Dynamic curative effect feedback traditional Chinese medicine sequential diagnosis and treatment logic mining method, system and medium

Abstract

The invention discloses a traditional Chinese medicine sequential diagnosis and treatment logic mining method and system based on dynamic curative effect feedback, and a medium, and the method comprises the steps: preprocessing traditional Chinese medicine sequential diagnosis and treatment data, constructing a diagnosis-time pair sequence containing multiple times of treatment of a patient, each diagnosis-time pair comprising a treatment state and curative effect feedback; based on the theory of traditional Chinese medicine, the symptoms and the prescriptions are mapped into structured vectors respectively, and a state conversion tuple is generated; calculating a dynamic evolution mode of a composite state of the state transition tuple through a state transition probability formula in combination with the curative effect feedback information; and extracting a composite state vector mapping rule by adopting a sequential rule mining algorithm, and generating a rule conforming to traditional Chinese medicine syndrome differentiation thinking in combination with the interpretable decision tree model. According to the method, a traditional Chinese medicine syndrome differentiation and treatment dynamic adjustment rule hidden in a medical record is mined, and visualization and regularization of diagnosis and treatment logic are achieved.

Description

Technical Field

[0001] The present invention relates to the field of medical data analysis technology, and in particular to a method, system, and medium for mining traditional Chinese medicine sequential diagnosis and treatment logic with dynamic efficacy feedback. Background Art

[0002] The core of TCM diagnosis and treatment lies in dynamic syndrome differentiation and treatment. In clinical practice, treatment principles and methods must be continuously adjusted based on the patient's condition evolution and efficacy feedback to form a sequential diagnosis and treatment logic. However, existing technologies face the following bottlenecks when mining and analyzing TCM time-series medical record data:

[0003] Traditional electronic health record (EHR) analysis methods often focus on the static associations of individual visits, making it difficult to capture the dynamic evolution of the "symptom-prescription" state between visits. For example, the Apriori algorithm can detect the co-occurrence patterns of symptoms and medications, but it cannot reveal the dynamic pathological transformation pathways such as "Qi deficiency and blood stasis → phlegm-dampness internal obstruction → Qi and Yin deficiency." While sequential pattern mining techniques can process time series data, they lack the ability to model complex states and cannot effectively integrate the driving effect of efficacy feedback on state transitions.

[0004] Existing research often treats efficacy as a general label, rather than a core conditional variable that triggers syndrome differentiation and adjustment. For example, time series models (such as LSTM and HMM) can predict disease progression, but their "black box" nature prevents explicit association with clinical decision-making logic, such as "ineffective treatment, then change prescription." This makes it difficult to quantitatively model the dynamic thinking of Traditional Chinese Medicine (TCM), which states that "if the treatment is effective, do not change the prescription; if the treatment is ineffective, change the prescription."

[0005] While deep learning models can handle complex temporal relationships, the rules they extract lack consistency with Traditional Chinese Medicine (TCM) theory. For example, a model might output the association "symptom A → medication B" but fail to explain its inherent connection to treatment principles such as "soothing the liver and relieving depression → regulating qi and activating blood circulation," making it difficult for clinicians to trust and apply the model. Summary of the Invention

[0006] Based on the above deficiencies of the existing technology, the present invention provides a method, system, and medium for mining the sequential diagnosis and treatment logic of traditional Chinese medicine with dynamic efficacy feedback, so as to realize the visualization, regularization, and clinical practicality of the diagnosis and treatment logic.

[0007] To solve the above technical problems, the first aspect of the present invention discloses a method for mining TCM sequential diagnosis and treatment logic with dynamic efficacy feedback, comprising:

[0008] Preprocess TCM time series diagnosis and treatment data to construct a sequence of consultation pairs containing multiple visits of the patient. Each consultation pair includes the consultation status and efficacy feedback.

[0009] Based on TCM theory, symptoms and prescriptions are mapped into structured vectors to generate state transition tuples.

[0010] Combined with efficacy feedback information, the dynamic evolution pattern of the composite state of the state transition tuple is calculated through the state transition probability formula; the sequential rule mining algorithm is used to extract the rules of composite state vector mapping, and combined with the interpretable decision tree model to generate rules that conform to the dialectical thinking of traditional Chinese medicine.

[0011] In some embodiments, the dynamic evolution pattern of the composite state is calculated using a state transition probability formula in combination with therapeutic effect feedback information, including:

[0012] Based on the diagnosis pair sequence, the state transition probability under different efficacy feedback is calculated to generate the state transition probability matrix;

[0013] The rules of previous state-efficacy-next state are extracted through the frequent state sequence mining algorithm, and strong association rules that meet the support and confidence thresholds are screened;

[0014] An interpretable decision tree model was constructed to predict subsequent prescription strategies based on efficacy feedback and previous status, and to extract rules that conform to TCM dialectical thinking.

