominant branch. If it does, the candidate takeover branch is determined as the subsequent dominant branch to perform subsequent auxiliary diagnosis.

[0006] On the other hand, an auxiliary diagnostic device based on a combined auxiliary diagnostic model is provided, the device comprising: The data acquisition and model invocation module is used to obtain auxiliary diagnosis guidance information based on the target diagnosis and treatment data, and to invoke a pre-built joint auxiliary diagnosis model. The joint auxiliary diagnosis model includes at least multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches, wherein a branch occlusion relationship is predetermined between the multiple normal auxiliary diagnosis branches and / or the multiple specialized auxiliary diagnosis branches. The dominant branch confirmation module is used to determine the current dominant branch among multiple normal auxiliary diagnosis branches and / or specialized auxiliary diagnosis branches based on the auxiliary diagnosis guidance information. The branch parsing and evaluation module is used to parse the target diagnosis and treatment data through the current dominant branch to generate the current dominant auxiliary diagnosis result, and to determine the degree of closure of the interpretation of the target diagnosis and treatment data by the current dominant branch; The candidate takeover branch selection module is used to, in response to the interpretation closure degree characterizing the instability of the current dominant branch interpretation, obtain one or more occluded auxiliary diagnostic branches associated with the current dominant branch based on the branch occlusion relationship, and determine at least one candidate takeover branch from the one or more occluded auxiliary diagnostic branches. The takeover assessment and branch switching module is used to parse the target diagnosis and treatment data through the candidate takeover branch to generate candidate auxiliary diagnosis results, and to determine whether the candidate takeover branch meets the branch takeover conditions for the current dominant branch. If it meets the conditions, the candidate takeover branch is determined as the subsequent dominant branch to perform subsequent auxiliary diagnosis.

[0007] In another aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described auxiliary diagnostic method.

[0008] In another aspect, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described auxiliary diagnostic method.

[0009] The aforementioned auxiliary diagnostic methods, devices, computer equipment, and storage media integrate multiple routine auxiliary diagnostic branches and multiple specialized auxiliary diagnostic branches into a joint auxiliary diagnostic model, and generate auxiliary diagnostic guidance information based on target diagnostic data to determine the current dominant branch among the routine auxiliary diagnostic branches and / or the specialized auxiliary diagnostic branches, so that routine auxiliary diagnostic capabilities and specialized auxiliary diagnostic capabilities can collaboratively participate in the auxiliary diagnostic process within the same auxiliary diagnostic framework. Furthermore, this embodiment establishes branch masking relationships between each auxiliary diagnostic branch in advance. When the interpretation of the target diagnosis and treatment data by the current dominant branch becomes unstable, a candidate takeover branch is determined from the masked auxiliary diagnostic branches associated with the current dominant branch based on the branch masking relationships. When the candidate takeover branch meets the branch takeover conditions for the current dominant branch, the candidate takeover branch performs subsequent auxiliary diagnosis. Based on this, the auxiliary diagnostic path can achieve orderly evolution and targeted takeover within a unified joint auxiliary diagnostic model according to a predetermined branch association structure. This facilitates the timely introduction of specialized auxiliary diagnostic capabilities as the target diagnosis and treatment data is continuously supplemented and related features gradually emerge. This improves the problems of insufficient connection between specialized auxiliary diagnostic capabilities and the real, continuous, and dynamically changing auxiliary diagnostic process, difficulty in determining the timing of intervention, and insufficient adaptability of auxiliary diagnostic paths. It also further improves the pertinence, continuity, and stability of the auxiliary diagnostic process, thereby enhancing the accuracy of auxiliary diagnosis in rare disease scenarios. Attached Figure Description

[0010] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1This is a diagram illustrating the application environment of an auxiliary diagnostic method in one embodiment; Figure 2 This is a flowchart illustrating an auxiliary diagnostic method in one embodiment; Figure 3 This is a structural block diagram of an auxiliary diagnostic device in one embodiment; Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0013] This application provides an auxiliary diagnostic method based on a combined auxiliary diagnostic model, which can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. Terminal 102 can collect, input, display, and update diagnosis-related information corresponding to the target object, such as current round of diagnosis-related data, historical rounds of diagnosis-related data, and subsequently added diagnosis-related data, and send the relevant data to server 104. Server 104 can process the diagnosis-related data, extract information reflecting the current diagnosis-related status and disease progression changes, and then call a pre-built joint auxiliary diagnosis model to perform auxiliary diagnosis. The joint auxiliary diagnosis model can consist of multiple auxiliary diagnosis branches oriented towards different interpretation objects, and is organized based on the masking association between branches, used to dynamically determine the auxiliary diagnosis path that is more suitable for undertaking the interpretation task during the diagnosis-related process. Server 104 can determine, verify, and adjust the currently dominant auxiliary diagnostic branch based on input data and changes in the diagnostic status. When the interpretability of the current auxiliary diagnostic path for the diagnostic data decreases, it can switch to a more suitable subsequent auxiliary diagnostic branch along the relationship between branches to continue performing auxiliary diagnosis. When the subsequent specialized diagnostic stage is completed or the superior auxiliary diagnostic path once again meets the interpretation requirements, it can also perform corresponding back-switch processing to achieve dynamic connection and adaptive adjustment in the auxiliary diagnostic process. Afterwards, server 104 can feed back the auxiliary diagnostic results, branch switching results, and / or corresponding prompts to terminal 102 for the user to view, confirm, or continue to supplement diagnostic data.

[0014] The network can be a wired network or a wireless network. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. Server 104 can be implemented as a standalone server or as a server cluster consisting of multiple servers.

[0015] In one embodiment, such as Figure 2 As shown, an auxiliary diagnostic method based on a combined auxiliary diagnostic model is provided, which is then applied to... Figure 1 Taking a 104 error on the server side as an example, the following steps are included: Step 201: Obtain auxiliary diagnosis guidance information based on the target diagnosis and treatment data, and call the pre-built joint auxiliary diagnosis model. The joint auxiliary diagnosis model includes at least multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches. Among them, there is a predetermined branch occlusion relationship between the multiple normal auxiliary diagnosis branches and / or the multiple specialized auxiliary diagnosis branches. Step 202: Determine the current dominant branch among multiple routine auxiliary diagnosis branches and / or specialized auxiliary diagnosis branches based on the auxiliary diagnosis guidance information; Step 203: Analyze the target diagnosis and treatment data through the current dominant branch to generate the current dominant auxiliary diagnosis result, and determine the degree of closure of the current dominant branch's interpretation of the target diagnosis and treatment data; Step 204: In response to the instability of the interpretation of the closed-loop degree characterization of the current dominant branch, one or more occluded auxiliary diagnostic branches associated with the current dominant branch are obtained based on the branch occlusion relationship, and at least one candidate takeover branch is determined from the one or more occluded auxiliary diagnostic branches. Step 205: Analyze the target diagnosis and treatment data through the candidate takeover branch to generate candidate auxiliary diagnosis results, and determine whether the candidate takeover branch meets the branch takeover conditions for the current dominant branch. If it meets the conditions, the candidate takeover branch is determined as the subsequent dominant branch to perform subsequent auxiliary diagnosis.

[0016] Specifically, this embodiment provides an auxiliary diagnosis method based on a joint auxiliary diagnosis model. By incorporating multiple routine auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches into the joint auxiliary diagnosis model, and generating auxiliary diagnosis guidance information based on target diagnosis and treatment data, the current dominant branch is determined among the routine auxiliary diagnosis branches and / or specialized auxiliary diagnosis branches. This enables routine auxiliary diagnosis capabilities and specialized auxiliary diagnosis capabilities to collaboratively participate in the auxiliary diagnosis process within the same auxiliary diagnosis framework. Furthermore, this embodiment establishes branch masking relationships between each auxiliary diagnostic branch in advance. When the interpretation of the target diagnosis and treatment data by the current dominant branch becomes unstable, a candidate takeover branch is determined from the masked auxiliary diagnostic branches associated with the current dominant branch based on the branch masking relationship. When the candidate takeover branch meets the branch takeover conditions for the current dominant branch, the candidate takeover branch performs subsequent auxiliary diagnosis. Based on this, the auxiliary diagnostic path can achieve orderly evolution and targeted takeover within a unified joint auxiliary diagnostic model according to a predetermined branch association structure. This facilitates the timely introduction of specialized auxiliary diagnostic capabilities as the target diagnosis and treatment data is continuously supplemented and related features gradually emerge. This improves the problems of insufficient connection between specialized auxiliary diagnostic capabilities and the real, continuous and dynamically changing auxiliary diagnostic process, difficulty in determining the intervention timing, and insufficient adaptability of auxiliary diagnostic paths. It also further improves the pertinence, continuity and stability of the auxiliary diagnostic process, thereby enhancing the accuracy of auxiliary diagnosis in rare disease scenarios.

[0017] In one embodiment, auxiliary diagnostic guidance information is obtained based on target diagnostic data, including: Obtain target treatment data, which includes at least one or more of the following: current round treatment data, historical round treatment data; The current round of diagnosis and treatment data may refer to newly collected or newly entered diagnosis and treatment related data in the current round, preferably including at least one of the following: chief complaint, present illness history, past medical history, physical examination records, laboratory test results, imaging examination results, pathology results, medication records, treatment response records, family history, and medical record text information; historical round of diagnosis and treatment data may refer to diagnosis and treatment data collected in one or more rounds prior to the current round. Construct a sequence of diagnostic and treatment characteristics based on current rounds of treatment data and / or historical rounds of treatment data; The diagnostic and treatment feature sequence refers to a set of diagnostic and treatment features organized in chronological order, preferably obtained through the following method: First, the current round of diagnostic and treatment data and the historical round of diagnostic and treatment data are structured and timestamped. Then, symptoms, signs, abnormal items of examination indicators, the magnitude of indicator changes, treatment response, disease progression events, etc. are extracted as corresponding diagnostic and treatment features. Finally, they are arranged according to the order of rounds, the order of occurrence, and / or the order of clinical events to obtain the diagnostic and treatment feature sequence. Based on the diagnostic and treatment feature sequence, at least one guiding feature is identified to characterize the current diagnostic and treatment status and the relationship between the disease course evolution, and auxiliary diagnostic guidance information is generated based on the guiding feature; Among them, guiding features include direct guiding features and trend guiding features; direct guiding features can refer to features that have a strong unique pointing relationship with a certain branch of explanation, such as characteristic gene mutations, characteristic metabolic abnormalities, characteristic imaging manifestations, or characteristic symptom combinations; trend guiding features can refer to temporal features that reflect the continuous aggravation, relief, prolongation, transformation, or concurrent evolution of the disease.

[0018] Preferably, the auxiliary diagnosis guidance information can be represented in any one or more of the following forms: feature label set, branch routing vector, branch priority ranking result, object candidate set or corresponding confidence distribution; Preferably, the auxiliary diagnosis guidance information can be obtained by inputting the guidance features into a preset guidance information generation module. The guidance information generation module can be a rule engine, a classification model, a sequence model, or a combination thereof. When using a rule engine, each rule in the rule base includes at least a guiding feature identifier, a triggering condition, and a target auxiliary diagnosis branch identifier. For direct guiding features, when the corresponding guiding feature appears in the diagnosis feature sequence, the auxiliary diagnosis branch uniquely corresponding to it is triggered. For trend guiding features, when at least two of the three most recent effective diagnosis rounds have the same direction of change, and the normalized change of the most recent round relative to the average of the three effective diagnosis rounds is not less than the change threshold, the corresponding auxiliary diagnosis branch is triggered. The reference range for the change threshold is 0.15 to 0.30, preferably 0.20. When using a classification model, the XGBoost multi-class classification model can be used, with the feature vector obtained by the guided feature encoding as input, and the output being the classification probability corresponding to each auxiliary diagnostic branch. The reference range for the number of decision trees in the XGBoost multi-class classification model is 100 to 300, preferably 200. The maximum depth of a single decision tree is referenced to be 3 to 6 layers, preferably 4 layers. When using a sequence model, the sequence model can be a gated recurrent unit model, i.e., a GRU model, with the aforementioned diagnosis and treatment feature sequence organized in chronological order as input. Output the temporal probabilities corresponding to each auxiliary diagnostic branch. The reference range for the length of the input sequence is 4 to 12 treatment rounds, preferably the most recent 8 treatment rounds. The reference range for the hidden layer dimension of the GRU model is 32 to 128, preferably 64. When the rule engine, classification model, and sequence model are used in combination, the branch matching results output by the rule engine, the classification probability output by the classification model, and the temporal probability output by the sequence model are converted to the range of 0 to 1, respectively. The reference range for the proportions of the three are 40% to 60%, 20% to 30%, and 20% to 30%, respectively, and the sum of the proportions of the three is 100%; preferably 50%, 25%, and 25%, respectively. Arrange the auxiliary diagnostic branches in descending order according to the merged results, and generate the branch priority ranking results and / or confidence distribution.

[0019] Specifically, in this embodiment, by acquiring the current round of treatment data and historical rounds of treatment data, and constructing a treatment feature sequence accordingly, guiding features that characterize the current treatment status and disease progression relationship are extracted from the treatment feature sequence to generate auxiliary diagnostic guidance information. This helps to ensure that the auxiliary diagnostic guidance information not only reflects the treatment performance at the current point in time, but also reflects the continuous evolution trend in the treatment process. This avoids the bias caused by branch guidance based solely on static data from a single round, and provides a more temporally continuous and disease progression-adaptive basis for determining the current dominant branch in the future.

