An intelligent pre-consultation analysis method and system for patient service

Abstract

This invention discloses an intelligent pre-diagnosis analysis method and system for patient services. The method includes: acquiring the patient's initial symptom information and determining the initial network density; comparing the initial network density with a preset seepage inflection point threshold; if the initial network density is less than the seepage inflection point threshold, determining that the symptom co-occurrence network is in a single disease state and outputting a single-cause diagnosis result; if the initial network density is not less than the seepage inflection point threshold, executing a multi-cause judgment process: calculating the diagnostic convergence of the symptom co-occurrence network; responding to a judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a multi-disease coexistence state, outputting a multi-cause diagnosis result, and completing the pre-diagnosis analysis. This invention provides an intelligent pre-diagnosis analysis method and system for patient services to solve the technical problems of existing pre-diagnosis systems lacking the ability to identify symptom network phase transitions and making it difficult to avoid missed diagnoses of comorbidities.

Description

Technical Field

[0001] This invention relates to the field of online medical data analysis, and in particular to an intelligent pre-diagnosis analysis method and system for patient services. Background Technology

[0002] Intelligent pre-diagnosis systems play a crucial role in improving the efficiency of medical services by collecting symptom information and inferring the causes of diseases through multiple rounds of interaction with patients. One of the core technologies for achieving accurate diagnosis is the identification of critical phase transitions in symptom co-occurrence networks. Its goal is to monitor the density evolution of the correlation network between symptoms in real time across the multidimensional symptoms reported by patients, accurately identify the critical moment when a symptom cluster transitions from a single attributable cause to a state of multiple co-occurrence, and switch diagnostic strategies in a timely manner to avoid missing comorbidities.

[0003] Currently, mainstream intelligent pre-diagnosis methods mainly include symptom accumulation-based inquiry-based diagnosis and knowledge graph-based disease reasoning methods. Symptom accumulation-based inquiry-based diagnosis primarily employs a single-cause convergence strategy. However, this method relies on the assumption that more accumulated symptoms lead to a clearer diagnosis, neglecting the co-occurrence relationships between symptoms. When the network density exceeds a critical point, the symptom cluster transitions into a state of multiple co-occurrence of diseases. This method lacks density monitoring capabilities and cannot identify the critical transition point, leading to continued inquiry based on a single-cause logic even when multiple diseases coexist, resulting in misdiagnosis or missed diagnosis. Knowledge graph-based disease reasoning methods rely on static graphs and lack the ability to model dynamic co-occurrence relationships of symptoms. Furthermore, the critical density threshold varies significantly among different populations, and a uniform standard leads to frequent misdiagnosis. Therefore, existing technologies lack real-time monitoring capabilities for the evolution of symptom co-occurrence network density and the ability to identify critical phase transitions. They cannot adaptively switch diagnostic strategies at critical moments, resulting in high rates of missed and misdiagnosed co-occurrences and insufficient accuracy in pre-diagnosis. Summary of the Invention

[0004] This invention provides an intelligent pre-diagnosis analysis method and system for patient services, which solves the technical problems of existing technologies that lack real-time monitoring capabilities for the evolution of symptom co-occurrence network density and critical phase transition identification capabilities, cannot adaptively switch diagnostic strategies at critical moments, and have high rates of missed diagnoses and misjudgments of comorbidities.

[0005] The present invention discloses the following technical solutions: In a first aspect, embodiments of the present invention provide an intelligent pre-diagnosis analysis method for patient services, the method comprising: Obtain the patient's initial symptom information, construct a symptom co-occurrence network based on the initial symptom information, and determine the initial network density of the symptom co-occurrence network; The initial network density is compared with a preset seepage inflection threshold. If the initial network density is less than the seepage inflection threshold, the symptom co-occurrence network is determined to be in a single disease state, and a single etiology diagnosis result is output. If the initial network density is not less than the seepage inflection threshold, a multi-etiology judgment process is executed. The multi-causal diagnosis process includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, responding to a judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-causal diagnosis results, and completing the pre-diagnosis analysis.

[0006] Optionally, the process of presetting the seepage inflection point threshold includes: Density evolution sequences were obtained from the symptom co-occurrence network; Calculate the percentage of nodes corresponding to different network densities in the density evolution sequence; Based on the rate of change of the node proportion, the network density corresponding to the maximum value of the rate of change is determined as the seepage inflection point threshold.

[0007] Optionally, the step of performing topology analysis on the symptom co-occurrence network and calculating the diagnostic convergence of the symptom co-occurrence network includes: When the initial network density is not less than the seepage inflection point threshold, it is determined that the symptom co-occurrence network has undergone a symptom phase transition. The topology of the symptom co-occurrence network is analyzed to obtain the symptom node features. Based on the symptom node features, the diagnostic convergence of the symptom is calculated.

[0008] Optionally, the step of performing topology analysis on the symptom co-occurrence network to obtain symptom node features includes: The topology of the symptom co-occurrence network is analyzed to determine multiple cluster centers in the symptom co-occurrence network; The state evolution features of the symptom co-occurrence network are extracted based on the cluster centers; Based on the state evolution characteristics, the symptom co-occurrence network is divided to obtain multiple connected components containing symptom node features.

[0009] Optionally, calculating the diagnostic convergence of the symptom based on the symptom node features includes: Obtain pathological evolution cycle data and clinical indication data from the medical record database; Extract node feature vectors from the symptom node features, and calculate the complication correlation degree based on the node feature vectors; In response to a judgment command that the correlation degree of the complication is greater than a preset correlation degree threshold, the node feature vector is weighted and fused according to the clinical indication data to obtain a feature aggregation tensor; Based on the feature aggregation tensor and the pathological evolution cycle data, the diagnostic convergence value corresponding to the symptoms is calculated.

