Medical history reconstruction system and method based on symptom evolution graph and anti-factual reasoning

Abstract

The invention provides a medical history reconstruction system and method based on a symptom evolution graph and anti-factual reasoning, and the system comprises a symptom spatio-temporal evolution graph construction module which maps the scattered symptom description of a patient into a four-dimensional spatio-temporal coordinate system; the medical history fragment intelligent splicing engine adopts an algorithm based on a minimum description length principle to reconstruct a most probable complete medical history timeline; the active symptom inducing module is used for generating a specific action instruction or an inducible question based on the current symptom characteristics; the anti-fact diagnosis verification framework is used for evaluating the contribution degree of each symptom to the final diagnosis and identifying key symptoms and redundant information; and the symptom description standardization converter is used for automatically converting the description into standard medical terms by using a semantic mapping network based on medical ontology, and meanwhile, a nuisance mark of original semantics is reserved. Through construction of the symptom spatio-temporal evolution graph, the system can reconstruct scattered and jumping narration of patients into a complete and coherent medical history timeline, and the symptom omission rate is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical artificial intelligence, in particular to a medical history reconstruction system and method based on symptom evolution map and counterfactual reasoning. BACKGROUND

[0002] In modern medical diagnosis, accurately obtaining and understanding patient medical history is a key prerequisite for doctors to make correct diagnoses. However, there is a common problem of incomplete and inaccurate medical history collection in clinical practice. When describing symptoms, patients often show highly unstructured characteristics. Their narratives are usually fragmented, jumping, and lack clear temporal order and logical relationships. Many patients have difficulty accurately recalling the specific time of symptom onset, often using vague time descriptions such as "a few days ago" or "for a while", making it difficult for doctors to establish an accurate timeline of symptom development.

[0003] Traditional medical history collection methods mainly rely on the experience and interrogation skills of doctors, gradually obtaining information through a question-and-answer approach. This method has many limitations: first, different doctors' interrogation levels and experience differ greatly, which may lead to the omission of key information; second, patients' memory biases and expression limitations affect the accuracy of information; third, patients often use dialects, colloquialisms, or emotional language to describe symptoms, such as "heart cavity blocked" or "head buzzing", which makes it difficult for doctors to understand and record these non-standardized descriptions.

[0004] More importantly, there are complex causal relationships and transmission mechanisms between many symptoms. For example, cervical spine lesions can cause headaches, dizziness, and further symptoms such as nausea, but patients often only focus on the most obvious or most disturbing symptoms, ignoring the internal connections between symptoms. At the same time, some latent symptoms may need specific triggering conditions to manifest, such as chest tightness symptoms only appearing during exercise for certain heart diseases, and patients may not feel any discomfort in a resting state, leading to important diagnostic clues being missed.

[0005] Existing electronic medical record systems, while standardizing medical history records to some extent, are mostly just electronic versions of paper medical records, lacking intelligent information extraction and analysis capabilities. These systems cannot automatically identify associations between symptoms, cannot extract structured information from patients' natural language descriptions, and cannot actively guide doctors to discover potential missing symptoms. In the face of complex cases, doctors still need to rely on personal experience to construct a spatiotemporal evolution map of symptoms in their minds, which is not only inefficient but also prone to judgment bias. SUMMARY

[0006] To overcome the shortcomings of existing technologies, this invention proposes a medical history reconstruction system and method based on symptom evolution mapping and counterfactual reasoning. Through active symptom induction technology, it can discover hidden symptoms that patients are unaware of or find difficult to describe. This is of great significance for the diagnosis of diseases with complex symptom evolution, such as neurological disorders and chronic diseases. The counterfactual reasoning framework further ensures that key symptoms are not overlooked, improving diagnostic accuracy.

[0007] To achieve the above objectives, this invention proposes an intelligent medical history reconstruction system based on symptom evolution maps and counterfactual reasoning, characterized by comprising:

[0008] The symptom spatiotemporal evolution map construction module maps the patient's scattered symptom descriptions to a four-dimensional spatiotemporal coordinate system, where three dimensions represent the body's anatomical location and one dimension represents the time axis. The symptom transmission path analysis algorithm automatically infers the causal relationship and transmission order between symptoms.