[0020] It is worth noting that in actual auxiliary diagnosis, different diseases, symptoms, and disease stages are often not isolated from each other, but rather exhibit a relationship of gradual refinement from the normal explanatory object to the specialized explanatory object, and gradual convergence from the higher-level explanatory scope to the lower-level explanatory scope. Some existing multi-branch collaborative models typically treat multiple branches as relatively parallel processing units for result fusion, weighted selection, or static routing distribution. They focus more on the collaborative utilization of the output results of multiple branches, but they do not adequately consider the higher-level constraints between different branches in terms of explanatory scope, disease stage, and object coverage. Therefore, when facing scenarios where the disease course continues to evolve, the explanatory focus gradually shifts downward, or dynamic switching is required between the higher-level explanatory path and the lower-level specialized explanatory path, problems such as duplicate branch interpretation, disordered competition, or unclear switching criteria may easily occur.

[0021] Therefore, this application provides an embodiment in which, before invoking a pre-built joint diagnostic model, the method further includes: Obtain multiple auxiliary diagnostic branches and determine the branch interpretation objects for these branches; Among them, auxiliary diagnosis branches can refer to sub-models, sub-networks, rule modules, expert system modules, or combinations thereof in the joint auxiliary diagnosis model that perform auxiliary diagnosis for specific branch interpretation objects; the branch interpretation objects can refer to the diseases, symptoms, disease stages, complications, subtypes, or combinations thereof that are the focus of the corresponding auxiliary diagnosis branches in interpretation. If the branch explanation object is a common disease and related symptoms, the corresponding auxiliary diagnosis branch is determined as a normal auxiliary diagnosis branch; if the branch explanation object is a rare disease and related symptoms, the corresponding auxiliary diagnosis branch is determined as a specialized auxiliary diagnosis branch. The classification of common diseases and related symptoms and rare diseases and related symptoms can be determined based on preset epidemiological thresholds, medical knowledge base annotations, clinical guideline classification results and / or expert annotation results; preferably, rare diseases can be marked according to a pre-stored rare disease catalog, incidence thresholds or guideline classification results; routine auxiliary diagnosis branches and specialized auxiliary diagnosis branches can also be pre-assigned corresponding labels when the branches are established; For multiple routine auxiliary diagnosis branches, based on the disease course evolution relationship between the multiple branch explanation objects, one or more specialized auxiliary diagnosis branches that have a disease course evolution association with the branch explanation objects of the routine auxiliary diagnosis branches are selected to construct multiple hybrid auxiliary diagnosis branch clusters, and the routine auxiliary diagnosis branches and the selected specialized auxiliary diagnosis branches are determined as auxiliary diagnosis branches within the cluster. Among them, the disease course evolution relationship can refer to the association between different branch explanation objects in terms of disease development, complications, stage migration, symptom evolution or changes in examination indicators; preferably, the disease course evolution relationship can be obtained through at least one of the following methods: an object relationship table based on clinical guidelines or medical knowledge graphs, an object transfer probability matrix obtained based on historical sample statistics, or a set of evolution paths determined based on expert rules; Among them, a hybrid auxiliary diagnosis branch cluster can refer to a local branch set formed by combining one or more specialized auxiliary diagnosis branches that are related to the disease course evolution with a normal auxiliary diagnosis branch as the core; preferably, when constructing multiple hybrid auxiliary diagnosis branch clusters, each normal auxiliary diagnosis branch can be limited to correspond to at least one branch cluster, and each specialized auxiliary diagnosis branch can be included in one or more branch clusters with evolutionary correlation. For multiple hybrid auxiliary diagnosis branch clusters, extract the diagnosis and treatment data related to the branch interpretation objects of the hybrid auxiliary diagnosis branch clusters respectively, and generate a hybrid input set of multiple branch clusters; The mixed input set of the branch cluster can refer to the subset of data selected from the original target diagnosis and treatment data that is commonly related to multiple intra-cluster auxiliary diagnosis branches of the corresponding mixed auxiliary diagnosis branch cluster; preferably, the mixed input set of the branch cluster can be obtained in the following way: first, extract relevant diagnosis and treatment data according to the symptom vocabulary, examination item table, abnormal indicator pattern table, disease course label and knowledge graph mapping relationship corresponding to the branch explanation object, and then splice, normalize and encode the relevant data of the same patient in different rounds to obtain the mixed input set of the corresponding branch cluster; The multiple branch clusters are mixed input sets are respectively input into the corresponding multiple mixed auxiliary diagnosis branch clusters to obtain multiple branch cluster output results; The branch cluster output results are analyzed to determine the branch interpretation coverage and output result bias among multiple intra-cluster auxiliary diagnostic branches of the corresponding hybrid auxiliary diagnostic branch cluster. Based on the branch interpretation coverage and output result bias, the intra-cluster branch occlusion relationship of the hybrid auxiliary diagnostic branch cluster is determined to characterize at least one intra-cluster auxiliary diagnostic branch's occlusion / masking relationship with the remaining multiple intra-cluster auxiliary diagnostic branches and the corresponding occlusion / masking level. Among them, the branch explanation coverage refers to the degree to which one auxiliary diagnosis branch covers the branch explanation objects, typical features of the objects, and related diagnosis and treatment information factors of another auxiliary diagnosis branch. Specifically, the branch explanation coverage can be obtained by: statistically analyzing the intersection ratio between the branch explanation object sets of two auxiliary diagnosis branches, and weighting it by combining the importance weight of each object, the degree of overlap of the typical features of the objects, and the degree of overlap of the explanation of the same diagnosis and treatment information factors to obtain the coverage value. The output bias can refer to the degree of bias shown by a certain auxiliary diagnostic branch within a branch cluster compared with other auxiliary diagnostic branches under the same input conditions in terms of output object orientation, output confidence, ranking priority, or interpretation stability. Preferably, the output bias can be obtained by comparing the object orientation score, confidence score, interpretation closure-loop related index, and result ranking position of multiple auxiliary diagnostic branches within the same cluster for the same mixed input set, and then normalizing the difference. If there is a disease progression relationship between the branch interpretation objects corresponding to the normal auxiliary diagnosis branches within multiple mixed auxiliary diagnosis branch clusters, then one or more mixed auxiliary diagnosis branch cluster groups are constructed based on multiple mixed auxiliary diagnosis branch clusters, and the intra-group branch cluster occlusion relationship between multiple mixed auxiliary diagnosis branch clusters is determined. Among them, the mixed auxiliary diagnosis branch cluster group can refer to the set of branch clusters formed by multiple mixed auxiliary diagnosis branch clusters that are related to each other in terms of disease course evolution; the branch cluster occlusion relationship within the group can refer to the directed occlusion relationship between different mixed auxiliary diagnosis branch clusters; preferably, the branch cluster occlusion relationship within the group can be obtained by comparing the disease course position, explanatory scope and cross-cluster result bias between the core normal auxiliary diagnosis branches of different branch clusters. Based on intra-cluster branch occlusion relationships and intra-group branch cluster occlusion relationships, the branch occlusion relationships between multiple normal auxiliary diagnostic branches and / or multiple specialized auxiliary diagnostic branches are determined; Based on the branch occlusion relationship, multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches are organized to generate a joint auxiliary diagnosis model; Among them, the joint auxiliary diagnosis model can be preferably organized into a hierarchical graph structure, a routing network structure, a tree structure, or a graph network structure, so that the calling order, masking relationship, and switching path between multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches can be uniformly managed through the branch masking relationship.

[0022] Specifically, in this embodiment, multiple auxiliary diagnosis branches are first classified into normal auxiliary diagnosis branches and specialized auxiliary diagnosis branches according to the branch interpretation object. Then, a hybrid auxiliary diagnosis branch cluster is constructed based on the disease course evolution relationship, the branch occlusion relationship within the cluster and the branch cluster occlusion relationship within the group are determined, and a global branch occlusion relationship is further formed to organize and generate a joint auxiliary diagnosis model. This is beneficial to incorporate the originally relatively independent normal auxiliary diagnosis capabilities and specialized auxiliary diagnosis capabilities into a unified auxiliary diagnosis structure with related hierarchical structures. As a result, subsequent branch selection, takeover and reversal can all be based on pre-organized structured relationships, improving the orderliness and controllability of the overall operation of the joint auxiliary diagnosis model.

[0023] In one specific embodiment, based on intra-cluster branch occlusion relationships and intra-group branch-cluster occlusion relationships, the branch occlusion relationships between multiple normal auxiliary diagnostic branches and / or multiple specialized auxiliary diagnostic branches are determined, specifically including: Determine whether any two auxiliary diagnostic branches are located in the same mixed auxiliary diagnostic branch cluster; In response to two auxiliary diagnostic branches being located in the same mixed auxiliary diagnostic branch cluster, the occlusion / occlusion relationship between the two auxiliary diagnostic branches and the corresponding occlusion level are determined based on the corresponding intra-cluster branch occlusion relationship. Among them, the intra-cluster branch occlusion relationship can refer to the directed relationship between different intra-cluster auxiliary diagnostic branches within the same mixed auxiliary diagnostic branch cluster, wherein the higher-level auxiliary diagnostic branch occludes the lower-level auxiliary diagnostic branch; the occlusion level can be used to characterize the occlusion strength, preferably divided into two, three, four or five levels, or can be represented by continuous numerical values. Specifically, the occlusion level can be obtained based on the weighted result of branch interpretation coverage and output result bias. The higher the coverage and the more obvious the bias, the higher the corresponding occlusion level. Preferably, the aforementioned branch interpretation coverage and output result bias are converted to the range of 0 to 1, and weighted according to a predetermined ratio to obtain the occlusion intensity. The reference range for the proportion of branch interpretation coverage is 50% to 70%, and the reference range for the proportion of output result bias is 30% to 50%, preferably 60% and 40% respectively. When the occlusion intensity reaches 0.40 and is lower than 0.55, it is determined to be level two occlusion; when it reaches 0.55 and is lower than 0.70, it is determined to be level three occlusion; when it reaches 0.70 and is lower than 0.85, it is determined to be level four occlusion; when it reaches 0.85, it is determined to be level five occlusion. When the occlusion intensity is lower than 0.40, a direct occlusion relationship between the two auxiliary diagnostic branches is not established based on the current branch cluster output result. In response to two auxiliary diagnostic branches located in different mixed auxiliary diagnostic branch clusters and the corresponding mixed auxiliary diagnostic branch clusters located in the same mixed auxiliary diagnostic branch cluster group, the inter-cluster occlusion direction between the two mixed auxiliary diagnostic branch clusters is determined based on the occlusion relationship of the corresponding intra-group branch clusters. In combination with the relative occlusion positions of the two auxiliary diagnostic branches within their respective mixed auxiliary diagnostic branch clusters, the occlusion / occlusion relationship and the corresponding occlusion level between the two auxiliary diagnostic branches are determined. If two auxiliary diagnosis branches are located in different mixed auxiliary diagnosis branch clusters, it is determined that there is no direct branch occlusion relationship between the two auxiliary diagnosis branches, or a weak occlusion relationship between the two auxiliary diagnosis branches is determined based on the overlap of objects and the correlation of disease course evolution between the branch interpretation objects corresponding to the two auxiliary diagnosis branches. Among them, a weak occlusion relationship can refer to a relationship in which, although there is no stable direct occlusion relationship between two auxiliary diagnosis branches, one auxiliary diagnosis branch has a limited explanatory coverage or limited priority guidance effect on the other auxiliary diagnosis branch in some objects, some scenarios, or some stages of the disease course; preferably, it can be determined as a weak occlusion relationship when both the object overlap degree and the disease course evolution correlation degree exceed the weak occlusion threshold, and the weak occlusion threshold can be set to, for example, 0.3 to 0.6; The occlusion / occlusion relationships among multiple auxiliary diagnostic branches are integrated to generate a branch occlusion relationship map. The branch occlusion relationship map at least represents the occlusion direction, occlusion level and cross-cluster transmission path among multiple auxiliary diagnostic branches. In response to the existence of bidirectional occlusion conflicts in the branch occlusion relationship graph, conflict resolution is carried out based on the corresponding occlusion level, inter-cluster occlusion direction priority, and branch interpretation coverage. The branch occlusion relationship graph can be represented by a directed weighted graph, where nodes correspond to auxiliary diagnostic branches, edges correspond to occlusion directions, edge weights correspond to occlusion levels or occlusion strengths, and cross-cluster paths can be used to characterize potential switching paths between different branch clusters. In response to bidirectional occlusion conflicts, the conflict resolution can preferably be carried out in the following order: first compare occlusion levels, then compare the priority of occlusion directions between clusters, and finally compare the branch interpretation coverage. If the conflict still cannot be resolved, expert rules or validation set performance can be introduced as the final judgment criteria.

[0024] Specifically, in this embodiment, by distinguishing between different situations where auxiliary diagnosis branches are in the same mixed auxiliary diagnosis branch cluster, the same mixed auxiliary diagnosis branch cluster group, or different mixed auxiliary diagnosis branch cluster groups, and combining intra-cluster occlusion relationships, intra-group branch cluster occlusion relationships, object overlap, and disease progression correlation, a branch occlusion relationship map is generated and bidirectional occlusion conflicts are resolved. This is beneficial for expanding local intra-cluster relationships into a branch relationship structure that can be called globally, thereby providing a unified relationship basis for the continuation, instability divergence, failure backtracking, and back-cutting upward of subsequent dominant branches, avoiding the problem of disordered competition or unclear switching paths among multiple auxiliary diagnosis branches.