[0015] In some embodiments, the state transition probability under different therapeutic effect feedback is calculated by the following formula:

[0016] P(State{i+1}|State i,Efficacy{i+1})=Count(State i,Efficacy{i+1},State{i+1}) / Sum{all S'}Count(State i,Efficacy{i+1},S')

[0017] Where P is the state transition probability, State i is the i-th state, Efficacy{i+1} is the i+1-th structured evaluation, and S' is the set of all possible next states under the current state State i and efficacy{i+1}.

[0018] In some embodiments, an interpretable decision tree model is constructed to predict subsequent prescription strategies based on efficacy feedback and previous status, and extract the following information in accordance with the TCM dialectical thinking rules:

[0019] Extract the defined features and target variables from the diagnosis-to-transformation dataset T and construct a feature matrix;

[0020] Use the C4.5 algorithm to calculate information gain and split information, use the information gain rate as the splitting criterion, and prioritize nodes with high gain rates as decision nodes to build a decision tree;

[0021] The pessimistic pruning algorithm is used to prune overfitting branches and retain the decision path that conforms to clinical experience;

[0022] Based on the decision tree model, rules that conform to traditional Chinese medicine dialectical thinking are extracted.

[0023] In some embodiments, overfitting branches are pruned using a pessimistic pruning algorithm to retain decision paths consistent with clinical experience, including:

[0024] Calculate the pessimistic error estimate of the subtree, the adjusted error number of the subtree, and the error number of leaf nodes after pruning;

[0025] Compare the adjusted error count of the subtree and the error count of the leaf node after pruning. If the error count of the leaf node after pruning is not greater than the adjusted error count of the subtree, perform the pruning operation.

[0026] In some embodiments, extracting rules that conform to TCM dialectical thinking includes:

[0027] Determine a therapeutic effect split point in the decision tree, and analyze a first branch and a second branch corresponding to the therapeutic effect split point; the first branch is a branch with good therapeutic effect, and the second branch is a branch with poor therapeutic effect;

[0028] Compare the differences between the first and second branches, count the proportion of prescription-maintaining rules in the first branch and the proportion of prescription-changing rules in the second branch, and quantify the driving weight of efficacy on syndrome differentiation decision-making;

[0029] Through pessimistic pruning, branches with support lower than the preset threshold are removed to extract the IF-THEN rule set that conforms to the dialectical thinking of traditional Chinese medicine.

[0030] In some embodiments, extracting rules for the previous state-efficacy-next state through frequent state sequence mining and screening strong association rules that meet support and confidence thresholds include:

[0031] Count the frequency of each state in all state sequences from the state sequence database D, calculate the support of a single state, filter out states with support greater than the minimum support, and generate a frequent 1-sequence set L1;

[0032] Iteratively generate candidate k-sequence set Lk, remove infrequent subsequences through pruning operation, calculate the support of candidate sequences, filter candidate sequences with support greater than the minimum support, and form frequent k-sequence set Lk;

[0033] Based on the frequent k-sequence set Lk and the state sequence database D, the rule support and confidence are calculated, and strong association rules are generated in combination with efficacy feedback.

[0034] In some embodiments, further comprising:

[0035] Visualization rules are generated with composite states as nodes and therapeutic effect feedback as edge weights; the visualization rules include state transition diagrams or flow charts; the composite states include symptom patterns and prescription strategies.

[0036] Secondly, a TCM sequential diagnosis and treatment logic mining system with dynamic efficacy feedback was disclosed, including:

[0037] The preprocessing module preprocesses TCM time-series diagnosis and treatment data to construct a sequence of consultation pairs containing multiple visits of the patient. Each consultation pair includes the consultation status and efficacy feedback.

[0038] The state transition tuple generation module maps symptoms and prescriptions into structured vectors based on TCM theory and generates state transition tuples.

[0039] The rule extraction module, combined with efficacy feedback information, calculates the dynamic evolution pattern of the composite state of the state transition tuple through the state transition probability formula; uses a sequential rule mining algorithm to extract the rules of composite state vector mapping, and combines it with an interpretable decision tree model to generate rules that conform to the dialectical thinking of traditional Chinese medicine.

[0040] In a third aspect, a computer storage medium is disclosed, on which a computer program is stored. When the computer program is executed by a processor, the method for mining the logic of sequential diagnosis and treatment of traditional Chinese medicine with dynamic efficacy feedback as described in any one of the above items is implemented.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] This invention deeply integrates dynamic efficacy feedback with the sequential diagnosis and treatment logic of Traditional Chinese Medicine (TCM). Based on structured vector mapping and state transition probability calculation, it transforms scattered diagnosis and treatment records into a dynamically evolving "symptom-prescription" knowledge network. This accurately quantifies the driving weight of efficacy feedback on syndrome differentiation decision-making, breaking through the limitations of static associations in traditional methods. Through sequential rule mining and interpretable decision tree models, it extracts the high-confidence rule that "if the effect is not good, do not change the prescription; if the effect is poor, change it" that conforms to TCM thinking. This constructs a decision path library for the multi-dimensional linkage of "efficacy-symptom-prescription", providing clinicians with evidence-based real-time decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of the TCM sequential diagnosis and treatment logic mining method with dynamic efficacy feedback provided by the present invention;