[0009] The intelligent medical history splicing engine uses an algorithm based on the minimum description length (MDL) principle to extract symptom fragments from the patient's non-linear and disjointed narrative. Through temporal constraint solving and medical common sense reasoning, it reconstructs the most likely complete timeline of the medical history.

[0010] The active symptom triggering module generates specific action instructions or triggering questions based on the current symptom characteristics. By monitoring the patient's response when executing the instructions in real time, it can discover hidden symptoms that the patient is unaware of or finds difficult to describe.

[0011] The counterfactual diagnostic verification framework assesses the contribution of each symptom to the final diagnosis and identifies key symptoms and redundant information by constructing counterfactual scenarios such as "removing a symptom" or "changing the order of symptom appearance".

[0012] The symptom description standardization converter uses a semantic mapping network based on medical ontology to automatically convert patients' colloquial and emotional descriptions into standard medical terms, while preserving the subtle differences in the original semantics.

[0013] Furthermore, the symptom spatiotemporal evolution map construction module specifically includes: an anatomical location recognition submodule, which identifies body parts in symptom descriptions through natural language processing and maps them to a standard anatomical coordinate system; and a symptom transmission path inference submodule, which infers possible symptom transmission paths based on the physiological laws of nerve conduction, blood circulation, and the lymphatic system, using the following symptom transmission path probability model:

[0014]

[0015] Wherein, P(S) j |S i ) indicates from symptom S i Transmission to symptoms Sj the conditional probability of d ij is the geodesic distance of symptoms i and j in the anatomical space, Δt ij is the time interval of symptom occurrence, θ ij is the angle between the symptom transmission direction and the physiological transmission path, λ is the distance attenuation parameter, α and β are the time and direction weight parameters, is the neighborhood set of symptom i, σ(·) is the sigmoid activation function; the time anchor extraction submodule identifies relative time markers and absolute time reference points from patient narratives and constructs a time sequence graph of symptom occurrence; the symptom intensity evolution modeling submodule uses Gaussian process regression to fit the curve of symptom intensity change over time.

[0016] Further, the medical history fragment intelligent splicing engine includes: a narrative segment division module that uses rules based on transition words and time words to divide patient narratives into independent segments; a segment temporal constraint solver that constructs a set of temporal consistency constraint equations and uses a constraint satisfaction algorithm to solve the optimal timeline, with the optimization objective function based on the MDL principle:

[0017]

[0018] where L MDL is the minimum description length loss, P(D|H) is the likelihood probability of symptom data D given the medical history hypothesis H, H(T i ) is the information entropy of the i-th time sequence segment, γ i is the importance weight of the i-th segment, Ω(G) is the complexity regular term of the temporal constraint graph G, and n is the total number of segments; a medical logic verification module that verifies the medical rationality of the reconstructed timeline based on a disease natural history knowledge base; a conflict resolution module that, when a temporal conflict occurs, resolves the ambiguity by asking key time nodes.

[0019] Further, the active symptom induction module includes: an induction action generator that generates specific body position changes, breathing patterns, or limb action instructions according to suspected diseases; a multi-modal response monitor that synchronously collects facial expressions, voice changes, and physiological parameters when the patient performs the action; a symptom induction evaluator that quantitatively evaluates the degree of symptom change caused by the induced action, using the following response intensity calculation formula:

[0020]

[0021] where R(a k ) is the response intensity of the k-th induced action a k , I m (t) is the signal intensity of the m-th monitoring index at time t, w mis the weight coefficient of the m-th modality, τ is the time decay parameter, t0 and t1 are the monitoring start and end time respectively, and M is the total number of monitoring modalities; a dynamic induction strategy optimizer, which continuously optimizes the selection strategy of the induction action based on historical induction effects.