[0025] In one specific embodiment, before invoking the joint diagnostic auxiliary model, the method further includes training the joint diagnostic auxiliary model, specifically including: Obtain a historical treatment sample set, where each sample includes at least: multi-round treatment data, target interpretation object annotation, corresponding disease course evolution annotation, and / or key treatment information factor annotation; The historical diagnosis and treatment sample set may consist of one or more of the following: historical case database, electronic medical record database, disease database, gene testing database, laboratory database, and expert-annotated sample database; preferably, each sample in the historical diagnosis and treatment sample set contains time-series diagnosis and treatment information that corresponds to a round, in order to support disease progression learning. Among them, the target interpretation object label can refer to the target disease, symptoms, disease stage, complication or subtype label given for each historical sample; the disease evolution label can refer to the disease progression, relief, outcome or transformation markers of cases between different rounds; the key diagnosis and treatment information factor label can refer to the diagnosis and treatment factors with high specificity, high risk or high decision value marked by experts or rule system; Based on the historical clinical sample set, a training input set and a validation input set are generated, and guidance feature supervision information, object pointing supervision information, explanation of closed loop state supervision information and branch occlusion relationship supervision information corresponding to the training input set and the validation input set are constructed respectively. Among them, the guiding feature supervision information can be obtained by annotating key temporal features, characteristic abnormal indicators, and typical disease progression events in historical samples; the object-oriented supervision information can be generated by the annotation of target interpretation objects and their hierarchical relationships; the interpretation closed-loop status supervision information can be generated by manual annotation, expert review and judgment, and / or based on the rule results of whether key factors are covered by the interpretation chain; the branch occlusion relationship supervision information can be obtained by combining the branch priority relationship defined by experts, the differences in the interpretation performance of each branch on the same object in historical cases, and the object coverage relationship. Preferably, the training input set and the validation input set can be divided according to a preset ratio, such as 7:3, 8:2 or a 7:2:1 division including the test set; the convergence condition can be set in the following way: when the decrease in the overall loss on the validation set is less than the preset convergence threshold within a preset number of consecutive rounds, or the increase in the overall performance index is less than the preset magnitude, the convergence condition is determined to be met. Based on the training input set, auxiliary diagnosis guidance information is generated respectively, and the training input set is input into multiple auxiliary diagnosis branches so that the multiple auxiliary diagnosis branches output object pointing results, auxiliary diagnosis results and diagnosis information interpretation chains respectively; The generation parameters of auxiliary diagnosis guidance information are optimized based on guidance feature supervision information, the object interpretation ability of multiple auxiliary diagnosis branches is optimized based on object pointing supervision information, the judgment parameters of interpretation closure degree are optimized based on interpretation closure state supervision information, and the branch occlusion relationship parameters between multiple auxiliary diagnosis branches are optimized based on branch occlusion relationship supervision information. When the guiding feature loss, object interpretation loss, interpretation loop determination loss, and branch occlusion relationship consistency loss meet the convergence condition, the trained joint auxiliary diagnosis model is obtained. Among them, the guiding feature loss, object interpretation loss, interpretation loop determination loss, and branch occlusion relationship consistency loss can correspond to one or more of the classification loss, ranking loss, regression loss, and structural consistency loss, respectively; preferably, multiple losses can be jointly optimized by setting weighting coefficients, and the weighting coefficients can be obtained by parameter tuning through the validation set; Based on the verification input set, determine the continuation threshold, preset confirmation threshold, first closed-loop threshold, second closed-loop threshold, preset backtracking threshold, and / or preset boosting threshold; Among them, the continuation threshold, preset confirmation threshold, first closed-loop threshold, second closed-loop threshold, preset backtracking threshold and / or preset enhancement threshold can preferably be determined by the optimization results of indicators on the verification input set. For example, the threshold that makes the F1 value, recall rate, stability index or switching accuracy reach the optimal value can be used as the corresponding preset threshold.

[0026] Specifically, in this embodiment, guidance feature supervision information, object pointing supervision information, explanation closed-loop state supervision information, and branch occlusion relationship supervision information are constructed based on historical diagnosis and treatment sample sets. The auxiliary diagnosis guidance information generation parameters, branch object interpretation ability, explanation closed-loop degree judgment parameters, and branch occlusion relationship parameters are jointly optimized. This helps to make the functional parts in the joint auxiliary diagnosis model form a coordinated and consistent optimization direction during the training phase, rather than converging independently. This improves the consistency and stability between auxiliary diagnosis guidance, dominant branch judgment, explanation closed-loop judgment, and branch switching logic, and provides a data basis for determining relevant threshold parameters.

[0027] In one embodiment, multiple auxiliary diagnostic branches are obtained, including: Based on the target interpretation object, obtain multiple candidate auxiliary diagnosis branches and determine the branch interpretation capability of multiple candidate auxiliary diagnosis branches; The target explanation object can be a set of target diseases, a set of target symptoms, a set of target syndromes, or a set of objects at its hierarchy to be included in the joint auxiliary diagnosis model; preferably, when obtaining multiple candidate auxiliary diagnosis branches based on the target explanation object, auxiliary diagnosis branches that have a label mapping relationship, object explanation relationship, or object similarity relationship with the target explanation object can be retrieved through the branch repository. The selection priority of multiple candidate auxiliary diagnostic branches is determined based on the overlap between the branch interpretation objects and the branch interpretation ability. Among them, branch explanation capability can refer to the comprehensive ability of candidate auxiliary diagnostic branches to identify, distinguish, rank, interpret results and adapt to time sequence of their branch explanation objects; preferably, branch explanation capability can be obtained through at least one of the following indicators: accuracy, recall, F1 score, AUC score, mean of explanation loop, coverage of key factors, confidence calibration error and cross-round stability. The overlap between the objects explained by the branches can be obtained by comparing the intersection ratio of the object sets corresponding to different candidate auxiliary diagnostic branches; preferably, the overlap of object sets, the overlap of typical object features, and the overlap of explanations of the same diagnostic factors can be considered at the same time to obtain a more accurate overlap result. Based on the selection priority, multiple candidate auxiliary diagnosis branches are selected sequentially as target auxiliary diagnosis branches, so that the coverage of the branch explanation objects of at least multiple target auxiliary diagnosis branches meets the preset threshold.

[0028] Specifically, the selection priority can be obtained in the following way: calculate the branch explanation ability score and redundancy suppression score for each candidate auxiliary diagnosis branch, wherein the redundancy suppression score is negatively correlated with the degree of overlap between it and the selected branch; then perform a weighted summation of the two to obtain the corresponding selection priority; the branch explanation object coverage can refer to the coverage ratio of the selected target auxiliary diagnosis branch to the entire set of target explanation objects, preferably, the preset threshold can be set to 0.8 to 0.95.

[0029] Specifically, in this embodiment, multiple candidate auxiliary diagnostic branches are first obtained based on the target explanation object and their branch explanation capabilities are determined. Then, the selection priority is determined by combining the overlap of branch explanation objects and branch explanation capabilities among the candidate auxiliary diagnostic branches. The target auxiliary diagnostic branches are then selected sequentially until the coverage of branch explanation objects meets the preset threshold. This helps to suppress the introduction of redundant branches while ensuring the coverage of explanation objects, thereby avoiding too many auxiliary diagnostic branches with highly overlapping explanation objects but limited contributions in the joint auxiliary diagnostic model, and improving the simplification of the model structure and the efficiency of branch configuration.

[0030] It is worth noting that in practical applications, auxiliary diagnostic capabilities are not static once built. As new disease knowledge, diagnostic rules, training samples, and better auxiliary diagnostic models emerge, some auxiliary diagnostic branches in the original joint auxiliary diagnostic model may have limited coverage of explanatory objects, insufficient explanatory power, or poor adaptability to new scenarios. Especially in rare diseases and complex disease courses, if the joint auxiliary diagnostic model relies solely on multiple auxiliary diagnostic branches pre-built during initialization, its overall auxiliary diagnostic capability is prone to lag over time.

[0031] Based on this, such as Figure 3 As shown, this application provides an embodiment in which, after organizing multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches based on branch occlusion relationships to generate a joint auxiliary diagnosis model, the method further includes: Set up a branch repository to store multiple candidate auxiliary diagnostic branches; The branch repository can be a local database, model repository, service registry or a combination thereof, used to store the parameter files, training versions, input and output interface definitions, branch interpretation object labels, performance evaluation results and scope of application of the auxiliary diagnostic branch; in addition to the branch addition event, the library change event can also include the branch update event, the branch offline event and the label modification event; Monitor library change events in the branch repository. If the library change event is a new branch event, then determine the new auxiliary branch and the corresponding branch explanation object. The branch explanation object corresponding to the newly added auxiliary diagnosis branch is compared with the branch explanation objects corresponding to multiple existing auxiliary diagnosis branches in the current joint auxiliary diagnosis model to determine whether there are overlapping branches with the same or overlapping branch explanation objects as the newly added auxiliary diagnosis branch. Among them, the overlapping branch can refer to an existing auxiliary diagnostic branch that has the same or high overlap with the newly added auxiliary diagnostic branch at the level of branch interpretation object; preferably, the existence of an overlapping branch can be determined when the degree of object overlap is greater than the object overlap threshold, and the object overlap threshold can be set to 0.5 to 0.8 for example. If there are no overlapping branches, the new auxiliary diagnosis branch is adapted based on the model structure of the current joint auxiliary diagnosis model. When the difference in branch interpretation accuracy before and after adaptation does not exceed the preset accuracy loss threshold, the branch occlusion relationship between the new auxiliary diagnosis branch and multiple existing auxiliary diagnosis branches in the current joint auxiliary diagnosis model is determined, and the new auxiliary diagnosis branch is integrated into the current joint auxiliary diagnosis model. The access adaptation process may include at least one of the following: input field mapping, input dimension alignment, output format normalization, confidence calibration, interface protocol adaptation, and relationship localization processing with the current branch occlusion relationship graph; a preset accuracy loss threshold can be used to limit the impact of new access on the overall performance of the existing joint auxiliary diagnosis model, and is preferably set to 1% to 5%; If there are overlapping branches, the auxiliary diagnostic branch with the highest degree of overlap of the branch explanation objects is selected as the reference branch, and the branch explanation object expansion degree of the newly added auxiliary diagnostic branch compared with the reference branch is determined. Based on the difference in the explanatory power of the newly added auxiliary diagnostic branch and the reference branch for the same or overlapping branch explanation objects, determine the degree of improvement in the branch explanation power of the newly added auxiliary diagnostic branch compared to the reference branch. Based on the differences in branch occlusion relationships between the newly added auxiliary diagnosis branch and the reference branch and other existing auxiliary diagnosis branches in the current joint auxiliary diagnosis model, the degree of correction of the branch occlusion relationship of the newly added auxiliary diagnosis branch compared with the reference branch is determined. Based on the branch explanation object expansion degree, branch explanation capability improvement degree, and branch occlusion relationship correction degree, the improvement degree of the newly added auxiliary diagnosis branch compared with the reference branch on the current joint auxiliary diagnosis model is determined. If the model lift meets the preset lift threshold, the reference branch will be replaced with the newly added auxiliary diagnosis branch, and the branch occlusion relationship between multiple auxiliary diagnosis branches in the current joint auxiliary diagnosis model will be updated.

[0032] Specifically, the branch explanation object expansion degree can be obtained in the following ways: first, count the number of objects, object levels, and typical characteristics of the newly added auxiliary diagnosis branch compared to the reference branch; then, normalize the newly added coverage relative to the original coverage of the reference branch to obtain the expansion degree; the branch explanation capability improvement degree can be obtained by comparing the performance index difference between the newly added auxiliary diagnosis branch and the reference branch on the same or overlapping branch explanation objects; the branch occlusion relationship correction degree can be obtained by comparing the degree of change caused by the occlusion direction, occlusion level, and switchable path between the two branches after they are connected to the joint auxiliary diagnosis model.

[0033] Preferably, the model improvement degree can be a weighted combination of the branch explanation object expansion degree, the branch explanation capability improvement degree, and the branch occlusion relationship correction degree; the preset improvement threshold can be set through the validation set, for example, a threshold that comprehensively improves the overall accuracy, rare disease recall rate, and path stability.

[0034] Specifically, in this embodiment, by setting up a branch repository and monitoring repository change events, when a new auxiliary diagnosis branch appears, the model improvement degree of the new auxiliary diagnosis branch is determined by combining the object overlap, the branch explanation object expansion degree, the branch explanation capability improvement degree, and the branch occlusion relationship correction degree. Based on this, new access or reference replacement is performed, which helps the joint auxiliary diagnosis model to have the ability to continuously evolve towards new diseases, new symptoms, or new specializations. Thus, without significantly damaging the stability of the existing model structure, better auxiliary diagnosis branches are dynamically absorbed and branch occlusion relationships are updated, improving the scalability and adaptability of the model throughout its life cycle.