[0044] Figure 2 This is a flow chart of step S3 of the TCM sequential diagnosis and treatment logic mining method with dynamic efficacy feedback provided by the present invention;

[0045] Figure 3 This is a flow chart of step S32 of the TCM sequential diagnosis and treatment logic mining method with dynamic efficacy feedback provided by the present invention;

[0046] Figure 4 This is a flow chart of step S33 of the TCM sequential diagnosis and treatment logic mining method with dynamic efficacy feedback provided by the present invention;

[0047] Figure 5 This is a flow chart of step S333 of the TCM sequential diagnosis and treatment logic mining method with dynamic efficacy feedback provided by the present invention;

[0048] Figure 6 This is a schematic diagram of the decision tree of the Traditional Chinese Medicine sequential diagnosis and treatment logic mining method with dynamic efficacy feedback provided by the present invention. DETAILED DESCRIPTION

[0049] For better understanding and implementation, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0050] The terms "including" and "having" and any variations thereof in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus that includes a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products or apparatus.

[0051] The embodiment of the present invention discloses a method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback, which extracts dynamic adjustment rules that are easy to understand and conform to traditional Chinese medicine thinking.

[0052] like Figure 1 As shown, the method includes:

[0053] Step S1: pre-processing TCM time series diagnosis and treatment data to construct a sequence of diagnosis pairs containing multiple visits of the patient, each diagnosis pair including the status of two adjacent visits, therapeutic effect feedback, and the status of the subsequent visit;

[0054] The TCM time series diagnosis and treatment data are read from the database, wherein the TCM time series diagnosis and treatment data include EMPI (Enterprise Master Patient Index, a unique value of the patient number), consultation date, prescription, symptoms and time series medical record data of the last consultation efficacy evaluation.

[0055] The TCM time series diagnosis and treatment data are preprocessed, including serialization and normalization. The serialization process is to group according to EMPI and sort each group in ascending order according to the date of visit. For each patient, an ordered visit sequence Vp = {Visit 1, Visit 2, ..., Visit k} is generated, where p is the patient and k is the total number of visits for the patient.

[0056] Normalization involves standardizing TCM names, symptom descriptions, and efficacy evaluations. Using a standard TCM dictionary, aliases and variant spellings are standardized, and dosage information can be selectively extracted or standardized. Using a standard TCM symptom terminology set, synonyms and negations are handled. Efficacy evaluations are mapped to predefined ordinal categorical variables, set to: {"significantly effective": 3, "effective": 2, "ineffective": 1, "exacerbated": 0}.

[0057] Based on the pre-processed TCM time series diagnosis and treatment data, a diagnosis pair sequence is constructed. For each patient p's diagnosis sequence Vp, if k>1, k-1 diagnosis pairs Pair i=(Visit{i},Visit{i+1}) are constructed, where i ranges from 1 to k-1. Each diagnosis pair contains:

[0058] EMPI p

[0059] Symptoms Raw i, the raw symptom list of visit i;

[0060] Prescription Raw i, the list of original prescriptions for visit i;

[0061] Efficacy{i+1}, a structured evaluation of the efficacy of the i-th prescription recorded at the i+1-th visit;

[0062] Symptoms Raw{i+1}, the raw symptom list of visit i+1;

[0063] Prescription Raw{i+1}, the original prescription list for the i+1th visit.

[0064] The diagnosis pair sequence is stored in an intermediate database or data structure.

[0065] Step S2: Based on TCM theory, symptoms and prescriptions are mapped into structured vectors to generate state transition tuples.

[0066] Based on rule mapping, a predefined "symptom combination-syndrome" rule library is used for matching. The original symptom list Symptoms Raw of each diagnosis is converted into a structured symptom pattern representation SymtomPattern. The symptom pattern representation of the i-th diagnosis SymtomPattern i and the symptom pattern representation of the i+1th diagnosis SymtomPattern{i+1} are output.

[0067] Based on the attributes of Chinese Pharmacopoeia TCM prescriptions, the four properties, five flavors, and meridians of each herb are mapped. Each original prescription list, Prescription Raw, is converted into a structured TCM prescription representation, PrescriptionStrategy. The output is PrescriptionStrategy i for the i-th original prescription list and PrescriptionStrategy{i+1} for the i+1th original prescription list.

[0068] Define the state of diagnosis i as State i = (SymptomPattern i, PrescriptionStrategy i).

[0069] Generate a state transition tuple Transition i = (State i, Efficacy{i+1}, State{i+1}) for each diagnosis pair Pair i.