[0010] Optionally, if the initial network density is less than the seepage inflection point threshold, the symptom co-occurrence network is determined to be in a single disease state, and a single etiology diagnosis result is output, including: When the initial network density is less than the seepage inflection point threshold, it is determined that the symptom co-occurrence network does not undergo symptom phase transition, and the judgment result that the patient is in a single disease state is obtained. Based on the judgment result, a single follow-up question is performed on the symptom co-occurrence network to obtain a single etiological diagnosis result.

[0011] Optionally, the multi-etiology judgment process includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, and, in response to a judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-etiology diagnostic results, and completing pre-diagnosis analysis, including: Patients whose diagnostic convergence is less than a preset convergence threshold are determined to be in a state of multiple coexisting diseases; Based on the coexistence of multiple diseases, a multi-cause inquiry instruction is generated, and the feedback result corresponding to the cause inquiry instruction is obtained. Based on the feedback results, the symptoms are analyzed and diagnosed to obtain multi-causal diagnostic results, thus completing the pre-diagnosis analysis.

[0012] Secondly, embodiments of the present invention provide an intelligent pre-diagnosis analysis system for patient services, comprising: An acquisition module is used to acquire the patient's initial symptom information, construct a symptom co-occurrence network based on the initial symptom information, and determine the initial network density of the symptom co-occurrence network. The judgment module is used to compare the initial network density with a preset seepage inflection point threshold. If the initial network density is less than the seepage inflection point threshold, the symptom co-occurrence network is determined to be in a single disease state, and a single etiology diagnosis result is output. If the initial network density is not less than the seepage inflection point threshold, a multi-etiology judgment process is executed. The analysis module, used in the multi-causal diagnosis process, includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, responding to a judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-causal diagnosis results, and completing the pre-diagnosis analysis.

[0013] Thirdly, another embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the intelligent pre-diagnosis analysis method for patient services as described above.

[0014] Another embodiment of the present invention provides a computer program product, including a computer program or instructions, which, when executed by a device, implement the steps of the aforementioned intelligent pre-diagnosis analysis method for patient services.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention, through acquiring initial patient symptoms and constructing a symptom co-occurrence network, transforms discrete symptom information into structured network data with topological relationships, providing a quantifiable analytical foundation for subsequent critical phase transition identification. By comparing the initial network density with a preset seepage inflection point threshold, the critical moment when a symptom cluster transitions from a single-cause attributable state to a multi-disease coexistence state is accurately identified, enabling adaptive switching of diagnostic strategies and avoiding misdiagnosis or missed diagnosis under complex conditions based on a single-cause logic. Furthermore, after determining that a symptom phase transition has occurred, by performing topological structure analysis on the network and calculating diagnostic convergence, the invention effectively distinguishes between complex single-disease states that are still convergent and true multi-disease coexistence states, providing reliable technical support for accurate follow-up questioning and final diagnosis. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, the 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 from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating an intelligent pre-diagnosis analysis method for patient services provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the specific implementation process of an intelligent pre-diagnosis analysis method for patient services provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of an intelligent pre-diagnosis analysis device for patient services provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the invention, are intended to cover non-exclusive inclusion.

[0020] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0021] See Figure 1 To address the technical problems of existing technologies lacking real-time monitoring capabilities for the evolution of symptom co-occurrence network density and the ability to identify critical phase transitions, thus failing to adaptively switch diagnostic strategies at critical moments and resulting in high rates of missed diagnoses and misdiagnoses of comorbidities, an embodiment of this invention provides an intelligent pre-diagnosis analysis method for patient services, comprising: S1, Obtain the patient's initial symptom information, construct a symptom co-occurrence network based on the initial symptom information, and determine the initial network density of the symptom co-occurrence network.

[0022] The process begins by acquiring the patient's initial symptom information, using symptom entities extracted from the initial symptom text as network nodes. Then, co-occurrence statistics from historical medical records or medical literature are retrieved to analyze the frequency of co-occurrence of any two symptoms in the same medical context. Based on this, weighted undirected connections are established between corresponding nodes, thus constructing a symptom co-occurrence network that reflects the degree of correlation between symptoms. On this basis, the system statistically analyzes the actual number of connections E and the total number of nodes N in the network, and calculates the ratio of these two values ​​based on the maximum number of edges N×(N−1) / 2 in undirected network theory, obtaining the initial network density ρ=2E / [N×(N−1)]. This density value serves as a core indicator for quantifying the overall clustering and complexity of the current symptom set, providing an objective quantitative basis for subsequently determining whether the symptom cluster has crossed the critical point of seepage phase transition.

[0023] S2, compare the initial network density with the preset seepage inflection point threshold. If the initial network density is less than the seepage inflection point threshold, determine that the symptom co-occurrence network is in a single disease state and output a single etiology diagnosis result; if the initial network density is not less than the seepage inflection point threshold, execute the multi-etiology judgment process.

[0024] The initial network density is calculated and compared with a pre-established inflection point threshold based on historical symptom network evolution data to determine whether the topology of the current symptom cluster has undergone a qualitative change. If the initial network density is greater than the inflection point threshold, it indicates that the edges connecting symptom nodes have crossed a critical density, and a globally connected component has been formed within the network. This indicates a symptom phase transition has occurred in the symptom co-occurrence network. At this point, the system triggers a deep analysis process for complex conditions, using a graph neural network to decompose the network topology to extract multiple connected components. The diagnostic convergence is calculated based on the multidimensional features of symptom nodes within each component to assess the concentration of the symptom set pointing to a specific disease. Conversely, if the initial network density is not greater than the inflection point threshold, it indicates that the symptom associations are still sparse, and no complex cross-system concurrent patterns have formed between the symptoms. The etiology is relatively clear, and the system determines that the patient is in a single disease state and directly performs single follow-up processing to output a single etiology diagnosis result.