[0022] Further, the counterfactual diagnosis verification framework comprises: a symptom importance quantification module, which quantifies the importance of a symptom by calculating the change in diagnosis probability after removing a single symptom, and adopts the following scoring formula:

[0023]

[0024] wherein, is the importance score of symptom s i , KL(·‖·) is the Kullback-Leibler divergence, P(D|S) is the diagnosis probability distribution given the symptom set S, P(D|S\{s i}) is the diagnosis probability distribution after removing symptom s i , is the conditional mutual information, S -i denotes the set of other symptoms except s i , φ is the balance parameter, denotes the mathematical expectation; a symptom combination effect analysis module, which identifies symptom combinations with synergistic effects; a diagnosis path visualization module, which displays the reasoning path from symptoms to diagnosis and the contribution weight of each path; and a diagnosis stability evaluation module, which evaluates the robustness of the diagnosis result through symptom perturbation testing.

[0025] Further, the symptom description standardization converter comprises: a dialect symptom dictionary, which records symptom description words in various dialects and their standard mappings; an emotional modifier filter, which identifies and labels emotional modification components, and extracts core symptom information; a fuzzy description quantifier, which converts fuzzy descriptions such as “a little pain” and “very painful” into standard pain scores; and a synonymous symptom clusterer, which identifies descriptions of different expressions but pointing to the same symptom, and adopts the following semantic similarity measure:

[0026]

[0027] wherein, Sim(d p ,d s ) is the similarity of patient description d p and standard term d s , v p and v s are the semantic embedding vectors of the patient description and the standard term respectively, W is a domain-specific transformation matrix, ‖·‖2 represents the L2 norm, EditDist(·,·) is the edit distance function, d p and d srepresents the character length of the description, and μ is the surface similarity weight coefficient.

[0028] Furthermore, it also includes: a symptom omission prediction module, which predicts accompanying symptoms that patients may miss or ignore based on existing symptoms and disease patterns; a medical history credibility assessment module, which assesses the overall credibility of medical history through multi-dimensional indicators such as symptom consistency and temporal rationality; and a doctor-patient communication optimization module, which analyzes the information transmission efficiency of doctor-patient dialogue in real time and prompts doctors to adjust their communication strategies.

[0029] The medical history reconstruction method based on symptom evolution maps and counterfactual reasoning is applicable to the aforementioned medical history reconstruction system based on symptom evolution maps and counterfactual reasoning, and includes the following steps:

[0030] S1: Receive the patient's initial chief complaint and extract key symptoms and time anchors;

[0031] S2: Construct a preliminary spatiotemporal evolution map of symptoms, and mark the anatomical location and temporal sequence of symptoms;

[0032] S3: Infer possible hidden symptoms based on symptom transmission patterns and generate targeted induction test plans;

[0033] S4: Perform symptom provocation tests, monitor and record patient responses;

[0034] S5: Integrate all symptom information and reconstruct a complete medical history timeline using the MDL principle;

[0035] S6: Verify key symptoms through counterfactual reasoning to assess the reliability of the diagnosis;

[0036] S7: Convert the patient's colloquial description into a standardized medical history record;

[0037] S8: Assess the completeness of the medical history, identify potential omissions, and ask further questions.

[0038] Furthermore, the medical history reconstruction in step S5 specifically includes: representing symptom fragments as nodes of a temporal constraint graph; defining temporal relationship constraints between fragments; using a constraint optimization algorithm to solve for the optimal timeline that satisfies all constraints; and verifying and correcting unreasonable temporal relationships based on medical knowledge.

[0039] Furthermore, the counterfactual reasoning in step S6 includes:

[0040] Construct a diagnostic model that includes a complete set of symptoms;

[0041] Remove symptoms one by one and recalculate the diagnosis probability;

[0042] Identify key symptoms that lead to significant changes in the probability of diagnosis;

[0043] Verify the accuracy of key symptoms by follow-up.