[0035] In one embodiment, the currently dominant branch is determined among multiple routine auxiliary diagnosis branches and / or specialized auxiliary diagnosis branches based on auxiliary diagnosis guidance information, including: The auxiliary diagnostic guidance information is analyzed to obtain guidance features, and the results of the most recent historical branch determination are obtained; The most recent branch determination results may include at least one or more of the following: the dominant branch identifier determined in the previous round or the most recent rounds, the corresponding round, the corresponding interpretation closure degree, the current dominant auxiliary diagnosis result, the summary of the diagnosis and treatment information interpretation chain, and the branch switching status information. Determine whether there are direct guiding features among the guiding features, and whether there is a unique correspondence between a direct guiding feature and an auxiliary diagnostic branch in the joint auxiliary diagnostic model; Among them, the direct guiding feature can refer to the feature that forms a unique or highly unique correspondence with a certain auxiliary diagnostic branch; preferably, the unique correspondence can be determined by a pre-established "guiding feature - auxiliary diagnostic branch" mapping table. For example, a certain characteristic gene variation, enzyme activity defect, abnormal combination of metabolic spectrum or combination of imaging features can directly correspond to a certain specialized auxiliary diagnostic branch. If a direct guiding feature exists, the auxiliary diagnostic branch corresponding to the direct guiding feature is determined as the current dominant branch. If the direct guiding feature does not exist and the result of the most recent historical branch determination is not empty, then the auxiliary diagnostic branch corresponding to the result of the most recent historical branch determination is determined as the current priority branch. When the matching degree between the guiding feature and the typical features of the object of the pre-stored branch explanation object corresponding to the current priority branch meets the continuation threshold, the current priority branch is determined as the current dominant branch. Among them, the typical features of the object can refer to the set of features that can represent a certain branch to explain the main clinical manifestations, key examination abnormalities, characteristic temporal changes or high-discrimination biomarkers of the object; preferably, the typical features of the object can be generated jointly by the statistical results of the branch training samples, the medical knowledge base and / or expert annotations; If the direct guiding feature does not exist and the result of the most recent historical branch determination is empty, then the current dominant branch is determined from multiple normal auxiliary diagnosis branches based on the matching degree between the guiding feature and the typical features of the object of the branch explanation object corresponding to multiple normal auxiliary diagnosis branches.

[0036] Specifically, the matching degree can be obtained in the following way: first, the typical features of the guiding features and the objects of each branch explanation are represented as a unified feature vector, then the similarity, overlap or weighted matching score between the two is calculated, and finally, the continuity weight of historical rounds and the priority weight of the most recent round are introduced as needed to obtain the final matching degree; the continuation threshold can preferably be set through the validation set, for example, it can be set to 0.6 to 0.8.

[0037] Specifically, in this embodiment, the existence of a direct guiding feature uniquely corresponding to a certain auxiliary diagnostic branch is prioritized. When no direct guiding feature exists, the branch continuation judgment is made by combining the historical most recent branch determination result and the continuation threshold. When there is no historical continuation basis, a matching selection is made from multiple normal auxiliary diagnostic branches. This is beneficial to ensure that the determination of the current dominant branch takes into account strong pointing scenarios, continuous disease course scenarios, and initial uncertain scenarios. This improves the adaptability of the dominant branch determination to different diagnosis and treatment stages and enables direct features related to rare diseases to trigger the intervention of specialized auxiliary diagnostic branches more promptly.

[0038] In one embodiment, in response to the absence of a direct guiding feature, the determination result of the most recent historical branch being non-empty, and the matching degree between the guiding feature and the typical features of the object explained by the current priority branch not meeting the continuation threshold, the method further includes: Remove the branch continuation priority from the current priority branch; Among them, branch continuation priority refers to the priority continuation qualification naturally enjoyed by the most recent historical branch when there is no direct guiding feature; after the branch continuation priority is removed, the system may no longer regard the current priority branch as the default leading branch, and instead switch to the redirection judgment. Based on the matching degree between the guiding features and the typical features of the objects of the branch interpretation objects corresponding to multiple normal auxiliary diagnosis branches and / or auxiliary diagnosis branches that have a masking relationship with the current priority branch, at least one redirection candidate branch is determined. The redirection candidate branches may include multiple normal auxiliary diagnosis branches and / or auxiliary diagnosis branches that have a masking relationship with the current priority branch; preferably, when the candidate range is determined, auxiliary diagnosis branches that are in the same mixed auxiliary diagnosis branch cluster, the same mixed auxiliary diagnosis branch cluster group or have a direct masking relationship with the current priority branch can be given priority to reduce interference from irrelevant branches. The current dominant branch is determined based on the matching degree of one or more redirection candidate branches, the branch shading relationship with the current priority branch, and the continuity of the course of the disease in the historical rounds of diagnosis and treatment data.

[0039] Specifically, the continuity of the disease course can be characterized in the following ways: compare whether the main symptoms, key abnormal indicators, treatment response trends, and current round guidance features in the historical round diagnosis and treatment data are consistent with the typical disease course path of the candidate branch explanation object; the higher the consistency, the higher the continuity of the disease course; the current dominant branch can be determined based on the weighted result of the matching degree of the redirected candidate branch, the continuity of the disease course, and the distance of the branch occlusion relationship, among which the closer the distance of the branch occlusion relationship, the higher the priority.

[0040] Specifically, in this embodiment, when the direct guiding feature does not exist and the result of the most recent historical branch determination is not empty, but the matching degree of the current priority branch does not meet the continuation threshold, the branch continuation priority of the current priority branch is removed. The redirection candidate branch and the current dominant branch are determined by combining the guiding feature matching degree, branch occlusion relationship and disease course continuity. This helps to avoid the system rigidly maintaining the historical continuation result when the disease course deviates, so that the dominant branch can timely switch to a more matching auxiliary diagnosis path as the diagnosis and treatment status changes, thereby improving the response capability of the auxiliary diagnosis process to disease course deviation or abnormal evolution.

[0041] In one embodiment, the current dominant auxiliary diagnostic result is generated by parsing the target diagnosis and treatment data through the current dominant branch, and the degree of closure of the interpretation of the target diagnosis and treatment data by the current dominant branch is determined, including: Input the target diagnosis and treatment data into the current dominant branch so that the current dominant branch can extract diagnosis and treatment information factors from the target diagnosis and treatment data, and perform inference based on multiple diagnosis and treatment information factors to generate the object pointing result of the target diagnosis and treatment data relative to the branch interpretation object corresponding to the current dominant branch; Among them, the diagnosis and treatment information factor can refer to the smallest explanatory unit extracted from the target diagnosis and treatment data to support object-oriented reasoning, preferably including symptom factors, sign factors, test factors, imaging factors, pathological factors, genetic factors, treatment response factors, time evolution factors and combinations thereof; The object pointing result can refer to the pointing judgment result of the current dominant branch on the target diagnosis and treatment data relative to its branch interpretation object and related sub-objects; preferably, the object pointing result can be represented as an object candidate set, object ranking result, object pointing score, object probability distribution and / or object credibility score; Based on the object pointing result, the current dominant auxiliary diagnosis result is generated. The current dominant auxiliary diagnosis result at least characterizes the degree of object pointing and credibility of the target diagnosis data to the branch explanation object and / or its sub-objects. Based on the inference contribution of multiple diagnostic and treatment information factors to the object-oriented result, determine the factor explanatory power of the current dominant branch to multiple diagnostic and treatment information factors under the object-oriented result; Among them, factor explanatory power can refer to the degree of explanatory contribution of a certain diagnostic and treatment information factor to the formation of the current object-oriented result; specifically, factor explanatory power can be obtained by calculating the contribution value, attention weight, feature importance score, causal effect strength or rule triggering strength of each diagnostic and treatment information factor in the current dominant branch reasoning process, and normalizing them as the corresponding factor explanatory power; Based on multiple diagnostic and treatment information factors, their corresponding explanatory power, and the object-oriented results, a diagnostic and treatment information interpretation chain is constructed. The interpretation chain of diagnostic and treatment information can refer to an interpretation path organized by multiple on-chain factors that can jointly support the result pointed to by the current object, according to logical dependencies, temporal chronology, pathological causality, or evidentiary support. Preferably, diagnostic and treatment information factors with an explanatory power higher than the on-chain threshold and a supporting relationship with the result pointed to by the current object can be included in the interpretation chain. Factors not included in the interpretation chain but still present in the target diagnostic and treatment data can be identified as off-chain factors. Medical information factors that have been included in the medical information interpretation chain are identified as on-chain factors, and medical information factors that have not been included in the medical information interpretation chain are identified as off-chain factors. The explanatory closure degree is determined based on the factor explanatory power of on-chain factors, the proportion of off-chain factors, and the characterization criticality of off-chain factors.

[0042] Specifically, the criticality of the off-chain factor can be determined in the following ways: if the off-chain factor has high specificity, high severity, high risk indication significance, strong conflict with the current branch's explanatory object, or is newly added in the current round and has a large change range, its criticality can be set to high; the explanatory closure degree can be obtained by combining the average factor explanatory power of on-chain factors, the proportion of off-chain factors, and the penalty weight of key off-chain factors, and can be normalized to the range of 0 to 1. The larger the value, the more complete the current dominant branch's explanation of the target diagnosis and treatment data; Preferably, the criticality of the dechain factor can be determined based on the number of preset criticality conditions met by the dechain factor. When the dechain factor does not meet any preset criticality condition, its criticality is set to 0; when it meets one preset criticality condition, its criticality is set to 0.5; when it meets two or more preset criticality conditions, its criticality is set to 1. When determining the explanatory closure degree, the average factor explanatory power of the chain factors is used as the first evaluation result; the value obtained by subtracting the proportion of the dechain factor to all diagnostic and treatment information factors from 1 is used as the second evaluation result; the value obtained by subtracting the average criticality of all dechain factors from 1 is used as the third evaluation result. When there is no dechain factor, the proportion of the dechain factor and the average criticality are both set to 0. The reference ranges for the proportions of the first evaluation result, the second evaluation result, and the third evaluation result are 40% to 60%, 20% to 40%, and 10% to 30%, respectively, preferably 50%, 30%, and 20%, respectively. Weight coefficients are set according to the corresponding proportions and multiplied with the corresponding evaluation results. Finally, the explanatory closure degree in the range of 0 to 1 is obtained based on the sum of the products.

[0043] Specifically, in this embodiment, by extracting and reasoning about the target diagnosis and treatment data through the current dominant branch, an object-oriented result, the current dominant auxiliary diagnosis result, and a diagnosis and treatment information interpretation chain composed of on-chain factors and off-chain factors are formed. Based on this, the degree of interpretation closure is determined. This helps to further refine the interpretation capability of the current dominant branch for the target diagnosis and treatment data from a single result output to a quantifiable and traceable interpretation structure, thereby providing a unified evaluation basis for subsequent interpretation state division, instability detection, failure judgment, and branch switching.

[0044] In one embodiment, after determining the degree of closure of the current dominant branch's interpretation of the target diagnostic data, the method further includes: Based on the degree of explanation closure, the explanation state of the current dominant branch is divided to obtain the explanation normal state, explanation unstable state or explanation failed state. If the degree of explanation closure is greater than or equal to the first closure threshold and the proportion of the off-chain factor is less than or equal to the first proportion threshold, then the explanation state is determined to be the explanation normal state. The first closed-loop threshold and the second closed-loop threshold can be set based on the statistical results of the validation set, expert experience and actual switching stability requirements, and the first closed-loop threshold must be greater than the second closed-loop threshold. Preferably, the first closed-loop threshold can be set to 0.75 to 0.9 and the second closed-loop threshold can be set to 0.4 to 0.7. The first proportion threshold can be used to limit the proportion of the dechaining factor, and is preferably set to 0.1 to 0.3. If the explanatory closure degree is less than the first closure threshold and greater than the second closure threshold, and / or there is a critical disconnection factor among the disconnection factors that meets the preset critical conditions, then the explanatory state is determined to be an explanatory instability state. The preset key conditions can refer to the off-chain factor meeting at least one of the following: having high specificity for the target explanatory object, representing involvement of important organs, constituting a significant conflict with the existing object-oriented results, belonging to the key attention factors marked by doctors, or belonging to the factors with a significantly increased degree of new abnormality; off-chain factors that meet the preset key conditions can be identified as key off-chain factors; "high specificity" can refer to the fact that the corresponding diagnosis and treatment information factor has a discriminative ability in the target explanatory object exceeding the preset specificity threshold; "significant conflict" can refer to the fact that the difference between the object-oriented results corresponding to the new diagnosis and treatment information factor and the existing object-oriented results exceeds the preset conflict threshold; "significant increase" can refer to the fact that the change in indicators in the current round of diagnosis and treatment data exceeds the preset change threshold; each threshold can be determined based on the statistical results of historical case data or the performance indicators of validation samples and the results of expert evaluation, and is not further limited here.

[0045] If the degree of explanation closure is less than or equal to the second closure threshold, and / or the object pointing degree or credibility corresponding to the current dominant auxiliary diagnosis result is less than the preset failure threshold, then the explanation state is determined to be an explanation failure state; wherein, the first closure threshold is greater than the second closure threshold; Among them, the object pointing degree can refer to the pointing strength of the current dominant auxiliary diagnosis result to its optimal object candidate, and the credibility can refer to the stability, confidence level or the reliability after calibration of the result; the preset failure threshold can be selected through the validation set, for example, it can be set to 0.3 to 0.5; preferably, when multiple state judgment conditions are met at the same time, the interpretation state can be determined according to the principle of priority of the higher risk, that is, the interpretation failure state is given priority, followed by the interpretation instability state.

[0046] Specifically, in this embodiment, by combining the degree of explanation closure, the proportion of off-chain factors, the situation of key off-chain factors, and the degree of object pointing or credibility, the explanation state of the current dominant branch is divided into the explanation normal state, the explanation unstable state, and the explanation failure state. This is beneficial to distinguish the operating state of the current dominant branch into different levels of abnormality, so that the subsequent maintenance, takeover or re-dominance processing logic can correspond to different problem intensities, avoiding treating all explanation abnormalities as the same situation, which would lead to a coarse switching strategy.