[0025] S3, the multi-causal judgment process includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, responding to the judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-causal diagnosis results, and completing the pre-diagnosis analysis.

[0026] The process involves obtaining the diagnostic convergence calculated in the previous step and comparing it with a preset convergence threshold. If the diagnostic convergence is lower than the preset threshold, it indicates that the disease probability distribution corresponding to the current symptom cluster is not concentrated on a few candidate diseases, but rather dispersed towards multiple different disease directions. Based on this, the system determines that the patient is in a state of multiple coexisting diseases and then triggers the multi-causal analysis engine. Starting from the initial symptoms, the engine traverses nodes and extracts associated paths in the medical knowledge graph, generates targeted multi-causal follow-up instructions, and obtains patient feedback. The feedback information is then structured and input into the comorbidity inference model for joint probability calculation. Finally, the system outputs a comorbidity diagnosis result containing multiple coexisting diseases and their confidence levels, thereby completing the intelligent pre-diagnosis analysis process for complex conditions.

[0027] Furthermore, such as Figure 2 As shown, this application provides a specific implementation process for an intelligent pre-diagnosis analysis method for patient services. First, the patient's initial symptom text is obtained. Entity extraction is performed on the initial symptom text using a random forest algorithm to obtain an initial symptom entity set. A symptom co-occurrence network is constructed based on the initial symptom entity set. The number of connection edges and the total number of nodes in the symptom co-occurrence network are obtained to determine the initial network density. Then, for the initial network density, a pre-established seepage inflection point threshold is obtained, and it is determined whether the initial network density is greater than the seepage inflection point threshold. If the initial network density is not greater than the seepage inflection point threshold, a single follow-up question is performed on the initial symptom entity set to obtain a single etiology diagnosis result. If the initial network density is greater than the seepage inflection point threshold, it is determined that the symptom co-occurrence network has undergone a symptom phase transition. A graph neural network algorithm is used to perform topological structure analysis on the symptom co-occurrence network that has undergone the symptom phase transition, obtaining multiple connected components. Then, the symptom node features contained in the multiple connected components are obtained, and the diagnostic convergence corresponding to the multiple connected components is calculated based on the symptom node features. Next, for the diagnostic convergence, a preset convergence threshold is obtained, and it is determined whether the diagnostic convergence is lower than the preset convergence threshold. If the diagnostic convergence is lower than the preset convergence threshold, it is determined that the patient is in a state of multiple coexisting diseases. Finally, a multi-cause inquiry instruction is generated for the state of multiple coexisting diseases to obtain the comorbidity diagnosis result.

[0028] In one embodiment, the process of setting the seepage inflection point threshold includes steps S201 to S203, each step of which is as follows: S201, Obtain the density evolution sequence from the symptom co-occurrence network.

[0029] Based on the constructed symptom co-occurrence network, the network was dynamically pruned by progressively removing low-weight edges or according to a co-occurrence strength threshold from high to low. The network density value and the proportion of nodes in the largest connected component were recorded at each pruning stage, generating a density evolution sequence reflecting the evolution of the network topology as the association strength changes. This sequence depicts the complete trajectory of the symptom network from a tightly connected state to fragmentation and disintegration, providing a data foundation for identifying abrupt changes in network density and the proportion of the largest connected component, and for establishing seepage inflection point thresholds offline.

[0030] S202, calculate the percentage of nodes corresponding to different network densities in the density evolution sequence.

[0031] Specifically, for each pruning stage in the generated density evolution sequence, a depth-first search or disjoint-set data structure algorithm is used to traverse all connected components of the symptom co-occurrence network at the current stage. The system identifies the largest connected component containing the most nodes and calculates the number of nodes in this largest connected component, dividing the total number of nodes in the current stage network by the number of nodes in this largest connected component. By traversing the entire density evolution sequence, the system establishes a synchronous mapping relationship between the node proportion of the largest connected component and the network density as it decreases. This mapping relationship clearly reflects the trend of the size change of the largest connected component in the symptom network as the edge density gradually decreases, providing crucial data support for subsequently locating the seepage inflection point based on the rate of change of the node proportion.

[0032] S203, based on the numerical change rate of the node proportion, the network density corresponding to the maximum value of the numerical change rate is determined as the seepage inflection point threshold.

[0033] Specifically, based on the proportion of the largest connected component nodes corresponding to each network density in the density evolution sequence, the changing trend of the node proportion with decreasing network density is numerically differentiated to calculate the rate of change of the node proportion within adjacent density intervals. Since the node proportion decreases sharply when the network transitions from a connected state to a fragmented state, its rate of change peaks near a critical point. The system identifies the network density corresponding to the maximum rate of change as the seepage inflection point threshold. This threshold characterizes the critical density value at which the symptom co-occurrence network abruptly changes from a macroscopically connected state to a multi-isolated cluster state, providing a pre-established quantitative benchmark for determining whether a phase transition has occurred in the patient's symptom cluster during subsequent online diagnosis.