[0070] Step S3: Combined with the therapeutic effect feedback information, the dynamic evolution pattern of the composite state is calculated using the state transition probability formula. Figure 2 As shown, the following steps are included:

[0071] Step S31: Calculate the composite state transition probability under different therapeutic effect feedbacks based on the diagnosis pair sequence and generate a state transition probability matrix;

[0072] The state transition probability under different efficacy evaluation conditions is calculated based on the state transition tuple Transition i. The composite state transition probability under different efficacy feedback is calculated using the following formula:

[0073] P(State{i+1}|State i,Efficacy{i+1})=Count(State i,Efficacy{i+1},State{i+1}) / Sum{all S'}Count(State i,Efficacy{i+1},S')

[0074] Where P is the state transition probability, State i is the i-th state, Efficacy{i+1} is the i+1-th structured evaluation, and S' is the set of all possible next states under the current state State i and efficacy{i+1}.

[0075] Based on the state transition probabilities, a state transition probability matrix is ​​constructed. The nodes of the state transition probability matrix are states, the edges are transitions, and the edge weights are expressed as state transition probabilities, which are displayed in layers by efficacy.

[0076] Step S32: extract the rules of previous state-efficacy-next state through the frequent state sequence mining algorithm, and select the strong association rules that meet the support and confidence thresholds. Figure 3 As shown, specifically including:

[0077] Step S321: Count the frequency of each state in all state sequences from the state sequence database D, calculate the support of a single state, select states with support greater than the minimum support, and generate a frequent 1-sequence set L1;

[0078] Step S322, iteratively generate a candidate k-sequence set Lk, remove infrequent subsequences through pruning operations, calculate the support of candidate sequences, and screen candidate sequences with support greater than the minimum support to form a frequent k-sequence set Lk;

[0079] Step S323: Based on the frequent k-sequence set Lk and the state sequence database D, the rule support and confidence are calculated, and strong association rules are generated in combination with the efficacy feedback.

[0080] Specifically, each patient p corresponds to a state sequence Sp, which is composed of states in chronological order, Sp =<State{p,1},State{p,2},...,State{p,k}> Wherein, the state State{p,i}=(SymptomPattern{p,i},PrescriptionStrategy{p,i}). A state sequence database D is constructed based on the state sequences Sp of multiple patients. A diagnosis-pair transition dataset T is constructed. The diagnosis-pair transition dataset T is composed of the set of state transition tuples Transition i in step S2 and is used for the subsequent calculation of the support and confidence of the rule.

[0081] A frequent state sequence mining algorithm is used to mine the state sequence database and the diagnosis pair conversion data set, identify frequent state sequences from the state sequence, use efficacy feedback as a conditional variable, extract the rules of previous state-efficacy-next state, and screen strong association rules that meet the support and confidence thresholds.

[0082] First, count the frequency of each single state State X across all state sequences from the state sequence database D. For example, the state sequence of patient A might be "Qi deficiency and blood stasis → Phlegm-dampness internal obstruction → Qi and Yin deficiency." Scan the state sequences of all patients and count the independent occurrence frequencies of individual states (such as "Qi deficiency and blood stasis"). Select states whose support exceeds a preset threshold to form a frequent 1-sequence set.

[0083] Calculate its support:

[0084] Support(State X)=Count(Sequences containing State X) / Total number of sequences in D

[0085] Here, Count(Sequences containing State X) refers to the number of patient state sequences that contain at least one state State X.

[0086] All the states State X that satisfy Support(State X)>=min sup are grouped into a frequent sequence set L1 of length 1. L1={<State X> |Support(State X)>=min sup}

[0087] Next, a candidate 2-sequence is generated by a join operation. For example, the frequent 1-sequences "Qi deficiency and blood stasis" and "phlegm dampness internal resistance" are combined into a candidate sequence "Qi deficiency and blood stasis → phlegm dampness internal resistance". Based on L{k-1}, a candidate k-sequence set Ck is generated. To generate C2, any two frequent states in the frequent sequence set L1 are taken.<State A> and<State B> , generate all possible 2-sequence candidates<State A,State B> For k>2, connecting State 1 and State 2∈L{k-1} requires that State 1 after removing the first element is equal to State 2 after removing the last element, that is, removing duplicate elements for connection operation. For example, if<A,B> and<B,C> In L2, the candidate is generated<A,B,C> .

[0088] These candidate sequences are pruned to remove those containing infrequent subsequences. If "phlegm dampness internal obstruction → qi and yin deficiency" is not frequent, its parent sequence is removed. For each generated candidate k-sequence c∈Ck, check whether all its (k-1)-length subsequences are in L{k-1}. If there is any subsequence that is not in L{k-1}, remove the candidate c from Ck. For example, for candidate<A,B,C> , need to check<A,B> and<B,C> Are they all in the candidate sequence L2?