[0044] The beneficial effects of the present application compared with the prior art are:

[0045] 1. The present application provides a medical history reconstruction system and method based on symptom evolution map and counterfactual reasoning. Through the construction of symptom spatio-temporal evolution map, the system can reconstruct the patient's scattered and jumping narrative into a complete and coherent medical history timeline, reducing the symptom omission rate. Especially through the active symptom induction technology, hidden symptoms that patients are unaware of or difficult to describe can be found, which is of great significance for the diagnosis of diseases with complex symptom evolution such as nervous system diseases and chronic diseases. The counterfactual reasoning framework further ensures that key symptoms are not overlooked, improving the accuracy of diagnosis.

[0046] 2. The present application provides a medical history reconstruction system and method based on symptom evolution map and counterfactual reasoning. The system shortens the average interrogation time from 18 minutes to 12 minutes, improving efficiency.

[0047] 3. The present application provides a medical history reconstruction system and method based on symptom evolution map and counterfactual reasoning. The automatic medical history recording and standardized conversion function significantly reduces the medical staff's paperwork burden, allowing doctors to devote more energy to diagnosis and patient communication.

[0048] 4. The present application provides a medical history reconstruction system and method based on symptom evolution map and counterfactual reasoning. The symptom description standardized converter can accurately understand various local dialects and colloquial expressions, breaking down the language barrier between doctors and patients. The system retains the subtle difference markers of the original semantics, achieving standardization of medical terminology without losing important information in patient expression. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the specific embodiments or prior art of the present application, the following will briefly introduce the drawings needed in the specific embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0050] Figure 1 is a schematic diagram of the system process of the present application DETAILED DESCRIPTION

[0051] The technical solutions of the present application will be described more clearly and completely by combining the drawings and the description of the preferred embodiments of the present application.

[0052] As Figure 1As shown, the system adopts a modular hierarchical architecture design, realizing an intelligent processing flow from the patient's unstructured symptom description to the standardized medical history reconstruction.

[0053] The input end of the system receives the patient's scattered and unstructured symptom descriptions, which often contain dialectal expressions, emotional modifications, and time sequence confusion. First, the symptom description standardization converter preprocesses the original input by mapping dialectal dictionaries, filtering emotional modification words, and quantifying fuzzy descriptions, converting the patient's colloquial expressions into standardized medical terminology, laying the foundation for subsequent processing.

[0054] The standardized symptom information flows to two core processing modules simultaneously. The medical history fragment intelligent splicing engine is responsible for processing the information reconstruction in the time sequence dimension. Through narrative segment division, the chaotic description is divided into independent fragments, and the optimization algorithm based on the minimum description length (MDL) principle is used to reconstruct the most possible symptom development timeline under the premise of meeting the medical logic check. In parallel with this, the symptom spatiotemporal evolution map construction module focuses on the mapping of the spatial dimension, determines the body position of the symptoms through anatomical positioning recognition, analyzes the physiological relationship between symptoms using conduction path reasoning, and constructs the propagation and evolution model of symptoms in the human body by combining time anchor extraction and intensity evolution modeling.

[0055] The processing results of the two modules converge in the symptom spatiotemporal data integration layer, forming a four-dimensional coordinate system, of which three dimensions represent anatomical spatial positions and one dimension represents the time axis. This four-dimensional representation method can comprehensively depict the spatiotemporal evolution characteristics of symptoms, providing a unified data framework for in-depth analysis.

[0056] Based on the integrated four-dimensional symptom data, the system enters the verification and enhancement stage. The active symptom induction module actively discovers hidden symptoms that the patient is unaware of or difficult to describe by generating specific induction actions and cooperating with multi-modal reaction monitoring. This module uses a symptom induction scoring mechanism to quantitatively evaluate the induction effect and continuously improves the induction scheme through dynamic strategy optimization. At the same time, the counterfactual diagnosis verification framework verifies the diagnosis from another angle, evaluates the contribution of each symptom to the diagnosis through symptom importance quantification, analyzes the combination effect between symptoms, and provides multi-dimensional verification of the reliability of the diagnosis through diagnosis path visualization and stability evaluation.