[0034] This embodiment achieves adaptive, data-driven offline calibration of the seepage inflection point threshold by obtaining the density evolution sequence from the symptom co-occurrence network, calculating the node proportions corresponding to different network densities, and determining the network density at which the rate of change of the node proportion reaches its maximum value as the seepage inflection point threshold. This preset method fully considers the topological differences of symptom association networks under different disease spectra and population characteristics, enabling the established seepage inflection point threshold to accurately reflect the critical density value at which symptom clusters change from a single attributable cause state to a multi-disease coexistence state in the current application scenario. This avoids the technical defects of frequent misjudgments of elderly patients or complex cases caused by using a uniform fixed threshold, thereby significantly improving the accuracy and individual adaptability of phase transition determination in subsequent online diagnosis, and providing quantitative critical benchmark support for reliable decision-making of intelligent pre-diagnosis systems in multi-disease coexistence scenarios.

[0035] In one embodiment, the topology of the symptom co-occurrence network is analyzed, and the diagnostic convergence of the symptom co-occurrence network is calculated, including steps S301 to S302, each of which is as follows: S301, when the initial network density is not less than the seepage inflection point threshold, it is determined that the symptom co-occurrence network has undergone a symptom phase transition, and the topology structure of the symptom co-occurrence network is analyzed to obtain the symptom node features.

[0036] When the system determines that the initial network density exceeds the seepage inflection point threshold, indicating that the symptom co-occurrence network has undergone a symptom phase transition, the system initiates a graph neural network algorithm to analyze the topology of the entire network. Through graph convolution operations, it aggregates the clinical attribute features of each node and the co-occurrence association features of its neighboring nodes, and combines topological statistics such as node centrality and betweenness to generate a comprehensive symptom node feature vector. Subsequently, based on these feature vectors, the system performs graph segmentation or spectral clustering operations on the network, separating groups of nodes with tightly connected internal structures and sparse external connections, thus obtaining multiple connected components. Each connected component contains a set of structured subgraphs with clear clinical co-occurrence patterns and corresponding symptom node features, providing a modular data foundation for subsequent diagnostic convergence calculations.

[0037] S302, Calculate the diagnostic convergence of the symptom based on the symptom node features.

[0038] The system uses the symptom node features extracted from each connected component as input. Initial weights are assigned to each symptom based on its clinical specificity; symptoms with higher specificity, such as typical rashes or specific neurological reflex abnormalities, are assigned higher weights, while non-specific symptoms, such as mild fever, are assigned lower weights. Subsequently, the system performs similarity matching between the weighted symptom features and candidate disease models in a pre-defined diagnostic hypothesis space. It then uses Bayesian networks or probabilistic graphical models to calculate the posterior probability of each candidate disease and form a probability distribution vector. Finally, an inverse proportional mapping is performed based on the information entropy of this probability distribution vector. Lower entropy values ​​indicate a greater concentration of the probability distribution towards a few diseases, resulting in higher diagnostic convergence, and vice versa. This quantitatively characterizes the degree to which the symptom set within the current connected component points to a specific disease.

[0039] This embodiment first performs topological analysis on the symptom co-occurrence network after a phase transition occurs to obtain symptom node features, and then calculates the diagnostic convergence based on these features. This achieves an effective connection and quantitative evaluation between network topology and clinical diagnostic decisions. Because the node features obtained from graph neural network analysis integrate the clinical attributes of the symptoms themselves with their structural importance within the network, the diagnostic convergence calculated based on these features can objectively reflect the degree to which the symptom set within the current connected component points to a specific disease conclusion, rather than simply relying on the number of symptoms or co-occurrence frequency for coarse-grained judgment. This effectively avoids the misdiagnosis or missed diagnosis problem caused by traditional schemes that forcibly attribute symptom groups to a single etiology after they cross the phase transition threshold. This provides a reliable quantitative basis for the system to accurately identify the coexistence of multiple diseases in complex symptom networks.

[0040] In one embodiment, topology analysis is performed on the symptom co-occurrence network to obtain symptom node features, including steps S401 to S403, each step of which is as follows: S401, Perform topology analysis on the symptom co-occurrence network to determine multiple cluster centers in the symptom co-occurrence network.

[0041] In this process, after determining that a phase transition has occurred in the symptom co-occurrence network, a graph topology-based clustering algorithm is used to cluster the symptom nodes in the network based on the comprehensive feature vectors extracted from each node using a graph neural network algorithm. By calculating the density distribution of nodes in the feature space and the structural similarity between nodes, local density maxima or centroid nodes with the smallest average distance from other nodes are identified in the feature space and determined as multiple cluster centers of the symptom co-occurrence network. These cluster centers are essentially the most representative and influential core symptom nodes in each symptom subgroup, effectively characterizing the typical features of different complication modules or pathological subtypes, providing clear anchor points for subsequent partitioning of connected components based on cluster centers and tracing the propagation path of core symptoms.

[0042] S402, extract the state evolution features of the symptom co-occurrence network based on the cluster centers.

[0043] Using multiple identified cluster centers as anchor points, the evolution of connection patterns of each cluster center as the network density changes from low to high or from high to low is tracked. By analyzing the increasing or decreasing trend of edge weights between cluster centers and their surrounding neighbors with varying thresholds, the expansion or contraction rate of the connected components to which the cluster centers belong, and the waxing and waning relationships of cross-cluster connection strengths between different cluster centers, state evolution features reflecting the dynamic changes in the topology of the entire symptom co-occurrence network are extracted. These features, in the form of time-sensitive multidimensional vectors, record key pattern parameters in the process of the network transitioning from a sparse, isolated cluster state to a dense, giant connected component state. This provides dynamic data support for subsequent use of state evolution features to partition the network into connected components and to quantitatively predict the concurrent trends of different pathological modules after symptom phase transitions.