[0047] In one embodiment, in response to the interpretation state indicating that the current dominant branch interpretation is normal, the method further includes: Maintain the current dominant branch as the current priority branch for subsequent rounds; Among them, the current priority branch can refer to the auxiliary diagnosis branch that is given priority to be used in subsequent rounds before the appearance of direct guiding features or obvious abnormal switching conditions; when the current dominant auxiliary diagnosis result, diagnosis and treatment information interpretation chain and interpretation closure degree are recorded as the historical most recent branch determination result, the current round timestamp, branch switching mark and key off-chain factors can also be recorded simultaneously for subsequent continued judgment and calling. Record the current dominant auxiliary diagnosis results, the chain of interpretation of diagnosis and treatment information, and the degree of closure of interpretation as the most recent historical branch to determine the results; In response to the existence of a candidate takeover branch in a pending takeover state, the pending takeover state of the candidate takeover branch is reduced or removed based on the closed-loop degree of the current dominant branch in subsequent rounds and the feature matching degree corresponding to the candidate takeover branch. The pending takeover status can refer to a transitional state in which the candidate takeover branch has not yet met the formal takeover conditions but has a certain potential takeover possibility; preferably, the pending takeover status can record the candidate takeover branch identifier, current feature matching degree, absorption degree, observation start round and effective observation period.

[0048] Specifically, reducing the pending takeover status of a candidate takeover branch can refer to lowering the takeover priority of the corresponding candidate takeover branch in subsequent rounds, shortening its retention time, or raising its subsequent confirmation threshold. Preferably, if the update explanation closure degree of the current dominant branch remains higher than the first closure threshold in consecutive preset rounds, and the update feature matching degree of the candidate takeover branch continuously decreases, its pending takeover status can be directly lifted.

[0049] Specifically, in this embodiment, when the interpretation state characterizes the current dominant branch as having a normal interpretation, it is maintained as the current priority branch for subsequent rounds. The current dominant auxiliary diagnosis result, the interpretation chain of diagnosis and treatment information, and the interpretation closure degree are recorded as the historical most recent branch determination results. At the same time, candidate takeover branches in the pending takeover state are reduced or removed. This is beneficial to maintain the continuity and stability of the auxiliary diagnosis process when the current diagnosis and treatment path still has good interpretability, thereby reducing unnecessary branch switching and avoiding interference to subsequent branch judgments caused by the long-term retention of the candidate takeover state.

[0050] In one embodiment, in response to the interpretation state indicating that the current dominant branch interpretation has failed, the method further includes: Remove the current dominant branch from the list of branches that are currently in control. Among them, the qualification for dominant continuation can refer to the qualification of the current dominant branch to be given priority in subsequent rounds; after the qualification for dominant continuation is revoked, the current dominant branch will no longer participate in the determination of the next round of dominant branch as the default continuation branch; If the updated auxiliary diagnosis guidance information contains a direct guidance feature, then the auxiliary diagnosis branch corresponding to the direct guidance feature will be identified as the new dominant branch. If there is no direct guidance feature in the updated auxiliary diagnosis guidance information, then based on the branch occlusion relationship, one or more occlusion auxiliary diagnosis branches that occlude the current dominant branch are obtained, and one or more occlusion auxiliary diagnosis branches are identified as backtracking candidate branches. Among them, the backtracking candidate branches can be selected from the superior auxiliary diagnosis branches that obscure the current dominant branch, the auxiliary diagnosis branches that have historically served as the dominant branch and have stable interpretations, and the normal auxiliary diagnosis branches that have a high degree of matching with the currently updated auxiliary diagnosis guidance information; the re-dominance adaptation degree can refer to the ability of the candidate backtracking branch to take over the current target diagnosis and treatment data again. The target diagnosis and treatment data are parsed using one or more backtracking candidate branches, one or more backtracking candidate auxiliary diagnosis results are generated, and the degree of re-dominance of one or more backtracking candidate branches to the target diagnosis and treatment data is determined. Specifically, the re-dominance fit can be obtained by weighting the candidate backtracking branches according to the following indicators: matching degree of guiding features of the updated target diagnosis and treatment data, estimated explanatory closure degree, matching degree with the continuity of historical disease course, and distance of the current branch occlusion path. The preset backtracking threshold can be preferably set through the validation set, for example, it can be set to 0.6 to 0.8. If there are multiple backtracking candidate branches that meet the preset backtracking threshold, the one with the highest re-dominance fit and the shortest branch occlusion path distance can be selected as the re-dominance branch. If there is a backtracking candidate branch whose re-dominance meets the preset backtracking threshold, then the re-dominance branch is determined from it, and the subsequent auxiliary diagnosis is performed by the re-dominance branch. If there is no candidate branch for backtracking that satisfies the preset backtracking threshold, the current dominant branch is re-determined among multiple normal auxiliary diagnosis branches based on the updated auxiliary diagnosis guidance information.

[0051] Specifically, in this embodiment, when the interpretation state characterizes the current dominant branch as having failed, its dominant continuation qualification is revoked, and the dominant branch is re-determined based on the direct guidance features in the updated auxiliary diagnosis guidance information, or the occluded auxiliary diagnosis branch is selected as a backtracking candidate branch based on the branch occlusion relationship and a new dominant branch is determined. This helps to distinguish between interpretation failure and interpretation instability and process them accordingly. Thus, when the current dominant branch can no longer bear the effective interpretation task, its continuation is terminated in time and it returns to a more suitable superior path or a new path is selected, thereby improving the accuracy and efficiency of dominant branch reconstruction.

[0052] In one embodiment, in response to the interpretation closure degree characterizing the instability of the current dominant branch, one or more obscured auxiliary diagnostic branches associated with the current dominant branch are obtained based on the branch obscuration relationship, and at least one candidate takeover branch is determined from the one or more obscured auxiliary diagnostic branches, including: Based on the branch occlusion relationship, identify one or more occluded auxiliary diagnostic branches associated with the current dominant branch, and determine the occlusion level of the current dominant branch on one or more occluded auxiliary diagnostic branches. The depth level of the current dominant branch and one or more corresponding occluded auxiliary branches is obtained by mapping the occlusion level. The depth level at least indicates that there is an occlusion relationship between the auxiliary branches in the shallow layer and the auxiliary branches in the deep layer, and the occlusion level of the auxiliary branches in the shallow layer to the auxiliary branches in the deep layer increases with the difference in branch depth. The depth level can refer to the relative hierarchical position of each auxiliary diagnosis branch in the branch occlusion relationship map; preferably, the layer where the normal auxiliary diagnosis branch is located can be defined as the shallower layer, and the layer where the specialized auxiliary diagnosis branch that forms an occlusion relationship with it is defined as the deeper layer, and the level can be determined step by step according to the occlusion level or the map topology. The next depth level after the current dominant branch is taken as the target depth level for diving, and one or more occluded auxiliary branches in the target depth level for diving are identified as candidate branches for diving. Obtain diagnostic and treatment information to interpret the chain, on-chain factors, and off-chain factors; Based on the branch interpretation objects corresponding to one or more potential candidate branches, determine the positive explanatory power of one or more potential candidate branches for off-chain factors, as well as their compatibility and maintenance capabilities for the interpretation chain of diagnostic and treatment information and factors on the chain. Among them, the positive explanatory power of the off-chain factor can refer to the ability of the potential candidate branch to interpret the off-chain factor that was not originally included in the explanatory chain by the current dominant branch as supporting evidence for its branch explanatory object; specifically, it can be obtained by comparing the typical features of the off-chain factor and the potential candidate branch object, the set of key supporting factors, and the co-occurrence relationship in historical training samples; the compatibility and maintenance ability of the diagnostic and treatment information explanatory chain and on-chain factors can refer to the ability of the potential candidate branch to continue to retain and take over the existing on-chain factors while explaining the newly added off-chain factor; Based on the positive explanatory power of the off-chain factors and the compatibility and maintenance of the diagnostic and treatment information interpretation chain and on-chain factors, the takeover adaptation degree of one or more potential candidate branches to the current dominant branch is determined respectively. Based on the takeover suitability, at least one candidate takeover branch is determined from one or more dive candidate branches.

[0053] Specifically, the depth level can be obtained by mapping the occlusion level in the following way: First, determine the superior occlusion branch and the inferior occlusion branch of each auxiliary diagnosis branch according to the branch occlusion relationship graph; then, start from the auxiliary diagnosis branch with no superior occlusion branch or the lowest occlusion level as the starting layer, and number the layers downwards along the occlusion direction to obtain the depth level corresponding to each auxiliary diagnosis branch; preferably, when diving, the next depth level of the current dominant branch is selected first to avoid excessive switching caused by crossing too many levels.

[0054] Specifically, the takeover fit can be obtained as follows: For each candidate branch, calculate its positive explanatory power score for the off-chain factors and its compatibility and maintenance score for the existing diagnostic and treatment information interpretation chain and on-chain factors, and then sum the two scores with weights to obtain the takeover fit. Preferably, the weights can be set according to the complexity of the disease, for example, increasing the weight of positive explanatory power when there are many off-chain factors, and increasing the weight of compatibility and maintenance when the existing chain is relatively stable.

[0055] Specifically, in this embodiment, when the interpretation of the current dominant branch becomes unstable, the associated occluded auxiliary diagnostic branches are obtained based on the branch occlusion relationship. Furthermore, the takeover fit is determined by combining the depth level mapped by the occlusion level, the target of the next depth level, and the positive explanatory ability of the candidate takeover branches for the off-chain factors and their compatibility and maintenance ability with the existing diagnostic information interpretation chain. This is beneficial to ensure that the branch switching in the unstable state is not an indiscriminate reselection among all auxiliary diagnostic branches, but rather a directional dive to a deeper and more refined interpretation path along the established occlusion relationship, thereby improving the targeting of the candidate takeover branches and the structural rationality of the switching path.

[0056] In one embodiment, candidate auxiliary diagnostic results are generated by parsing target diagnostic data through candidate takeover branches, and it is determined whether the candidate takeover branch meets the branch takeover conditions for the current dominant branch. If it does, the candidate takeover branch is determined as the subsequent dominant branch to perform subsequent auxiliary diagnoses, including: Maintain the current dominant branch to execute the current round of auxiliary diagnosis, and input the target diagnosis and treatment data into at least one candidate takeover branch, so that the candidate takeover branch can perform factor extraction and inference based on the target diagnosis and treatment data, generate candidate object pointing results, and generate corresponding candidate auxiliary diagnosis results and candidate takeover diagnosis and treatment information interpretation chain; Among them, the format of the candidate object pointing result and the candidate auxiliary diagnosis result can be consistent with the output of the current dominant branch so as to make horizontal comparisons; the candidate takeover diagnosis and treatment information interpretation chain can be the interpretation path constructed by the candidate takeover branch for the target diagnosis and treatment data, which is used to evaluate whether it can effectively absorb the off-chain factors of the current dominant branch. Based on the degree of inclusion of the off-chain factor corresponding to the current dominant branch in the interpretation chain of one or more candidate takeover diagnosis and treatment information, determine the degree of absorption of the off-chain factor by one or more candidate takeover branches respectively; Pre-inference is performed based on one or more candidate takeover diagnosis and treatment information interpretation chains to obtain the predicted diagnosis and treatment features corresponding to one or more candidate takeover branches respectively; The predicted diagnostic features refer to the diagnostic features that the candidate takeover branch infers based on the existing target diagnostic data and the candidate takeover diagnostic information interpretation chain, which may appear, be enhanced, or persist in subsequent rounds. Preferably, the inference can be performed through time series models, rule deduction, knowledge graph deduction, or a combination thereof. New diagnostic features refer to diagnostic features that appear, are significantly enhanced, or undergo changes in status in subsequent rounds of diagnostic data compared to the previous round. In response to receiving subsequent rounds of treatment data, new treatment features are extracted from the subsequent rounds of treatment data, and the feature matching degree between the new treatment features and the predicted treatment features corresponding to one or more candidate takeover branches is determined respectively; Based on subsequent rounds of diagnosis and treatment data and the current dominant branch, determine the updated explanatory closure degree of the current dominant branch in subsequent rounds, and determine the trend of the explanatory closure degree of the current dominant branch based on the updated explanatory closure degree and the explanatory closure degree of the previous rounds. Based on the feature matching degree corresponding to one or more candidate takeover branches and the trend of the change in the explanatory closure degree of the current dominant branch, a set of candidate takeover branches that meet the branch takeover conditions is determined. The branch takeover conditions include at least: the corresponding feature matching degree meets the preset confirmation threshold, and the trend of the change in the explanatory closure degree indicates that the explanatory closure degree of the current dominant branch continues to decrease. If the candidate takeover branch set contains only one candidate takeover branch, the candidate takeover branch is determined as the subsequent dominant branch to replace the current dominant branch and perform subsequent auxiliary diagnostics. In response to the presence of multiple candidate takeover branches in the candidate takeover branch set, the subsequent dominant branch is determined based on the feature matching degree and / or the degree of absorption of the off-chain factor corresponding to the multiple candidate takeover branches. If the set of candidate takeover branches is empty, the current dominant branch is maintained, and the takeover status of at least one candidate takeover branch is kept pending confirmation.

[0057] Specifically, the absorption degree of off-chain factors can be obtained by: statistically analyzing the proportion of off-chain factors corresponding to the current dominant branch that are included as on-chain factors or interpreted as key supporting factors by the candidate takeover treatment information interpretation chain; if factor importance is also considered, the absorption degree can be obtained by further weighting according to the key weight of each off-chain factor.