[0057] It is worth noting that the active symptom induction module also feeds back the newly discovered symptom information to the data integration layer through a feedback loop mechanism, realizing the dynamic optimization and continuous improvement of the system.

[0058] Finally, the system outputs the complete history reconstruction results, including the standardized history timeline, key symptom annotations, and diagnosis confidence assessment, providing comprehensive, accurate, and reliable information support for doctors' diagnosis decisions. Through multi-level intelligent processing and cross-validation, the entire system effectively solves the problems of information loss, time sequence confusion, and non-standard expression in traditional history collection, significantly improving the quality of history information and the accuracy of diagnosis.

[0059] As a specific embodiment, the present application provides an intelligent history reconstruction system and method based on symptom evolution map and counterfactual reasoning, which can reconstruct a complete and accurate history timeline from patients' scattered and nonlinear symptom descriptions. The system adopts a distributed architecture design, with the front end interacting with patients through a Web interface or mobile APP and the back end deployed on a cloud server, including multiple functional modules working together.

[0060] When the patient starts describing the symptoms, the system first analyzes the input in real time through natural language processing technology. For example, the patient says "I started having a headache on the right side yesterday, and last night I felt a little uncomfortable in my left shoulder, and oh, I've been feeling stiff in my neck since last Tuesday," the system will immediately identify three symptom fragments and their time markers. The anatomical positioning identification sub-module maps "right head" to the (x = + 15, y = 0, z = + 180) position in the standard anatomical coordinate system, "left shoulder" to (x = -20, y = + 30, z = + 140), and "neck" to (x = 0, y = 0, z = + 150), with the origin at the center of gravity of the human body and the unit in centimeters.

[0061] The symptom transmission path reasoning sub-module then starts working, calculating the transmission probability between symptoms. Taking neck stiffness to shoulder discomfort as an example, the system first calculates the geodesic distance between the two symptoms in the anatomical space d = 18.7 cm, considering the physiological transmission path along the trapezius muscle and cervical nerve, the time interval Δt = 5 days, and the angle θ = 15° between the transmission direction and the cervical nerve. Substituting the formula gives P(shoulder discomfort | neck stiffness) = 0.73, indicating that there is a high causal relationship between the two symptoms. The system uses a distance decay parameter λ = 0.05, weight parameters α = 0.3 and β = 0.2, which are obtained through machine learning training on 100,000 clinical cases.

[0062] After constructing the initial symptom profile, the medical history fragment intelligence stitching engine starts to reconstruct the complete timeline. The system segments the patient's narrative into independent fragments, each containing a symptom description and time information. For the above case, the system constructs a temporal constraint graph G with 3 nodes, the constraint relations between nodes include: "neck stiffness" must be at least 3 days earlier than "shoulder discomfort", "shoulder discomfort" must be at least 1 day earlier than "headache". The system uses a constraint satisfaction algorithm to solve, the optimization goal is to minimize the MDL loss function. In the calculation process, the likelihood probability P(D|H) is estimated by the Bayesian network, the information entropy H(Ti) of each temporal fragment is calculated according to the complexity of the symptom description, and the constraint graph complexity Ω(G) is set as the logarithm of the number of edges. After iterative optimization, the system reconstructs the timeline: neck stiffness on day 1 → left shoulder discomfort on day 5 → right side headache on day 6.

[0063] To discover hidden symptoms, the active symptom induction module generates a targeted test plan. Based on the neck symptoms, the system instructs the patient: "Please slowly turn your head to the left and then to the right, and tell me how you feel." The multi-modal response monitor works simultaneously, capturing the patient's facial expressions through the camera and detecting a slight frown (facial action unit AU4 intensity of 2) when the head is turned 25° to the right, and the microphone detects a slight inhalation sound, indicating a pain response. The symptom induction scorer calculates the response intensity R=3.7, with facial expression contributing a weight w1=0.4, sound change weight w2=0.3, action fluency weight w3=0.3, and time decay parameter τ=2 seconds. This result suggests the presence of a hidden symptom of limited cervical rotation.