[0044] S403, based on the state evolution characteristics, the symptom co-occurrence network is divided to obtain multiple connected components containing symptom node features.

[0045] Based on the extracted state evolution features—namely, the evolution trajectory of the connection patterns of each cluster center during the dynamic changes in network density and the expansion rate of its connected components—spectral clustering or graph cut algorithms are used to partition the symptom co-occurrence network. Cluster centers with similar evolution trajectories and tightly connected internal nodes, along with their associated nodes, are grouped into the same subgraph. Simultaneously, edges sparsely connected to external nodes are cut, resulting in multiple connected components with relatively independent internal topologies and low external coupling. Then, during the partitioning process, the system simultaneously retains and updates a comprehensive feature vector—a fusion of clinical attributes and topological statistics—for each symptom node within each connected component. This ensures that each connected component contains a structured set of nodes with complete symptom node features, providing clear functional modular input for accurate calculation of diagnostic convergence and comorbidity diagnosis inference.

[0046] This embodiment first identifies multiple cluster centers in the symptom co-occurrence network, then extracts state evolution features based on these cluster centers, and finally partitions the network based on these features to obtain multiple connected components containing symptom node features. This achieves refined and dynamic modular decomposition of complex symptom networks after phase transition. The scheme uses cluster centers as the core anchor points for each symptom subgroup and leverages state evolution features to characterize the evolution trajectory of connection patterns during density changes in different cluster centers. Therefore, when partitioning connected components, it not only relies on the static topology but also incorporates the dynamic evolution law of the network. This makes the partitioned connected components more closely resemble the independent evolution paths of different complication modules or pathological subtypes in clinical practice, effectively avoiding symptom attribution errors caused by traditional static fragmentation methods. It provides a structurally clear and clinically semantically explicit subgraph input for subsequent diagnostic convergence calculations and comorbidity inference for each connected component.

[0047] In one embodiment, the diagnostic convergence of the symptom is calculated based on the symptom node features, including steps S501 to S504, each step of which is as follows: S501 retrieves pathological evolution cycle data and clinical indication data from the medical record database.

[0048] The system uses core symptom nodes and their associated paths within connected components as search indexes to initiate structured query requests to a pre-established and continuously updated medical record database. This database connects to the medical record repository in the hospital information system and accesses public academic resource platforms such as PubMed and CNKI via open search APIs. It extracts corresponding clinical indication data and pathological evolution cycle data by matching symptom combinations with disease identifier codes. The clinical indication data includes objective quantitative information such as laboratory test indicators, physical examination results, and medical imaging features. The pathological evolution cycle data records the time span and symptom evolution patterns of the disease from the incubation period, prodromal period, onset period to the final stage. Both provide standardized clinical reference for subsequent calculation of complication correlation and weighted fusion of feature aggregation tensors.

[0049] S502, extract node feature vectors from the symptom node features, and calculate the complication correlation degree based on the node feature vectors.

[0050] Specifically, numerical node feature vectors are extracted from the symptom node features contained in each connected component. These vectors integrate clinical symptom attributes such as anatomical location, duration, and severity, as well as topological statistics such as node centrality, betweenness, and clustering coefficient. Subsequently, the system uses the extracted node feature vectors as input and, based on the statistical association frequency between diseases and symptoms in a pre-defined medical knowledge graph, calculates the co-occurrence probability and causal correlation between each symptom node in each connected component and multiple known diseases. Then, through weighted summation or cosine similarity measurement, the system comprehensively obtains the complication association degree value corresponding to each connected component. This numerical value quantitatively reflects the tightness of coupling between the current symptom cluster and specific complication pathways, providing a structured quantitative input for subsequent weighted fusion processing combined with clinical indication data.

[0051] S503, in response to the judgment instruction that the correlation degree of the complication is greater than the preset correlation degree threshold, the node feature vector is weighted and fused according to the clinical indication data to obtain the feature aggregation tensor.

[0052] The calculated complication correlation degree is compared with a preset correlation degree threshold. If the complication correlation degree is greater than the preset threshold, it indicates a significant coupling relationship between the symptom cluster and the specific complication path within the current connected component, at which point the system triggers the feature fusion process. Specifically, the system uses clinical indicator data obtained from the medical record database as dynamic weight coefficients to perform weighted fusion processing on the attributes of each dimension in the node feature vector. That is, different fusion weights are assigned according to the strength of the statistical correlation between symptoms and clinical indicators. Through multi-dimensional feature splicing and linear transformation, the weighted node feature vectors are integrated into a unified high-order feature aggregation tensor. This tensor effectively incorporates correction factors of clinical objective test indicators while retaining the original symptom topological information, providing an information-enhanced structured input for subsequent calculation of diagnostic convergence values ​​in conjunction with the pathological evolution cycle.

[0053] S504, Based on the feature aggregation tensor and the pathological evolution cycle data, calculate the diagnostic convergence value corresponding to the symptoms.

[0054] The system uses a weighted fusion-derived feature aggregation tensor as input data, while simultaneously referencing pathological evolution cycle data from a medical record database as a temporal reference. The multi-dimensional symptom association information contained in the feature aggregation tensor is aligned and mapped with the time window parameters of the corresponding disease stage in the pathological evolution cycle. Based on this, the system calculates the conditional probability distribution of each candidate disease at each stage of the pathological evolution cycle, and uses the Shannon information entropy of this probability distribution vector as the basic metric. Finally, an inverse proportional mapping function converts the entropy value into a diagnostic convergence value corresponding to the symptom. A lower entropy value indicates a more concentrated match between the symptom features and the specific disease on the pathological timeline, resulting in a higher convergence value; conversely, a higher entropy value indicates a lower convergence value. This achieves a comprehensive quantitative evaluation that integrates static symptom association features with dynamic disease temporal evolution patterns.