[0089] Rescan the state sequence database D, traverse the state sequences Sp of all patients, calculate the support of candidate sequences, and retain the sequences that meet the threshold as frequent k-sequences. The support is calculated using the following formula:

[0090] Support(c)=Count(Sequences containing c as a subsequence) / Totalnumber of sequences in D

[0091] Among them, Sequences containing c as a subsequence is the state sequence containing c as a subsequence, and Total number of sequences in D is the total number of sequences in the state sequence database D.

[0092] This process is repeated until no higher-order frequent sequences can be generated. The principle is based on the "prior property" of the Apriori algorithm - if a sequence is not frequent, its supersequence must be infrequent, thus greatly reducing the generation of invalid candidates. Each element in L2 is of the form<State A,State B> , indicating that State A followed by State B is a relatively frequent pattern. Hierarchical screening significantly reduces computational complexity while retaining statistically significant high-frequency patterns, providing a reliable foundation for subsequent rule extraction. For example, if the support for "liver depression and qi stagnation → qi stagnation and blood stasis" reaches 3%, it indicates that this pattern recurs clinically and may reflect a common pathological mechanism.

[0093] Since the rule form is State A->[Efficacy]->State B, the most important thing is the frequent 2-sequence set L2. Each element in the sequence set L2 is of the form<State A,State B> , indicating that State A followed by State B is a relatively frequent pattern.

[0094] Based on the frequent k-sequence set Lk and the state sequence database D, the rule support and confidence are calculated, and the strong association rules are generated by combining the efficacy feedback. Traverse the sequence set L2, for each frequent 2-sequence in the sequence set L2<State A,StateB> , obtain the efficacy value Efficacy X for that state, where 3 = markedly effective, 2 = effective, 1 = ineffective, and 0 = worsening. In the diagnosis pair transformation dataset T, find all tuples of the form (State A, Efficacy X, State B). This represents the numerator of the rule State A->[Efficacy = X]->State B. Also, find all tuples of the form (State A, Efficacy X, Any State) in the diagnosis pair transformation dataset T. This represents the total number of times State A appears and the subsequent efficacy evaluation is X. This serves as the denominator for calculating the confidence.

[0095] The rule support is calculated using the following formula:

[0096] Support(Rule:State A->[Efficacy=X]->State B)=Count(State A,EfficacyX,State B) / Total number of transitions in T

[0097] Total number of transitions in T is the total number of tuples in the diagnosis pair transition dataset T.

[0098] The confidence of the rule is calculated using the following formula:

[0099] Confidence(Rule:StateA->[Efficacy=X]->State B)=Count(State A,Efficacy X,State B) / Count(State A,Efficacy X)

[0100] If Count(State A, Efficacy X) is 0, the confidence of the rule is undefined or 0.

[0101] All rules that meet both the minimum rule support and the minimum confidence are retained to generate strong association rules. By formulating conditional probabilities, the driving effect of efficacy on state transition is quantified. For example:

[0102] The rule “liver depression and qi stagnation → [efficacy = ineffective] → qi stagnation and blood stasis” has a support rate of 1.2% and a confidence level of 85%, indicating that this adjustment strategy has high credibility when the efficacy is ineffective.

[0103] The confidence level of the rule "tonifying Qi and activating blood circulation → [efficacy = marked effect] → tonifying Qi and activating blood circulation (maintaining the prescription)" reached 90%, which directly verified the clinical experience of traditional Chinese medicine that "if the prescription is effective, do not change it".

[0104] Step S33: construct an interpretable decision tree model, use efficacy feedback and previous status as features to predict subsequent prescription strategies, and extract the rules of TCM dialectical thinking.

[0105] Step S331: extract the defined features and target variables from the state transition tuple Transition i and construct a feature matrix; use the C4.5 algorithm, with the information gain rate as the splitting criterion, and preferentially select nodes with high information gain rates as decision nodes;

[0106] From the above state transition tuples, extract the features and target variables defined above for each tuple, construct an MxN feature matrix X (M transition instances, N-1 features) and an Mx1 target vector Y. Randomly split the dataset F into a training set Ftrain (70%) and a test set Ftest (30%), train on Ftrain, and evaluate on Ftest.

[0107] The selected C4.5 algorithm library is configured and used, and a training set F train is input into the C4.5 algorithm, where the training set F includes features X train and targets Y train.

[0108] Calculate the information entropy Entropy(S) of the current node dataset S relative to the target variable PrescriptionStrategy{i+1}.

[0109] Entropy(S)=-Σ[P(cj)*log2(P(cj))]

[0110] Where P(cj) is the proportion of category cj (a certain PrescriptionStrategy{i+1}) in the dataset S.

[0111] For each candidate feature A, such as SymptomPattern i, Efficacy{i+1}, etc.), calculate its information gain Gain(S,A) and split information SplitInfo(S,A).

[0112] Gain(S,A)=Entropy(S)-Σ[(|Sv| / |S|)*Entropy(Sv)]

[0113] Where Sv is the subset of S where feature A takes value v.