[0058] Specifically, the feature matching degree can be obtained by comparing the consistency between the predicted diagnosis and treatment features corresponding to the candidate takeover branch and the newly added diagnosis and treatment features in subsequent rounds in terms of feature type, directionality, magnitude of change, time position, and feature importance, and then normalizing the comparison results. The trend of the explained closure degree can be obtained by comparing the direction of change, magnitude of change, or slope of the sliding window of the explained closure degree in previous rounds and subsequent rounds. Preferably, when the explained closure degree continuously decreases within consecutive preset rounds, it can be considered to meet the trend condition of "continuous decrease". The preset confirmation threshold can preferably be set to 0.7 to 0.9.

[0059] Furthermore, when the candidate takeover branch set includes multiple candidate takeover branches, the subsequent dominant branch can be determined by weighting and sorting the feature matching degree and the degree of off-chain factor absorption. If necessary, the interpretation closed-loop prediction value and the branch occlusion level distance of the candidate takeover branches can also be further referenced.

[0060] Specifically, in this embodiment, while maintaining the current dominant branch in performing the current round of auxiliary diagnosis, at least one candidate takeover branch is introduced to perform parallel analysis of the target diagnosis and treatment data. The branch takeover conditions are determined by combining the degree of absorption of the off-chain factors by the candidate takeover diagnosis and treatment information interpretation chain, the predicted diagnosis and treatment features obtained by pre-inference, the feature matching degree between the newly added diagnosis and treatment features in subsequent rounds and the predicted diagnosis and treatment features, and the changing trend of the interpretation closure degree of the current dominant branch. This is beneficial to base the takeover judgment on the verification of future diagnosis and treatment evolution, rather than directly switching based on the local anomalies in the current round, thereby reducing the risk of false switching and improving the reliability of the confirmation of subsequent dominant branch replacement.

[0061] In one embodiment, after maintaining the current dominant branch and retaining the pending takeover status of at least one candidate takeover branch in response to an empty set of candidate takeover branches, the method further includes: In response to receiving new rounds of treatment data, update the predicted treatment features, feature matching degree, and absorption degree of off-chain factors corresponding to the candidate takeover branches that are in the pending takeover status; The pending takeover status can be set with an expiration period. Preferably, the expiration period can correspond to a preset number of rounds, such as 1 to 3 rounds, or a preset time length. If the status has not been changed to confirmed takeover after the expiration period, it can be automatically deactivated. Specifically, updating the predicted diagnosis and treatment features, feature matching degree, and absorption degree of off-chain factors corresponding to the candidate takeover branch that is in the pending takeover state can be achieved by re-inputting the new round of diagnosis and treatment data into the candidate takeover branch, re-executing the candidate diagnosis and treatment information interpretation chain construction and predicted feature generation process, and performing incremental comparison with the previous results. Based on the update explanation of the current dominant branch and the update feature matching degree of the candidate takeover branch in the pending takeover state, determine whether the pending takeover state has been converted to the confirmed takeover state. Specifically, whether the pending takeover status is converted to a confirmed takeover status can be determined as follows: if the updated feature matching degree is higher than the confirmation threshold for a consecutive preset number of rounds, and the trend of the update explanation closure degree of the current dominant branch continues to decline, then the pending takeover status is converted to a confirmed takeover status; the preset revocation threshold is preferably lower than the preset confirmation threshold to form a hysteresis interval between confirmation and revocation, thereby reducing frequent fluctuations in the candidate takeover status. Preferably, the preset revocation threshold can be set to 0.4 to 0.6; In response to the transition from pending takeover status to confirmed takeover status, the corresponding candidate takeover branch is included in the candidate takeover branch set to determine the subsequent dominant branch; If the explanation closure degree of the current dominant branch returns to the normal explanation state, and / or the update feature matching degree of the candidate takeover branch in the pending takeover state is lower than the preset revocation threshold, then the corresponding pending takeover state is lifted.

[0062] Specifically, in this embodiment, when the candidate takeover branch set is empty, at least one candidate takeover branch is retained in the pending takeover state. After the arrival of new rounds of diagnosis and treatment data, its predicted diagnosis and treatment characteristics, feature matching degree and absorption degree of off-chain factors are continuously updated. Then, combined with the current dominant branch's explanation of the closure degree change trend, it is decided whether to switch to the confirmed takeover state or remove the pending takeover state. This is beneficial to retain a flexible observation window in the transition stage when the dominant branch has not completely lost the basis for continuation and the potential takeover branch has not been fully verified, thereby improving the robustness and continuity of the dominant branch switching decision.

[0063] In one embodiment, after designating the candidate takeover branch as the subsequent leading branch and performing subsequent auxiliary diagnostics, the method further includes: In response to the update of the target diagnosis and treatment data, it is determined whether the branch back-cutting trigger condition is met. The branch back-cutting trigger condition is used to characterize that the phased auxiliary diagnosis task corresponding to the subsequent dominant branch has been completed, and / or to characterize that at least one occluded auxiliary diagnosis branch that occludes the subsequent dominant branch meets the dominant adaptation condition again for the updated target diagnosis and treatment data. Among them, the branch back-cutting trigger conditions may include at least one of the following: the completion rate of the phased auxiliary diagnosis task corresponding to the subsequent dominant branch reaches the preset completion threshold, the matching degree of the updated auxiliary diagnosis guidance information with the typical features of the object of the superior auxiliary diagnosis branch that obscures it increases again, the specialized interpretation advantage of the subsequent dominant branch for the new data decreases, and the estimated interpretation closure degree of the superior auxiliary diagnosis branch is higher than that of the subsequent dominant branch. Among them, the phased auxiliary diagnostic task can refer to the specific verification, screening, attribution, classification or exclusion tasks to be completed after diving into the subsequent dominant branch; the completion rate of the phased auxiliary diagnostic task can be obtained in the following way: statistically analyze the proportion of key diagnostic features, key examination items or key classification conditions that the subsequent dominant branch needs to focus on, which have been met, explained or excluded, and combine them with the weight of key conditions to obtain the completion rate result. Preferably, the branch cutback trigger condition can require one or more consecutive rounds to be met in order to reduce premature cutback caused by short-term fluctuations; the preset completion threshold can be set through the validation set and expert experience, for example, it can be set to 0.7 to 0.9; If the branch back-cutting trigger condition is met, based on the branch occlusion relationship, one or more occlusion auxiliary diagnostic branches that occlude the subsequent dominant branch are obtained, and at least one candidate back-cutting branch is determined from the one or more occlusion auxiliary diagnostic branches. The updated target diagnosis and treatment data is analyzed by candidate back-cutting branches to generate back-cutting candidate auxiliary diagnosis results, and it is determined whether the candidate back-cutting branches meet the branch back-cutting conditions for the subsequent dominant branches. If they meet the conditions, the candidate back-cutting branches are determined as the dominant branches after back-cutting to perform subsequent auxiliary diagnoses.

[0064] Specifically, in this embodiment, after a candidate takeover branch becomes the subsequent dominant branch, the system further updates the target diagnosis and treatment data to determine whether the branch back-cutting trigger condition is met. When the condition is met, it obtains one or more occluding auxiliary diagnosis branches that occlude the subsequent dominant branch based on the branch occlusion relationship, thereby determining the candidate back-cutting branch. This is beneficial because after the system completes the takeover, it does not remain permanently in a deeper or more specialized auxiliary diagnosis path, but can implement reverse back-cutting when the phased task is completed or the upper-level auxiliary diagnosis path regains its adaptability, thereby maintaining the dynamic balance between deep specialized interpretation and upper-level overall interpretation in the entire auxiliary diagnosis process.

[0065] In one embodiment, the updated target diagnosis and treatment data is parsed through candidate back-cutting branches to generate back-cutting candidate auxiliary diagnosis results, and it is determined whether the candidate back-cutting branches meet the branch back-cutting conditions for the subsequent dominant branch, including: The updated target diagnosis and treatment data is input into at least one candidate back-cutting branch, so that the candidate back-cutting branch generates back-cutting candidate object pointing result, back-cutting candidate auxiliary diagnosis result and back-cutting candidate diagnosis and treatment information interpretation chain; Among them, the output format of the candidate object pointing back, the candidate auxiliary diagnosis result back, and the interpretation chain of the candidate diagnosis and treatment information back can preferably be consistent with the corresponding output format of the current dominant branch and the candidate takeover branch, so as to make a unified comparison. Based on the phased task completion rate of the branch explanation object corresponding to the subsequent leading branch, determine whether the specialized auxiliary diagnosis phase of the subsequent leading branch has been completed. Whether the specialized auxiliary diagnosis stage is completed can be determined based on at least one of the following: the key specialized object corresponding to the subsequent dominant branch has been identified or excluded, the key subtyping task has been completed, the key decoupling factor has been absorbed and explained, and the stage verification indicators have met the preset requirements; the completion degree of the stage task can preferably use the same or compatible calculation logic as the completion degree in the aforementioned back-cutting trigger conditions. Based on the updated target diagnosis and treatment data, determine the degree of re-dominance adaptation of one or more candidate back-cutting branches to the updated target diagnosis and treatment data, as well as the ability to maintain upper-level compatibility with the interpretation chain of diagnosis and treatment information corresponding to the subsequent dominant branches. Based on the completion rate of the phased task, the degree of re-leadership adaptation, and the ability to maintain compatibility with higher-level systems, a set of candidate back-cutting branches that meet the branch back-cutting conditions is determined. If the set of candidate backcut branches contains only one candidate backcut branch, then the candidate backcut branch is determined as the dominant branch after the backcut. In response to the fact that the candidate back-cutting branch set includes multiple candidate back-cutting branches, the dominant branch after back-cutting is determined based on the re-dominance adaptation degree and / or the upper-level compatibility preservation capability corresponding to the multiple candidate back-cutting branches. If the set of candidate back-cut branches is empty, the subsequent dominant branch will continue to perform subsequent auxiliary diagnostics.

[0066] Specifically, the re-dominance fit can be obtained by comparing the degree of matching of the guiding features of the candidate back-cutting branch with the updated target diagnosis and treatment data, the stability of the object pointing, the estimated explanatory closure degree, and the degree of fit with the continuity of the historical course of the disease, and then summing the above results with weights; the ability to maintain superior compatibility can refer to the ability of the candidate back-cutting branch to retain and accept the effective on-chain factors in the diagnosis and treatment information interpretation chain established by the subsequent dominant branch after back-cutting. This can be obtained by statistically analyzing the proportion of on-chain factors of the subsequent dominant branch that are continued to be included in its interpretation chain by the candidate back-cutting branch or can be smoothly mapped to its superior interpretation chain, and combining the weights of key on-chain factors.

[0067] Specifically, the branch reversal criteria can be determined by comprehensively considering the completion rate of the phased task, the degree of re-dominance adaptation, and the ability to maintain higher-level compatibility. Preferably, a comprehensive reversal score threshold can be set. When the comprehensive reversal score of a candidate reversal branch is higher than the threshold, it is included in the candidate reversal branch set. If multiple candidate reversal branches meet the criteria, the one with higher re-dominance adaptation and stronger ability to maintain higher-level compatibility can be selected as the dominant branch after the reversal. If none of them meet the criteria, the subsequent dominant branch is maintained to continue performing subsequent auxiliary diagnosis.

[0068] Specifically, in this embodiment, the updated target diagnosis and treatment data is input into the candidate back-cutting branch to generate the back-cutting candidate object pointing result, the back-cutting candidate auxiliary diagnosis result, and the back-cutting candidate diagnosis and treatment information interpretation chain. The branch back-cutting conditions are determined by combining the completion degree of the corresponding stage task of the subsequent leading branch, the degree of re-leading adaptation of the candidate back-cutting branch to the updated target diagnosis and treatment data, and the ability to maintain the upper-level compatibility of the diagnosis and treatment information interpretation chain of the subsequent leading branch. This helps to establish the back-cutting judgment on the common constraints of "whether the down-stage is completed" and "whether the upper-level path can be stably inherited", thereby avoiding the back-cutting action from becoming a simple reverse switch and improving the stability of the leading branch after back-cutting in inheriting the subsequent auxiliary diagnosis process.

[0069] It is worth noting that in actual clinical practice, target diagnostic and treatment data can represent not only a single disease, a single symptom, or a single stage of disease, but also two or more coexisting explanatory objects simultaneously. For example, underlying diseases and complications may coexist, multiple symptoms may coexist, or a dominant symptom and accompanying symptoms may coexist. In this case, some diagnostic and treatment information factors in the target data can be explained by the diagnostic and treatment information explanation chain corresponding to the current dominant branch, while other diagnostic and treatment information factors are not necessarily used to represent the instability or failure of the current dominant branch; they can also be used to represent other coexisting explanatory objects besides those explained by the branch corresponding to the current dominant branch.

[0070] Based on this, in one embodiment provided by this application, after determining the current dominant branch, it is further possible to identify whether there are mixed symptoms and / or complications in the target medical data, specifically including: If the interpretation state corresponding to the current dominant branch is either in a normal interpretation state or an unstable interpretation state, and there are remaining unexplained interpretation information factors in the target diagnosis and treatment data, then the method further includes: Based on the remaining unexplained diagnostic and treatment information factors, at least one candidate coexisting factor group is identified; Among them, the remaining unexplained diagnostic and treatment information factors may refer to off-chain factors that have not been absorbed by the diagnostic and treatment information interpretation chain corresponding to the current dominant branch, and / or diagnostic and treatment information factors that, although included in the diagnostic and treatment information interpretation chain, have an explanatory power lower than the preset interpretation threshold; the candidate coexisting factor group may refer to a factor group consisting of multiple diagnostic and treatment information factors that are related in organ systems, occur at similar times, have similar characteristics, or point to the same additional explanatory object in clinical sense. Based on at least one candidate coexistence factor group, at least one concurrent candidate branch is determined from multiple auxiliary diagnostic branches; Call at least one concurrent candidate branch to identify object pointers for at least one candidate coexistence factor group and obtain at least one concurrent candidate object pointer result; Based on the result pointed to by at least one concurrent candidate object, determine whether the target diagnostic data corresponds to at least one coexisting interpretation object; In response to the target diagnostic data corresponding to at least one coexisting explanatory object, while maintaining the current dominant branch, the auxiliary diagnostic branch corresponding to at least one coexisting explanatory object is determined as the concurrent identification branch, and a hybrid auxiliary diagnostic result is generated based on the current dominant branch and at least one concurrent identification branch.