[0055] This embodiment obtains pathological evolution cycle data and clinical indication data from a medical record database, and calculates the complication correlation degree based on node feature vectors. When the complication correlation degree exceeds a preset threshold, the node feature vectors are weighted and fused according to clinical indication data to obtain a feature aggregation tensor. Finally, the diagnostic convergence value is calculated in conjunction with the pathological evolution cycle, realizing a comprehensive quantitative assessment that integrates static symptom correlation information with dynamic disease temporal evolution patterns. Because this scheme introduces clinical indication data as a dynamic weight coefficient to correct node features, and provides time window references for each stage of the disease through pathological evolution cycle data, the calculated diagnostic convergence degree can objectively reflect the degree of matching and concentration of symptom clusters in the pathological temporal dimension, rather than simply relying on static statistics of symptom co-occurrence. This significantly improves the clinical reliability and discrimination accuracy of the diagnostic convergence degree in multi-disease coexistence scenarios, effectively avoiding the problem of misdiagnosis or missed diagnosis of comorbidities caused by ignoring the temporal evolution characteristics of the disease.

[0056] In one embodiment, if the initial network density is less than the seepage inflection point threshold, the symptom co-occurrence network is determined to be in a single disease state, and a single etiology diagnosis result is output, including steps S601 to S602, each step of which is as follows: S601, when the initial network density is less than the seepage inflection point threshold, it is determined that the symptom co-occurrence network does not undergo symptom phase transition, and the judgment result that the patient is in a single disease state is obtained.

[0057] The calculated initial network density is compared with a pre-established seepage inflection point threshold. If the initial network density is less than the seepage inflection point threshold, the symptom co-occurrence network is determined to have not reached the critical density required for symptom phase transition. This means the edges connecting symptom nodes in the network are relatively sparse, and a massively connected component spanning the entire network has not yet been formed; each symptom still exists as an isolated local cluster. This network state indicates that the patient's current symptom cluster does not possess the complex characteristics of multi-system concurrent lesions in its topological structure, and the etiology is relatively singular and clear. Based on this, the system determines that the patient is in a single disease state and triggers subsequent single-question processing and single-cause reasoning, thereby avoiding the introduction of overly complex multi-disease analysis logic in simple disease scenarios.

[0058] S602, based on the judgment result, perform single follow-up processing on the symptom co-occurrence network to obtain a single etiology diagnosis result.

[0059] Based on the patient's assessment of a single disease state, the system triggers a single-threaded vertical data mining strategy. This involves identifying the core symptom with the highest weight or most representative characteristics from the initial symptom entity set, and progressively questioning along its single pathological evolution path, exploring highly vertical details such as the nature of pain, duration, triggering factors, and accompanying signs. Throughout the questioning process, a single-disease decision tree model is invoked to filter out interfering questions related to irrelevant physiological systems. The system then transforms the patient's feedback text from each round of questioning into structured features through semantic parsing, inputting it into a single-cause inference engine for joint probability calculation. Finally, it outputs a single-cause diagnosis result pointing to the specific disease name and its confidence level. This allows for accurate diagnosis with the shortest inference path in scenarios with simple and clear symptoms, effectively reducing the patient's interaction burden and improving consultation efficiency.

[0060] This embodiment determines that the symptom co-occurrence network does not undergo a phase transition and confirms that the patient is in a single disease state when the initial network density is less than or equal to the seepage inflection point threshold. This triggers targeted single-cause follow-up processing to output a single-cause diagnosis, effectively simplifying and accelerating the consultation strategy in scenarios with simple and clear symptoms. Because this scheme only enables a single-threaded consultation mode with vertical depth mining when the network density has not crossed the critical value, it avoids using complex general practice screening logic when symptom associations are sparse and the cause is singular. This significantly shortens the computational path for diagnostic reasoning, reduces the cognitive burden and fatigue of patients during human-computer interaction, and ensures the accuracy of single-cause diagnosis by focusing on vertical details of core symptoms. This greatly improves the operational efficiency and user experience of the pre-consultation system while ensuring diagnostic quality.

[0061] In one embodiment, the multi-etiology judgment process includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, and, in response to a judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-etiology diagnostic results, and completing the pre-diagnosis analysis, including steps S701 to S703, each step of which is as follows: S701, patients whose diagnostic convergence is less than a preset convergence threshold are determined to be in a state of multiple coexisting diseases.

[0062] The system compares the calculated diagnostic convergence with a preset convergence threshold. If the convergence is less than the threshold, it indicates that the posterior probability distribution of the current symptom cluster is relatively dispersed, without a steep probability peak concentrated in a few disease directions. Instead, it points to multiple different candidate diseases with relatively uniform probability. Based on this, the system determines that the patient is in a state of multiple coexisting diseases, meaning that their clinical manifestations are not dominated by a single cause, but rather a complex pathological pattern involving two or more independent or mutually influential diseases. This determination provides a clear decision-making basis for subsequent multi-cause inquiry commands and comorbidity diagnosis reasoning, ensuring that the system can automatically switch to a processing flow suitable for multi-disease analysis in complex disease scenarios.

[0063] S702, based on the coexistence of multiple diseases, generate a multi-cause inquiry instruction and obtain the feedback result corresponding to the cause inquiry instruction.