[0064] The counterfactual diagnosis verification framework then evaluates the diagnostic contribution of each symptom. The system constructs an initial diagnosis model containing all symptoms, calculating a probability of cervical spondylosis of 0.82. Through counterfactual analysis, removing "neck stiffness" reduces the diagnosis probability to 0.31, removing "headache" reduces it to 0.65, and removing "shoulder discomfort" reduces it to 0.71. Using the importance score formula, the importance score of neck stiffness I=2.45, headache I=0.98, and shoulder discomfort I=0.67. The KL divergence measures the degree of change in the diagnosis distribution, the conditional mutual information captures the dependency relationship between symptoms, and the balance parameter is set to 0.4. The results show that neck stiffness is a key symptom, and the system therefore generates follow-up questions: "Is your neck stiffness persistent or intermittent? Does your neck make a noise when you turn it?"

[0065] The symptom description standardizer converts the patient's colloquial description into medical terminology. When the patient says "neck is as hard as being choked," the system first extracts the emotional modifier "as hard as being choked" and identifies the core symptom as "neck discomfort." Through a pre-trained medical language model, the patient's description is encoded into a 768-dimensional semantic vector vp, which is compared with "neck muscle spasm," "neck compression," "neck pain," and other standard terms in the library. The system uses a domain-specific transformation matrix W (768 x 768 dimensions), which is trained on medical text corpus through contrastive learning. The calculation result shows that the similarity with "neck compression" is the highest, Sim = 0.87, of which the semantic similarity contribution is 0.72 and the surface similarity contribution is 0.15.

[0066] The system also includes a symptom omission prediction module that predicts possible missed accompanying symptoms based on the constructed symptom evolution map and medical knowledge base. For cervical spondylosis cases, the system suggests that "finger numbness" (prediction probability 0.65) and "sleep quality decline" (prediction probability 0.58) may exist. The medical history credibility assessment module comprehensively assesses the reliability of the reconstructed medical history, considering the consistency of symptom timing (score 0.89), the rationality of anatomical location (score 0.92), and the compliance of symptom evolution with medical rules (score 0.87), with an overall credibility of 0.89.

[0067] In actual deployment, the system uses a microservice architecture, with each functional module deployed independently and communicating through a message queue. The natural language processing module uses the BERT-base model, fine-tuned on 1 million medical conversation data. The symptom transmission path reasoning uses a graph neural network, which includes 5 layers of graph convolution layers. The temporal constraint solver uses the open-source OR-Tools library. Multi-modal monitoring uses OpenCV for video processing and a pre-trained facial action unit detection model. All patient data is encrypted and stored in compliance with HIPAA privacy standards.

[0068] The system has been in trial operation in the Department of Neurology of a certain third-grade hospital for 3 months, handling 2847 patients. Compared with traditional interrogation, the average interrogation time is shortened from 18 minutes to 12 minutes, the symptom omission rate is reduced from 23% to 7%, and the diagnostic accuracy is improved from 81% to 89%. Especially for elderly patients and children with unclear symptom expression, the system significantly improves the completeness of symptom recognition through multi-modal monitoring and active induction. Doctor feedback shows that the structured medical history record generated by the system greatly reduces the clerical work burden, allowing doctors to devote more energy to diagnosis and decision-making.

[0069] The above detailed description merely describes the preferred embodiment of the application, and is not intended for restricting the protection scope of the application. Without departing from the design concept and spirit of the application, various modifications, replacements and improvements of the technical solutions of the application made by those skilled in the art according to the description and drawings provided by the application should all belong to the protection scope of the application. The protection scope of the application is defined by the claims.