[0071] In this embodiment, based on the remaining unexplained diagnostic and treatment information factors, at least one candidate coexisting factor group is determined, including: The remaining unexplained diagnostic and treatment information factors are clustered or associated with each other to obtain multiple candidate factor sets. Based on the co-occurrence relationship, temporal proximity relationship and / or organ system association relationship among the diagnostic and treatment information factors in multiple candidate factor sets, at least one candidate co-occurrence factor group is determined from multiple candidate factor sets.

[0072] Preferably, the remaining unexplained diagnostic and treatment information factors are clustered using a hierarchical agglomerative clustering algorithm with average connectivity, including: Based on the factor type, abnormal direction, and normalized abnormality degree of each remaining unexplained diagnostic and treatment information factor, a corresponding factor feature vector is constructed; wherein, the factor type includes at least one of symptoms, signs, laboratory test results, imaging examination results, and pathological results, and the abnormal direction includes elevation, reduction, positivity, negativity, or morphological abnormality; Each remaining unexplained diagnostic and treatment information factor is used as an initial cluster. The feature similarity between any two diagnostic and treatment information factors is determined based on the factor feature vectors. The average feature similarity between the diagnostic and treatment information factors in the two clusters is used as the cluster similarity between the two clusters. The feature similarity can be determined using cosine similarity. Two clusters whose cluster similarity reaches a preset clustering threshold are merged, and the cluster similarity is determined and the clusters are merged repeatedly until the cluster similarity between any two clusters does not reach the preset clustering threshold. The clusters that ultimately contain at least two remaining unexplained diagnostic and treatment information factors are determined as the candidate factor set; wherein, the reference range of the preset clustering threshold is 0.60 to 0.80, preferably 0.70; For example, when multiple remaining unexplained diagnostic information factors belong to the same factor type, have the same abnormal direction, and the cluster similarity between their corresponding feature vectors reaches 0.70, they can be classified into the same candidate factor set.

[0073] Preferably, the remaining unexplained diagnostic and treatment information factors are subjected to association aggregation processing, which can be performed using the FP-Growth association rule algorithm, including: The remaining unexplained treatment information factors are organized according to the treatment round, and the remaining unexplained treatment information factors appearing in the same treatment round are formed into a set of transaction items. The number of occurrences of a factor itemset consisting of at least two remaining unexplained diagnostic information factors in all valid diagnostic rounds is counted, and the proportion of the occurrences to the total number of valid diagnostic rounds is determined as the support of the corresponding factor itemset. The set of factor itemsets whose support reaches a preset support threshold is determined as the candidate factor set; The reference range for the preset support threshold is 0.20 to 0.40, preferably 0.30; For example, if three remaining unexplained treatment information factors appear together in three out of ten valid treatment rounds, the support of the corresponding factor itemset is 0.30, and the factor itemset can be identified as a candidate factor set.

[0074] Specifically, for any set of candidate factors, when at least two of them appear in the same treatment round, a feature co-occurrence relationship is determined between the at least two treatment information factors; when at least two of them appear in two adjacent treatment rounds, a temporal proximity relationship is determined between the at least two treatment information factors; when at least two of them belong to the same organ system, or their respective organ systems are directly associated in a pre-stored organ system association table, an organ system association relationship is determined between the at least two treatment information factors; the set of candidate factors that satisfies at least one of the feature co-occurrence relationship, temporal proximity relationship, or organ system association relationship is determined as a candidate coexisting factor group, preferably the set of candidate factors that satisfies at least two of them is determined as the candidate coexisting factor group; For example, if two diagnostic information factors in a candidate factor set appear in adjacent diagnostic rounds and belong to the same organ system, the candidate factor set can be identified as a candidate coexisting factor group.

[0075] In this embodiment, based on at least one candidate coexistence factor group, at least one concurrent candidate branch is determined from multiple auxiliary diagnostic branches, including: Remove auxiliary diagnostic branches from multiple auxiliary diagnostic branches that have an exclusive relationship or a strong substitution relationship with the current dominant branch; Among them, object exclusivity relationship can refer to the relationship that the two branches of the explanatory object are usually not established at the same time, or the explanatory necessity of the other branch of the explanatory object is significantly reduced when the explanatory object of one branch is established; strong substitution relationship can refer to the relationship that once the corresponding auxiliary diagnosis branch is established, it is more suitable as the current dominant branch of the target diagnosis and treatment data for overall interpretation. At least one concurrent candidate branch is determined based on the matching degree between at least one candidate coexistence factor group and the typical features of the objects corresponding to multiple remaining auxiliary diagnostic branches.

[0076] In this embodiment, at least one concurrent candidate branch is invoked to identify object pointers for at least one candidate coexistence factor group, resulting in at least one concurrent candidate object pointer result, including: Input at least one candidate coexistence factor group into at least one concurrent candidate branch; Each concurrent candidate branch outputs its corresponding concurrent candidate auxiliary diagnosis result and concurrent candidate diagnosis and treatment information interpretation chain. Based on the concurrent candidate auxiliary diagnosis results and the concurrent candidate diagnosis and treatment information interpretation chain, at least one concurrent candidate object is identified as pointing to a result.

[0077] In this embodiment, determining whether the target diagnostic data corresponds to at least one coexisting interpretation object based on the result of at least one concurrent candidate object includes: Determine the factor coverage of at least one concurrent candidate branch for the corresponding candidate coexisting factor group; Determine the degree to which at least one concurrent candidate branch absorbs the off-chain factor corresponding to the current dominant branch; Determine the object coexistence compatibility and inter-chain compatibility between at least one concurrent candidate branch and the current dominant branch; Among them, object coexistence compatibility can refer to the degree of possibility that the branch interpretation object corresponding to the current dominant branch and the branch interpretation object corresponding to the concurrent candidate branch may coexist, co-occur, be secondary to or occur in parallel in clinical practice; chain compatibility can refer to the degree of conflict or continuity between the concurrent candidate diagnosis and treatment information interpretation chain corresponding to the concurrent candidate branch and the diagnosis and treatment information interpretation chain corresponding to the current dominant branch while interpreting the candidate coexistence factor group. Based on factor coverage, absorption, object coexistence compatibility, and inter-chain compatibility, determine the concurrent support of at least one concurrent candidate branch; In response to the existence of concurrent candidate branches with a concurrent support greater than a preset concurrent threshold, at least one coexisting interpretation object is determined to correspond to the target diagnostic data.

[0078] In this embodiment, in response to the existence of multiple concurrent candidate branches whose concurrent support respectively meets a preset concurrent threshold, the multiple concurrent candidate branches can be jointly determined as a set of concurrent identification branches; at this time, a hybrid auxiliary diagnosis result is generated based on the current dominant branch and at least one concurrent identification branch, including: Determine the dominant interpretation object corresponding to the current dominant branch; Identify the coexistence interpretation objects corresponding to at least one concurrent identification branch; Based on the order of occurrence, direction of object association, triggering relationship of key factors and / or propagation relationship between the dominant explanatory object and at least one coexisting explanatory object, determine the relationship label between at least one coexisting explanatory object and the dominant explanatory object; The relationship label may include at least one of the following: coexisting disease relationship, accompanying disease relationship, complication relationship, and secondary relationship; Based on the dominant interpreter, at least one coexisting interpreter, and at least one relation label, a hybrid auxiliary diagnosis result is generated.

[0079] In this embodiment, in response to at least one coexisting explanatory object being induced, progressed, or triggered by the dominant explanatory object, the corresponding coexisting explanatory object is identified as a complication object; in response to at least one coexisting explanatory object and the dominant explanatory object not having a clear triggering direction but allowing stable coexistence, the corresponding coexisting explanatory object is identified as a mixed disease object or a coexisting disease object. The preset concurrency threshold can be determined by historical sample statistics, validation set optimization results, and / or expert experience. Preferably, the preset concurrency threshold can be set to 0.6 to 0.85, but is not limited to this.

[0080] Specifically, through the above implementation method, when the current dominant branch still maintains its ability to explain the main treatment pathway, the switching of the dominant branch is not directly triggered. Instead, the additional explanatory objects represented by the remaining unexplained treatment information factors are further identified. This helps to distinguish between the two different situations of "insufficient explanation by the current dominant branch" and "simultaneous coexistence of multiple symptoms or complications", thereby improving the ability to identify mixed symptoms and complications and maintaining the continuity of the dominant treatment pathway.

[0081] It should be understood that, although Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0082] In some embodiments of this application, the term "round" may refer to any one of the following: a single medical visit, a round of consultations, a single examination update, or a data update cycle within a preset time window; the term "object" may refer to a disease, symptom, syndrome, complication, disease stage, subtype, or a combination thereof. The term "preset threshold" and "preset conditions" may be determined by at least one of the following methods: based on statistical results of historical samples, based on validation set performance optimization results, based on clinical guidelines or medical knowledge bases, based on expert experience, and / or based on tolerance for misjudgments, missed judgments, or frequent switching in the target scenario. At least one of the following: object pointing degree, credibility, matching degree, adaptability, completion degree, absorption degree, and compatibility retention capability may preferably be normalized to a range of 0 to 1, where a larger value generally indicates a higher corresponding capability, degree, or stability, but this is not a limitation.

[0083] It should be noted that, unless otherwise specified, the weighted results, weighted processing, or weighted combinations involved in the above embodiments are all converted to the range of 0 to 1, and corresponding weights are set for each, with the sum of each weight being 1; the product of each evaluation result and its corresponding weight is calculated, and the sum of each product is used as the corresponding weighted result; for evaluation results whose increased value indicates an increased adverse impact, the converted value of the evaluation result is first subtracted from 1 before participating in the weighted calculation; the weights can be determined according to the degree of influence of each evaluation result on the result to be determined, or they can be determined by adjusting parameters according to the actual application scenario, without further limitations here.

[0084] In one embodiment, such as Figure 3 As shown, an auxiliary diagnostic device based on a combined auxiliary diagnostic model is provided, comprising: The data acquisition and model invocation module is used to obtain auxiliary diagnosis guidance information based on the target diagnosis and treatment data, and to invoke the pre-built joint auxiliary diagnosis model. The joint auxiliary diagnosis model includes at least multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches. Among them, there is a predetermined branch occlusion relationship between the multiple normal auxiliary diagnosis branches and / or the multiple specialized auxiliary diagnosis branches. The dominant branch confirmation module is used to determine the current dominant branch among multiple normal auxiliary diagnosis branches and / or specialized auxiliary diagnosis branches based on auxiliary diagnosis guidance information; The branch parsing and evaluation module is used to parse the target diagnosis and treatment data through the current dominant branch to generate the current dominant auxiliary diagnosis and treatment results, and to determine the degree of closure of the current dominant branch's interpretation of the target diagnosis and treatment data; The candidate takeover branch selection module is used to respond to the instability of the interpretation of the current dominant branch in terms of the interpretation of the closed-loop degree characterization. Then, based on the branch occlusion relationship, it obtains one or more occluded auxiliary diagnostic branches associated with the current dominant branch, and determines at least one candidate takeover branch from the one or more occluded auxiliary diagnostic branches. The takeover assessment and branch switching module is used to parse the target diagnosis and treatment data through candidate takeover branches to generate candidate auxiliary diagnosis results, and to determine whether the candidate takeover branch meets the branch takeover conditions for the current dominant branch. If it does, the candidate takeover branch is determined as the subsequent dominant branch to perform subsequent auxiliary diagnosis.

[0085] For specific limitations regarding the auxiliary diagnostic device, please refer to the limitations of the auxiliary diagnostic method above, which will not be repeated here. Each module in the aforementioned auxiliary diagnostic device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0086] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores auxiliary diagnostic data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements an auxiliary diagnostic method based on a joint auxiliary diagnostic model.

[0087] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0088] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described auxiliary diagnostic method.

[0089] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described auxiliary diagnostic method.

[0090] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0091] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0092] The above embodiments merely illustrate several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. An auxiliary diagnostic method based on a combined auxiliary diagnostic model, characterized in that, include: Based on the target diagnosis and treatment data, auxiliary diagnosis guidance information is obtained, and a pre-constructed joint auxiliary diagnosis model is invoked. The joint auxiliary diagnosis model includes at least multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches, wherein a branch occlusion relationship is predetermined between the multiple normal auxiliary diagnosis branches and / or the multiple specialized auxiliary diagnosis branches. Based on the auxiliary diagnosis guidance information, determine the current dominant branch among the multiple normal auxiliary diagnosis branches and / or the specialized auxiliary diagnosis branches; The target diagnosis and treatment data is analyzed by the current dominant branch to generate the current dominant auxiliary diagnosis result, and the degree of closure of the interpretation of the target diagnosis and treatment data by the current dominant branch is determined. In response to the interpretation closure degree characterizing the instability of the current dominant branch interpretation, one or more occluded auxiliary diagnostic branches associated with the current dominant branch are obtained based on the branch occlusion relationship, and at least one candidate takeover branch is determined from the one or more occluded auxiliary diagnostic branches. The candidate takeover branch is used to parse the target diagnosis and treatment data to generate candidate auxiliary diagnosis results, and it is determined whether the candidate takeover branch meets the branch takeover conditions of the current dominant branch. If it does, the candidate takeover branch is determined as the subsequent dominant branch to perform subsequent auxiliary diagnosis.