[0064] Based on the determination that the patient is in a state of multiple coexisting diseases, the system activates a multi-causal analysis engine. Using the identified coexisting diseases as target nodes, it traverses the medical knowledge graph to extract the intersection symptoms and specific differential symptoms of each target disease. Centering on multiple disease nodes, its known symptoms are designated as observed nodes, and unqueried symptoms are designated as candidate nodes. The edge weights between each symptom and disease are determined based on the disease-symptom statistical frequency in the medical knowledge graph. For each candidate symptom node, the system calculates the conditional entropy of the current disease set.

[0065] In the formula, This represents the posterior probability of each disease given the presence or absence of the symptom. This probability is calculated using Bayes' theorem based on the prior disease probabilities and the symptom-disease correlation. Information gain is defined as the difference between the disease uncertainty before querying for the symptom and the expected conditional entropy after the query.

[0066] Information gain values ​​are calculated for all candidate symptoms. In practical applications, symptom nodes with an information gain greater than 0.3 are typically selected, or the top 5 symptom nodes with the highest information gain are selected. The comorbidity network updates after each new symptom answer, converting the symptom into an observed node, recalculating the posterior probability distribution of each disease, and updating the information gain ranking of the remaining candidate symptoms. The instruction generation module converts the selected high-value symptom nodes into medical inquiry statements in natural language, thereby generating multi-causal follow-up instructions. Subsequently, the system receives the feedback text returned by the patient in response to each round of follow-up questions, extracts new symptom entities and sign descriptions through semantic parsing and named entity recognition technology, and uses this as structured input data for subsequent updates to the comorbidity inference model, thus completing the targeted collection of key identification information in the state of multiple coexisting diseases.

[0067] S703, Based on the feedback results, the symptoms are analyzed and diagnosed to obtain multi-causal diagnosis results, and the pre-diagnosis analysis is completed.

[0068] The system extracts newly added symptom entities and sign descriptions from the patient's follow-up questions regarding multiple causes through semantic parsing, and merges these with the initial symptom information to form a symptom feature set. Subsequently, using this symptom feature set as input, the system performs joint inference in the comorbidity diagnosis model. Through Bayesian posterior inference, it recalculates the corrected posterior probabilities of each coexisting disease at each stage of the pathological evolution cycle. Disease nodes with probability values ​​exceeding a preset threshold are then matched and verified using a comorbidity pattern rule base. Finally, the system outputs a complete multi-cause diagnosis result, including the primary disease, coexisting diseases, and their respective confidence levels. This enables accurate identification of multiple coexisting diseases in complex medical scenarios, completing the entire analysis process for intelligent pre-diagnosis for patients.

[0069] This embodiment first generates multi-causal inquiry instructions and obtains patient feedback after determining that the patient is in a state of multiple coexisting diseases. Then, based on the feedback results, it performs diagnostic analysis and outputs multi-causal diagnostic results, constructing a complete adaptive consultation and reasoning closed loop for complex conditions. Because this scheme only activates the multi-causal analysis engine after confirming the coexistence of multiple diseases, it selects high-value symptoms for targeted inquiry based on the criterion of maximizing information gain. This effectively avoids redundant questioning of patients in single-disease scenarios. At the same time, by integrating the feedback results into the comorbidity diagnosis model after semantic parsing for joint probability updates, the final output multi-causal diagnostic results fully integrate the initial symptom information and the discriminative features obtained from the inquiry. This significantly improves the comprehensiveness and accuracy of comorbidity identification, thereby ensuring efficient consultation for patients with mild symptoms while ensuring that complex patients can obtain accurate pre-consultation conclusions without omitting coexisting diseases. This fully demonstrates the system's adaptive diagnostic capabilities under different disease complexities.

[0070] In one exemplary embodiment, such as Figure 3As shown, an intelligent pre-diagnosis analysis system for patient services is provided, including: The acquisition module 801 is used to acquire the patient's initial symptom information, construct a symptom co-occurrence network based on the initial symptom information, and determine the initial network density of the symptom co-occurrence network. The judgment module 802 is used to compare the initial network density with a preset seepage inflection point threshold. If the initial network density is less than the seepage inflection point threshold, the symptom co-occurrence network is determined to be in a single disease state, and a single etiology diagnosis result is output. If the initial network density is not less than the seepage inflection point threshold, a multi-etiology judgment process is executed. The analysis module 803 is used in the multi-cause judgment process, which includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, and in response to a judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-cause diagnosis results, and completing the pre-diagnosis analysis.

[0071] In one embodiment, a computer program product is provided, including a computer program or instructions that, when executed by a device, implement the steps of a patient-oriented intelligent pre-diagnosis analysis method as described above.

[0072] 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 in the above method embodiments.

[0073] This technical solution deeply integrates random forest entity extraction with graph neural network topology analysis and incorporates the percolation phase transition critical mechanism of symptom co-occurrence networks to construct an intelligent diagnostic strategy adaptive switching system specifically for pre-consultation scenarios. Its core benefit lies in overcoming the limitations of existing technologies that are limited to single-cause inquiry or passive symptom accumulation. It achieves dynamic monitoring and critical point identification of symptom cluster network density during the consultation process. Through real-time network density comparison and diagnostic convergence calculation, the system proactively determines whether the current symptom cluster is in a stage where it can converge to a single diagnosis or has crossed the critical point into a state of multiple co-occurrence. Based on this, it promptly switches inquiry strategies, providing a quantitative decision-making basis for the accurate identification of comorbid patients and achieving a leap from single-disease inquiry to multi-cause investigation. Simultaneously, this solution fully leverages the respective advantages of random forest in text entity extraction, graph neural networks in topology analysis, and percolation theory in critical state determination, solving the core problems of traditional pre-consultation systems lacking the ability to identify symptom network phase transitions and struggling to avoid missed diagnoses of comorbidities.