[0114] SplitInfo(S,A)=-Σ[(|Sv| / |S|)*log2(|Sv| / |S|)]

[0115] That is, for all possible values ​​v of feature A.

[0116] Calculate the gain ratio GainRatio(S,A) for each feature A:

[0117] GainRatio(S,A)=Gain(S,A) / SplitInfo(S,A)

[0118] If SplitInfo(S,A) is 0, the gain rate is also 0 or special processing is used.

[0119] The feature A* with the highest gain ratio is selected as the splitting criterion for the current node. Based on different values ​​of A*, the dataset S is divided into subsets Sv. For each non-pure subset Sv, that is, a subset containing multiple target categories, the C4.5 algorithm is recursively called to construct a subtree. All samples in the node belong to the same category or there are no remaining features for splitting.

[0120] Step S332: Prune overfitting branches using a pessimistic pruning algorithm to retain decision paths that are consistent with clinical experience.

[0121] Step S3321, calculating the pessimistic error estimate of the subtree, the adjusted error number of the subtree, and the error number of the leaf node after pruning;

[0122] Step S3322: Compare the adjusted error count of the subtree and the error count of the leaf node after pruning. If the error count of the leaf node after pruning is not greater than the adjusted error count of the subtree, perform the pruning operation.

[0123] Starting from the bottom of the tree, visit each non-leaf node (Internal Node) upwards. For each non-leaf node T, consider pruning it, that is, replacing the entire subtree with T as the root with a leaf node.

[0124] Calculate the pessimistic error estimate for a subtree. Let Subtree(T) denote the entire subtree rooted at node T. Traverse all leaf nodes in Subtree(T). For each leaf node L in Subtree(T): N(L): The total number of training examples that reach this leaf node. E(L): The number of training examples that are misclassified at this leaf node.

[0125] Calculate the Pessimistic Error Count for Subtree, which includes adjusting the error count using continuity correction, adding an error of 0.5 to each leaf node to compensate for the uncertainty in estimating the generalization error from the training data.

[0126] Adjusted Errors(Subtree(T))=Σ[E(L)+0.5], summed over all leaf nodes L in Subtree(T).

[0127] Calculate the pessimistic error estimate after pruning node T: If node T is replaced by a leaf node, the predicted category of the leaf node will be the category with the largest number of training samples reaching node T (Majority Class). Let N(T) be the total number of training samples reaching node T,

[0128] Let E'(T) be the number of training examples that would be misclassified at node T if it were to become a leaf node predicting its majority class. Apply continuity correction to calculate the adjusted error count after pruning node T (Pessimistic ErrorCount if Pruned):

[0129] Adjusted Errors(T as Leaf)=E'(T)+0.5

[0130] Compare the pessimistic error count for retaining the subtree, Adjusted Errors(Subtree(T)), with the pessimistic error count for pruning node T to a leaf, Adjusted Errors(T as Leaf). Compare the subtree's adjusted error count with the leaf node's error count after pruning. If the leaf node's error count after pruning is no greater than the subtree's adjusted error count, perform the pruning operation. Otherwise, do not prune and retain node T and its subtree.

[0131] Step S333: Extraction rules based on the decision tree model, such as Figure 5 Shown, including:

[0132] Step S3331: Determine the efficacy split point in the decision tree, and analyze the first branch and the second branch corresponding to the efficacy split point; the first branch is the branch with good efficacy, and the second branch is the branch with poor efficacy;

[0133] Step S3332: Compare the differences between the first branch and the second branch, count the proportion of prescription-maintaining rules in the first branch and the proportion of prescription-changing rules in the second branch, and quantify the driving weight of efficacy on syndrome differentiation decision-making;

[0134] Step S3333: Remove branches whose support is lower than a preset threshold through pessimistic pruning to extract the IF-THEN rule set that conforms to the dialectical thinking of traditional Chinese medicine.

[0135] Specifically, starting from the root node of the pruned decision tree, we traverse the path to each leaf node. Each path corresponds to a rule. Each internal node on the path represents an AND clause in the IF condition, including the feature + value / range. The leaf node represents the predicted result of the THEN part, PrescriptionStrategy{i+1}.

[0136] For example, if PrescriptionStrategy i = "tonifying Qi and strengthening the spleen" AND Efficacy{i+1} = "ineffective" AND SymptomPattern{i+1} = "internal accumulation of dampness and heat" THEN PrescriptionStrategy{i+1} = "clearing heat, removing dampness and strengthening the spleen" IF PrescriptionStrategy i = "promoting blood circulation and removing blood stasis" AND Efficacy{i+1} = "significantly effective" THEN PrescriptionStrategy{i+1} = "promoting blood circulation and removing blood stasis" (prescription-defining logic)

[0137] When observing a decision tree, focus on features (SymptomPattern i, PrescriptionStrategy i, Efficacy{i+1}, SymptomPattern{i+1}) that appear near the root of the tree. Identify patterns that represent common syndrome differentiation and adjustment strategies, such as maintaining the prescription, adding flavors, changing the prescription, and targeting concurrent syndromes.