2. The auxiliary diagnostic method based on a combined auxiliary diagnostic model according to claim 1, characterized in that, The auxiliary diagnostic guidance information obtained based on the target diagnostic data includes: Acquire target treatment data, which includes at least one or more of the following: current round treatment data, historical round treatment data; Based on the current round of treatment data and / or the historical round of treatment data, construct a treatment feature sequence; Based on the diagnostic and treatment feature sequence, at least one guiding feature is determined to characterize the current diagnostic and treatment status and the relationship between the disease course evolution, and the auxiliary diagnostic guidance information is generated based on the guiding feature.

3. The auxiliary diagnostic method based on a combined auxiliary diagnostic model according to claim 1, characterized in that, Before invoking the pre-built joint diagnostic model, the method further includes: Obtain multiple auxiliary diagnostic branches and determine the branch interpretation objects of the multiple auxiliary diagnostic branches; If the branch explanation object is a common disease and related symptoms, the corresponding auxiliary diagnosis branch is determined as a normal auxiliary diagnosis branch; if the branch explanation object is a rare disease and related symptoms, the corresponding auxiliary diagnosis branch is determined as a specialized auxiliary diagnosis branch. For multiple routine auxiliary diagnosis branches, based on the disease course evolution relationship between the multiple branch explanation objects, one or more specialized auxiliary diagnosis branches that have a disease course evolution association with the branch explanation objects of the routine auxiliary diagnosis branches are selected to construct multiple hybrid auxiliary diagnosis branch clusters, and the routine auxiliary diagnosis branches and the selected specialized auxiliary diagnosis branches are determined as auxiliary diagnosis branches within the cluster. For multiple hybrid auxiliary diagnosis branch clusters, extract the diagnosis and treatment data related to the branch interpretation object of each hybrid auxiliary diagnosis branch cluster to generate multiple branch cluster hybrid input sets; The multiple branch cluster mixed input sets are respectively input into the corresponding multiple mixed auxiliary diagnosis branch clusters to obtain multiple branch cluster output results; The branch cluster output results are analyzed to determine the branch interpretation coverage and output result bias among the multiple intra-cluster auxiliary diagnostic branches of the corresponding hybrid auxiliary diagnostic branch cluster. Based on the branch interpretation coverage and the output result bias, the intra-cluster branch occlusion relationship of the hybrid auxiliary diagnostic branch cluster is determined to characterize at least one intra-cluster auxiliary diagnostic branch's occlusion / masking relationship relative to the remaining multiple intra-cluster auxiliary diagnostic branches and the corresponding occlusion / masking level. In response to the existence of the disease progression association between the branch interpretation objects corresponding to the normal auxiliary diagnosis branches within the multiple hybrid auxiliary diagnosis branch clusters, one or more hybrid auxiliary diagnosis branch cluster groups are constructed based on the multiple hybrid auxiliary diagnosis branch clusters, and the intra-group branch cluster occlusion relationship between the multiple hybrid auxiliary diagnosis branch clusters is determined. Based on the intra-cluster branch occlusion relationship and the intra-group branch cluster occlusion relationship, determine the branch occlusion relationship between multiple normal auxiliary diagnostic branches and / or multiple specialized auxiliary diagnostic branches; Based on the branch occlusion relationship, multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches are organized to generate a joint auxiliary diagnosis model.

4. The auxiliary diagnostic method based on a combined auxiliary diagnostic model according to claim 3, characterized in that, The acquisition of multiple auxiliary diagnostic branches includes: Based on the target interpretation object, obtain multiple candidate auxiliary diagnosis branches, and determine the branch interpretation capability of the multiple candidate auxiliary diagnosis branches; The selection priority of the multiple candidate auxiliary diagnostic branches is determined based on the overlap between the branch interpretation objects of the multiple candidate auxiliary diagnostic branches and the branch interpretation capability. Based on the selection priority, multiple candidate auxiliary diagnostic branches are selected sequentially as target auxiliary diagnostic branches, so that the coverage of the branch explanation objects of at least multiple target auxiliary diagnostic branches meets a preset threshold.

5. The auxiliary diagnostic method based on a combined auxiliary diagnostic model according to claim 1, characterized in that, The step of determining the current dominant branch among multiple routine auxiliary diagnosis branches and / or specialized auxiliary diagnosis branches based on the auxiliary diagnosis guidance information includes: The auxiliary diagnostic guidance information is analyzed to obtain guidance features, and the historical most recent branch determination results are obtained; Determine whether there is a direct guidance feature among the guidance features, wherein the direct guidance feature has a unique correspondence with an auxiliary diagnosis branch in the joint auxiliary diagnosis model; In response to the existence of the direct guidance feature, the auxiliary diagnostic branch corresponding to the direct guidance feature is determined as the current dominant branch; In response to the absence of the direct guidance feature and the determination result of the most recent historical branch being non-empty, the auxiliary diagnostic branch corresponding to the determination result of the most recent historical branch is determined as the current priority branch, and when the matching degree between the guidance feature and the object typical feature of the pre-stored branch explanation object corresponding to the current priority branch meets the continuation threshold, the current priority branch is determined as the current dominant branch. In response to the absence of the direct guiding feature and the empty result of the historical most recent branch determination, the current dominant branch is determined from the multiple normal auxiliary diagnosis branches based on the matching degree between the guiding feature and the typical features of the object of the branch explanation object corresponding to the multiple normal auxiliary diagnosis branches.

6. The auxiliary diagnostic method based on a combined auxiliary diagnostic model according to claim 1, characterized in that, The step of generating a current dominant auxiliary diagnosis result by parsing the target diagnosis and treatment data through the current dominant branch, and determining the closure degree of the interpretation of the target diagnosis and treatment data by the current dominant branch, includes: The target diagnosis and treatment data is input into the current dominant branch, so that the current dominant branch extracts diagnosis and treatment information factors from the target diagnosis and treatment data, and performs inference based on multiple diagnosis and treatment information factors to generate an object pointing result of the target diagnosis and treatment data relative to the branch interpretation object corresponding to the current dominant branch; Based on the object pointing result, the current dominant auxiliary diagnosis result is generated. The current dominant auxiliary diagnosis result at least characterizes the degree and credibility of the object pointing of the target diagnosis data to the branch explanation object and / or its sub-objects. Based on the inference contribution of multiple diagnostic and treatment information factors to the object-directed result, determine the factor explanatory power of the current dominant branch to the multiple diagnostic and treatment information factors under the object-directed result; Based on multiple diagnostic and treatment information factors, the corresponding factor explanatory power, and the object pointing result, a diagnostic and treatment information interpretation chain is constructed. The diagnostic and treatment information factors that have been included in the diagnostic and treatment information interpretation chain are identified as on-chain factors, and the diagnostic and treatment information factors that have not been included in the diagnostic and treatment information interpretation chain are identified as off-chain factors. The explanatory closure degree is determined based on the factor explanatory power of the on-chain factors, the proportion of the off-chain factors, and the characterization criticality of the off-chain factors.

7. The auxiliary diagnostic method based on a combined auxiliary diagnostic model according to claim 6, characterized in that, In response to the interpretation closure degree characterizing the instability of the current dominant branch, one or more occluded auxiliary diagnostic branches associated with the current dominant branch are obtained based on the branch occlusion relationship, and at least one candidate takeover branch is determined from the one or more occluded auxiliary diagnostic branches, including: Based on the branch occlusion relationship, determine one or more occluded auxiliary diagnostic branches associated with the current dominant branch, and determine the occlusion level of the current dominant branch on one or more of the occluded auxiliary diagnostic branches; The depth level of the current dominant branch and one or more corresponding occluded auxiliary diagnostic branches is obtained by mapping the occlusion level. The depth level at least indicates that there is an occlusion relationship between the auxiliary diagnostic branch in the shallow layer and the auxiliary diagnostic branch in the deep layer, and the occlusion level of the auxiliary diagnostic branch in the shallow layer to the auxiliary diagnostic branch in the deep layer increases with the branch depth difference. The next depth level after the current dominant branch is taken as the target depth level for descent, and one or more of the occluded auxiliary branches in the target depth level for descent are identified as candidate branches for descent. Obtain the diagnostic information interpretation chain, the on-chain factors, and the off-chain factors; Based on the branch interpretation objects corresponding to one or more of the potential candidate branches, determine the positive explanatory power of one or more of the potential candidate branches for the off-chain factors, as well as the compatibility and maintenance ability of the diagnostic information interpretation chain and the factors on the chain. Based on the positive explanatory power of the off-chain factors and the compatibility and maintenance of the diagnostic information interpretation chain and the on-chain factors, the takeover adaptation degree of one or more of the potential candidate branches to the current dominant branch is determined respectively. Based on the takeover suitability, at least one candidate takeover branch is determined from one or more of the dive candidate branches.

8. The auxiliary diagnostic method based on a combined auxiliary diagnostic model according to claim 7, characterized in that, The step of parsing the target diagnosis and treatment data through the candidate takeover branch to generate candidate auxiliary diagnosis results, and determining whether the candidate takeover branch meets the branch takeover conditions for the current dominant branch, and if so, determining the candidate takeover branch as the subsequent dominant branch to perform subsequent auxiliary diagnosis, includes: The current dominant branch continues to perform the current round of auxiliary diagnosis, and the target diagnosis and treatment data is input into at least one of the candidate takeover branches, so that the candidate takeover branches perform factor extraction and inference based on the target diagnosis and treatment data, generate candidate object pointing results, and generate corresponding candidate auxiliary diagnosis results and candidate takeover diagnosis and treatment information interpretation chains; Based on the degree of inclusion of the off-chain factor corresponding to the current dominant branch in one or more of the candidate takeover diagnosis and treatment information interpretation chains, the degree of absorption of the off-chain factor by one or more of the candidate takeover branches is determined respectively; Pre-inference is performed based on one or more of the candidate takeover diagnosis and treatment information interpretation chains to obtain the predicted diagnosis and treatment features corresponding to one or more of the candidate takeover branches respectively; In response to receiving subsequent rounds of treatment data, new treatment features are extracted from the subsequent rounds of treatment data, and the feature matching degree between the new treatment features and the predicted treatment features corresponding to one or more candidate takeover branches is determined respectively; Based on the subsequent rounds of diagnosis and treatment data and the current dominant branch, the updated explanatory closure degree of the current dominant branch in the subsequent rounds is determined, and the trend of the explanatory closure degree of the current dominant branch is determined according to the updated explanatory closure degree and the explanatory closure degree of the previous rounds. Based on the feature matching degree corresponding to one or more of the candidate takeover branches and the trend of the change in the explanatory closure degree of the current dominant branch, a set of candidate takeover branches that meet the branch takeover conditions is determined, wherein the branch takeover conditions include at least: the corresponding feature matching degree meets a preset confirmation threshold, and the trend of the change in the explanatory closure degree indicates that the explanatory closure degree of the current dominant branch continues to decrease. If the candidate takeover branch set contains only one candidate takeover branch, then the candidate takeover branch is determined as the subsequent dominant branch to replace the current dominant branch and perform subsequent auxiliary diagnostics. In response to the fact that the candidate takeover branch set includes multiple candidate takeover branches, the subsequent dominant branch is determined based on the feature matching degree corresponding to the multiple candidate takeover branches and / or the degree of absorption of the dechaining factor; In response to the set of candidate takeover branches being empty, the current dominant branch is maintained, and the takeover status of at least one of the candidate takeover branches is retained as pending confirmation.

9. An auxiliary diagnostic device based on a combined auxiliary diagnostic model, characterized in that, The device includes: The data acquisition and model invocation module is used to obtain auxiliary diagnosis guidance information based on the target diagnosis and treatment data, and to invoke a pre-built joint auxiliary diagnosis model. The joint auxiliary diagnosis model includes at least multiple normal auxiliary diagnosis branches and multiple specialized auxiliary diagnosis branches, wherein a branch occlusion relationship is predetermined between the multiple normal auxiliary diagnosis branches and / or the multiple specialized auxiliary diagnosis branches. The dominant branch confirmation module is used to determine the current dominant branch among multiple normal auxiliary diagnosis branches and / or specialized auxiliary diagnosis branches based on the auxiliary diagnosis guidance information. The branch parsing and evaluation module is used to parse the target diagnosis and treatment data through the current dominant branch to generate the current dominant auxiliary diagnosis result, and to determine the degree of closure of the interpretation of the target diagnosis and treatment data by the current dominant branch; The candidate takeover branch selection module is used to, in response to the interpretation closure degree characterizing the instability of the current dominant branch interpretation, obtain one or more occluded auxiliary diagnostic branches associated with the current dominant branch based on the branch occlusion relationship, and determine at least one candidate takeover branch from the one or more occluded auxiliary diagnostic branches. The takeover assessment and branch switching module is used to parse the target diagnosis and treatment data through the candidate takeover branch to generate candidate auxiliary diagnosis results, and to determine whether the candidate takeover branch meets the branch takeover conditions for the current dominant branch. If it meets the conditions, the candidate takeover branch is determined as the subsequent dominant branch to perform subsequent auxiliary diagnosis.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images