[0074] For the device embodiments, since they basically correspond to the method embodiments, the relevant details can be found in the descriptions of the method embodiments. The device embodiments described above are merely illustrative; components described as separate parts may or may not be physically separate, and components shown as units may or may not be physical units, meaning they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0075] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A patient-oriented intelligent pre-diagnosis analysis method, characterized in that, include: Obtain the patient's initial symptom information, construct a symptom co-occurrence network based on the initial symptom information, and determine the initial network density of the symptom co-occurrence network; The initial network density is compared with a preset seepage inflection threshold. If the initial network density is less than the seepage inflection threshold, the symptom co-occurrence network is determined to be in a single disease state, and a single etiology diagnosis result is output. If the initial network density is not less than the seepage inflection threshold, a multi-etiology judgment process is executed. The multi-cause diagnosis process includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, and, in response to a judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-cause diagnosis results, and completing the pre-diagnosis analysis.

2. The intelligent pre-diagnosis analysis method for patient services as described in claim 1, characterized in that, The process of setting the seepage inflection point threshold includes: Density evolution sequences were obtained from the symptom co-occurrence network; Calculate the percentage of nodes corresponding to different network densities in the density evolution sequence; Based on the rate of change of the node proportion, the network density corresponding to the maximum value of the rate of change is determined as the seepage inflection point threshold.

3. The intelligent pre-diagnosis analysis method for patient services as described in claim 2, characterized in that, The process of performing topology analysis on the symptom co-occurrence network and calculating the diagnostic convergence of the symptom co-occurrence network includes: When the initial network density is not less than the seepage inflection point threshold, it is determined that the symptom co-occurrence network has undergone a symptom phase transition. The topology of the symptom co-occurrence network is analyzed to obtain the symptom node features. Based on the symptom node features, the diagnostic convergence of the symptom is calculated.

4. The intelligent pre-diagnosis analysis method for patient services as described in claim 3, characterized in that, The topology analysis of the symptom co-occurrence network to obtain symptom node features includes: The topology of the symptom co-occurrence network is analyzed to determine multiple cluster centers in the symptom co-occurrence network; The state evolution features of the symptom co-occurrence network are extracted based on the cluster centers; Based on the state evolution characteristics, the symptom co-occurrence network is divided to obtain multiple connected components containing symptom node features.

5. The intelligent pre-diagnosis analysis method for patient services as described in claim 3, characterized in that, The step of calculating the diagnostic convergence of the symptom based on the symptom node features includes: Obtain pathological evolution cycle data and clinical indication data from the medical record database; Extract node feature vectors from the symptom node features, and calculate the complication correlation degree based on the node feature vectors; In response to a judgment command that the correlation degree of the complication is greater than a preset correlation degree threshold, the node feature vector is weighted and fused according to the clinical indication data to obtain a feature aggregation tensor; Based on the feature aggregation tensor and the pathological evolution cycle data, the diagnostic convergence value corresponding to the symptoms is calculated.

6. The intelligent pre-diagnosis analysis method for patient services as described in claim 1, characterized in that, If the initial network density is less than the seepage inflection point threshold, then the symptom co-occurrence network is determined to be in a single disease state, and a single etiology diagnosis result is output, including: When the initial network density is less than the seepage inflection point threshold, it is determined that the symptom co-occurrence network does not undergo symptom phase transition, and the judgment result that the patient is in a single disease state is obtained. Based on the judgment result, a single follow-up question is performed on the symptom co-occurrence network to obtain a single etiological diagnosis result.

7. The intelligent pre-diagnosis analysis method for patient services as described in claim 1, characterized in that, The multi-etiology judgment process includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, and, in response to a judgment command that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-etiology diagnostic results, and completing the pre-diagnosis analysis, including: Patients whose diagnostic convergence is less than a preset convergence threshold are determined to be in a state of multiple coexisting diseases; Based on the coexistence of multiple diseases, a multi-cause inquiry instruction is generated, and the feedback result corresponding to the cause inquiry instruction is obtained. Based on the feedback results, the symptoms are analyzed and diagnosed to obtain multi-causal diagnostic results, thus completing the pre-diagnosis analysis.

8. A patient-oriented intelligent pre-diagnosis analysis system, characterized in that, The system includes: An acquisition module is used to acquire the patient's initial symptom information, construct a symptom co-occurrence network based on the initial symptom information, and determine the initial network density of the symptom co-occurrence network. The judgment module is used to compare the initial network density with a preset seepage inflection point threshold. If the initial network density is less than the seepage inflection point threshold, the symptom co-occurrence network is determined to be in a single disease state, and a single etiology diagnosis result is output. If the initial network density is not less than the seepage inflection point threshold, a multi-etiology judgment process is executed. The analysis module, used in the multi-causal diagnosis process, includes: performing topology analysis on the symptom co-occurrence network, calculating the diagnostic convergence of the symptom co-occurrence network, responding to a judgment instruction that the diagnostic convergence is less than a preset convergence threshold, determining that the patient is in a state of multiple co-occurrence of diseases, outputting multi-causal diagnosis results, and completing the pre-diagnosis analysis.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program or instructions, which, when executed by a communication device, implement the intelligent pre-diagnosis analysis method for patient services as described in any one of claims 1-7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by the communication device, they implement the intelligent pre-diagnosis analysis method for patient services as described in any one of claims 1-7.

Images