[0138] For example, for a node that splits on Efficacy_{i+1}, we need to clarify the circumstances under which efficacy is considered. We need to trace the path from the tree root all the way back to the efficacy split node. The split conditions of all other nodes along the path collectively define the "context" or "preconditions" of this node.

[0139] For example, if the path is IF PrescriptionStrategy_i == "Soothing Liver and Resolving Depression" AND SymptomPattern_i == "Liver Depression and Qi Stagnation," and a split occurs at Efficacy_{i+1}, this means: for a patient previously treated with the "Soothing Liver and Resolving Depression" strategy for "Liver Depression and Qi Stagnation," how will subsequent decisions differ based on the efficacy feedback? The path determines the context of this efficacy.

[0140] Continue down the branch that represents "good efficacy" (e.g., Efficacy{i+1}>1.5 or Efficacy{i+1}== "significantly effective" or Efficacy{i+1}== "effective"). Observe whether this branch splits further. If so, what features (e.g., Symptom Pattern{i+1} or a subdivision of Prescription Strategy i) determine the split? What Prescription Strategy{i+1} is predicted by the final leaf node? Record the subsequent prescription strategies recommended by the model for the "good efficacy" path.

[0141] Similarly, continue down the branch that represents "poor efficacy," for example, Efficacy{i+1} <= 1.5, Efficacy{i+1} == "ineffective," or Efficacy{i+1} == "exacerbated." Observe the subsequent splits in this branch to see if the features used differ from those in the "good efficacy" branch. The final leaf node predicted by the structured TCM prescription, Prescription Strategy{i+1}, is what. Record the subsequent prescription strategies recommended by the model for this "poor efficacy" path.

[0142] like Figure 6 As shown in the figure, when the previous prescription strategy (PrescriptionStrategyi) is known to be "Yi Qi Huo Xue", the model determines the next prescription strategy (PrescriptionStrategyi+1) based on the efficacy feedback (Efficacyi+1) and the current symptom pattern (SymptomPatterni+1). <div class＝"node">Represents a decision node, which contains a judgment condition. A box with a green background <div class＝"leaf">represents a leaf node, giving the final prediction result, i.e., PrescriptionStrategy_{i+1}. Lines represent decision branches, and the text next to them, such as "Yes (poor efficacy)" and "No (good efficacy)", explains the conditions for selecting that branch. The entire decision tree structure is read from top to bottom and from left to right.

[0143] Find the node in the graph that makes a judgment based on Efficacy_{i+1}. In this example, it's node B. Its condition is Efficacy_{i+1} <= 1.5. This is used to distinguish between "poor efficacy" (ineffective / exacerbated) and "good efficacy" (effective / significantly effective). This split occurs based on the assumption that the "Yinqi Huoxue" strategy was used previously. Analyze the good efficacy branch and the poor efficacy branch, and compare the differences between the two branches.

[0144] Starting from node B in the diagram, choosing "No," which is path BD, once the Efficacy{i+1} condition satisfies "Good Efficacy," the path immediately reaches leaf node D. Leaf node D's prediction is "Yinqi Huoxue." This indicates that when the previous "Yinqi Huoxue" strategy was effective or significantly effective, the model has learned that the best practice is to continue using that strategy (defensive). In this case, the model concludes that no further judgment based on the current symptom pattern, SymptomPattern{i+1}, is necessary.

[0145] Starting from node B in the diagram, selecting "Yes" would lead to paths BCE, FG, or H. After Efficacy{i+1} determines "poor efficacy," this path does not immediately conclude, but instead continues to split. The next split is based on the current symptom pattern, SymptomPattern_{i+1}. Node C determines internal phlegm-dampness obstruction, while node F determines worsening qi stagnation. Depending on the current symptom, different prescription strategies are ultimately chosen: if internal phlegm-dampness obstruction is present, node E predicts strengthening the spleen, resolving phlegm, and activating blood circulation; if worsening qi stagnation is present, node G predicts strengthening qi-regulating and blood-activating measures. Otherwise, node H predicts adjusting the qi-invigorating and blood-activating prescription. This indicates that when the previous "qi-invigorating and blood-activating" strategy is ineffective or worsens, the model no longer maintains a fixed prescription. Instead, it determines that the current specific condition requires a reassessment, meaning that SymptomPattern_{i+1} becomes the key factor in determining the next strategy, and poor efficacy triggers re-differentiation. The final strategy is to adjust or change the prescription based on the current syndrome differentiation results.

[0146] Comparing the two branches, the "good efficacy" path is short and direct, while the "poor efficacy" path is longer and requires more judgment. In the "good efficacy" scenario, the efficacy itself is the primary determinant; in the "poor efficacy" scenario, the efficacy is the trigger, and the current symptoms become the primary determinant of subsequent responses. Those with a "good efficacy" strategy tend to maintain the original strategy, while those with a "poor efficacy" strategy tend to change or adjust it.