Claims

1. A system for reconstructing medical history based on symptom evolution map and counterfactual reasoning, characterized in that, Comprise: Symptom spatiotemporal evolution graph construction module, mapping patients' scattered symptom descriptions to a four-dimensional spatiotemporal coordinate system, where three dimensions represent anatomical locations and one dimension represents the time axis, automatically inferring the causal relationship and transmission order between symptoms through symptom transmission path analysis algorithm; Medical history fragment intelligent splicing engine, using an algorithm based on the minimum description length principle to extract symptom fragments from patients' nonlinear and jump-like narratives, reconstructing the most possible complete medical history timeline through temporal constraint solving and medical common sense reasoning; Active symptom induction module, generating specific action instructions or inducing questions based on current symptom characteristics, discovering hidden symptoms that patients are unaware of or difficult to describe by monitoring patients' reactions when executing instructions in real time; Counterfactual diagnosis verification framework, evaluating the contribution of each symptom to the final diagnosis by constructing counterfactual scenarios that remove certain symptoms or change the order of symptom appearance, identifying key symptoms and redundant information; Symptom description standardization converter, using a semantic mapping network based on medical ontology to automatically convert patients' popularized and emotional descriptions into standard medical terminology while preserving the subtle differences in original semantics.

2. The symptom evolution map and counterfactual reasoning based medical history reconstruction system according to claim 1, characterized in that, The symptom spatiotemporal evolution graph construction module specifically comprises: Anatomy positioning identification submodule, identifying body parts in symptom descriptions through natural language processing and mapping them to a standard anatomical coordinate system; where P(S j |S i ) denotes the conditional probability of transmitting from symptom S i to symptom S j , d ij is the geodesic distance between symptoms i and j in the anatomical space, Δt ij is the time interval between symptom occurrences, θ ij is the angle between the symptom transmission direction and the physiological transmission path, λ is the distance decay parameter, and α and β are the time and direction weight parameters, is the neighborhood set of symptom i, and σ(·) is the sigmoid activation function. Symptom transmission path reasoning submodule, inferring possible transmission paths of symptoms based on the physiological laws of nerve conduction, blood circulation, and lymphatic system, using the following symptom transmission path probability model: Time anchor extraction submodule, identifying relative time markers and absolute time reference points from patient narratives to construct a time sequence relationship graph of symptom appearance; 3. The symptom evolution map and counterfactual-based medical history reconstruction system of claim 1, wherein, Symptom intensity evolution modeling submodule, using Gaussian process regression to fit the curve of symptom intensity change over time. The medical history fragment intelligent splicing engine includes: Narrative fragment segmentation module, using rules based on transition words and time words to segment patient narratives into independent fragments; where L MDL is the minimum description length loss, P(D|H) is the likelihood probability of the symptom data D given the history hypothesis H, H(T i ) is the information entropy of the i-th time slice, γ i is the importance weight of the i-th slice, Ω(G) is the complexity regularizer of the temporal constraint graph G, and n is the total number of slices. Fragment temporal constraint solver, constructing a set of temporal consistency constraint equations and solving the optimal timeline using constraint satisfaction algorithms, with the optimization objective function based on the MDL principle:

4. The symptom evolution map and counterfactual-based medical history reconstruction system of claim 1, wherein, Medical logic verification module, verifying the medical rationality of the reconstructed timeline based on the disease natural history knowledge base; Conflict resolution module, when there is a temporal conflict, resolving the ambiguity by asking key time nodes. The active symptom induction module includes: Inducing action generator, generating specific body position changes, breathing patterns, or limb action instructions based on suspected diseases; Multi-modal response monitor, synchronously collecting facial expressions, voice changes, and physiological parameters when patients perform actions; wherein R(a k ) is the response intensity of the kth induced action a k , I m (t) is the signal intensity of the mth monitoring index at time t, w m is the weight coefficient of the mth mode, τ is the time decay parameter, t0 and t1 are the monitoring start and end times, respectively, and M is the total number of monitoring modes; Symptom induction scorer, quantitatively evaluating the degree of symptom change caused by inducing actions using the following response intensity calculation formula:

5. The symptom evolution map and counterfactual-based medical history reconstruction system of claim 1, wherein, Dynamic induction strategy optimizer, continuously optimizing the selection strategy of inducing actions based on historical induction effects. where, is the importance score of the symptom s i , KL(·‖·) is the Kullback-Leibler divergence, P(D|S) is the diagnostic probability distribution given the symptom set S, P(D|S\{s i}) is the diagnostic probability distribution after removing the symptom s i , I(·;·|·) is the conditional mutual information, S -i denotes the set of other symptoms except s i , φ is the balance parameter, denotes the mathematical expectation; The counterfactual diagnosis verification framework includes: Symptom importance quantification module, quantifying symptom importance by calculating the change in diagnosis probability after removing a single symptom, using the following scoring formula: Symptom combination effect analysis module, identifying symptom combinations with synergistic effects; A diagnostic path visualization module that presents the reasoning path from symptoms to diagnosis and the contribution weight of each path; A diagnostic stability evaluation module that assesses the robustness of the diagnosis result through symptom perturbation testing.

6. The symptom evolution map and counterfactual-based medical history reconstruction system of claim 1, wherein, The symptom description standardization converter includes a dialect symptom dictionary that records symptom description vocabulary in various dialects and their standard mappings; An emotional modifier filter that identifies and labels emotional modifiers, extracting core symptom information; A fuzzy description quantifier that converts fuzzy descriptions into standard pain scores; A synonymous symptom clusterer that identifies descriptions of different expressions but pointing to the same symptom, using the following semantic similarity measure: where Sim(d p ,d s ) is the similarity of the patient description d p to the standard term d s , v p and v s are the semantic embedding vectors of the patient description and the standard term respectively, W is a domain-specific transformation matrix, ||·||2 denotes the L2 norm, EditDist(·,·) is the edit distance function, d p and d s denote the length of the description respectively, and μ is the surface similarity weight coefficient.

7. The symptom evolution map and counterfactual-based medical history reconstruction system of claim 1, wherein, Also includes: A symptom omission prediction module that predicts possible missed or overlooked accompanying symptoms based on existing symptoms and disease patterns; A medical history credibility evaluation module that evaluates the overall credibility of the medical history through multidimensional indicators; A doctor-patient communication optimization module that analyzes the information transmission efficiency of doctor-patient dialogue in real time and prompts doctors to adjust communication strategies.

8. The method for reconstructing medical history based on symptom evolution map and counterfactual reasoning, applied to the system for reconstructing medical history based on symptom evolution map and counterfactual reasoning according to any one of claims 1-7, characterized in that, The steps include: S1: Receive the patient's initial complaint, extract key symptoms and time anchors; S2: Construct a preliminary symptom spatiotemporal evolution graph, label the anatomical location and time sequence of symptoms; S3: Based on the symptom transmission rule, infer the possible hidden symptoms and generate targeted evocative test schemes; S4: Perform symptom evocative testing, monitor and record patient responses; S5: Integrate all symptom information and use the minimum description length principle to reconstruct the complete medical history timeline; S6: Verify key symptoms through counterfactual reasoning and evaluate the reliability of the diagnosis; S7: Convert the patient's colloquial description into a standardized medical history record; S8: Evaluate the completeness of the medical history, identify potential missing information and conduct supplementary inquiries.

9. The method of claim 8, wherein the method further comprises: The medical history reconstruction in step S5 specifically includes: Representing symptom fragments as nodes of a time-constrained graph; Defining time sequence relationship constraints between fragments; Using constraint optimization algorithms to solve the optimal timeline that satisfies all constraints; Based on medical knowledge, verify and correct unreasonable time sequence relationships.

10. The method of claim 8, wherein the method further comprises: The counterfactual reasoning in step S6 includes: Construct a diagnostic model of the complete symptom set; Remove symptoms one by one and recalculate the diagnosis probability; Identify key symptoms that cause significant changes in diagnosis probability; Verify the accuracy of key symptoms through follow-up questions.

Images