[0147] Analyzing this decision tree clearly shows that the efficacy feedback, Efficacy_{i+1}, acts as a "decision logic converter" in the model. Based on the effectiveness of the previous treatment, it directs decisions to different branches: a simple path that favors "maintaining the status quo" (defensive treatment), and a complex path that triggers "reassessment and strategy adjustment." In this complex path, the current symptom pattern, SymptomPattern_{i+1}, takes over from the efficacy and becomes the key to determining the specific adjustment direction. This intuitively embodies the TCM mindset of dynamic diagnosis and treatment adjustments based on efficacy feedback.

[0148] The decision tree model obtained above is evaluated for model performance. The predictive ability and generalization of the pruned C4.5 decision tree model are verified using multi-dimensional indicators. The test set Ftest containing the feature Xtest is input into the trained model, and the predicted result Ypred is output. It is then compared with the true label Ytest to calculate the following indicators:

[0149] 1. Overall accuracy: The ratio of correctly predicted samples to the total number of samples, reflecting the global prediction ability of the model.

[0150] 2. Confusion Matrix: This matrix displays the number of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) for each category, revealing the model's misjudgment patterns for specific categories. For example, "Qi deficiency and blood stasis" could be misjudged as "phlegm-dampness internal obstruction."

[0151] 3. Precision: {Precision} = TP / (TP+FP), which measures the proportion of results predicted by the model as a certain category that are actually correct, avoiding the risk of over-adjusting prescriptions, such as the risk of "activating blood circulation and removing blood stasis" being incorrectly recommended.

[0152] 4. Recall: TP / (TP+FN), which evaluates the model's ability to capture real cases and reduce the risk of missed diagnoses, such as "Qi stagnation and blood stasis" not being identified.

[0153] 5. F1-Score: 2(Precision-Recall) / (Precision+Recall), which comprehensively balances precision and recall. This is particularly suitable for scenarios with an imbalance in TCM syndrome categories, such as robustness assessment when there are few "significantly effective" cases.

[0154] Step S4: Generate a state transition diagram or a flow chart with the composite state as a node and the therapeutic effect feedback as an edge weight.

[0155] Taking composite states, such as the combination of symptom patterns and prescription strategies, as nodes and efficacy feedback as edge weights, a directed graph structure is constructed based on the probability matrix, and a state transition diagram is generated to intuitively display the high-frequency state evolution path.

[0156] Sequential rules or decision tree logic are decomposed into multi-level judgment nodes through flowcharts, such as efficacy branches and syndrome conditions, converting complex rules into interactive graphical paths, such as "ineffective efficacy → differentiation of phlegm and dampness → adding flavors to clear heat". When selecting typical patient cases, their complete status trajectory is traced back, and the differentiation logic of each step is marked to form a visual report of the diagnosis and treatment path. For example, the initial diagnosis is "liver depression and qi stagnation" → the follow-up visit is "qi stagnation and blood stasis", and "ineffective efficacy triggers the strengthening of qi regulation and blood circulation"). The graphical interface lowers the understanding threshold of the dynamic differentiation logic of traditional Chinese medicine, assisting doctors to quickly locate key adjustment nodes; quantified probability and rule confidence improve the credibility of decision-making, and in actual applications, high-probability paths can be recommended first); typical cases verify the consistency between the system and clinical practice, and promote the inheritance of experience and the implementation of intelligent auxiliary diagnosis and treatment.

[0157] The present invention has achieved the quantitative mining and visual presentation of dynamic dialectical logic for the first time by systematically integrating efficacy feedback information with TCM time-series diagnosis and treatment data. The technical solution transforms traditional static and scattered diagnosis and treatment records into a dynamically associated knowledge network, revealing the law of the evolution of the "diagnosis-prescription" state between diagnosis sessions with the efficacy, and extracting the degree of influence of quantitative efficacy on subsequent decision-making through probability calculation and rule extraction. For example, the probability of adhering to the prescription in "significantly effective" cases is 90%, while the probability of changing the prescription in "ineffective" cases exceeds 70%. At the clinical level, the system can provide doctors with a reference for dialectical adjustment of similar cases to reduce decision-making uncertainty; in the field of scientific research, it provides data support for theoretical research on syndrome evolution, prescription compatibility, etc. (such as mining the common transformation path of "damp-heat → phlegm and blood stasis"). In addition, the generated state transition diagram and decision tree flow chart, combined with the full-cycle state evolution trajectory analysis of typical medical cases, intuitively restore the dynamic dialectical thinking of TCM, form a reusable teaching case library, and help the inheritance of clinical experience and the innovation of TCM education model.

[0158] This application also provides a TCM sequential diagnosis and treatment logic mining system with dynamic efficacy feedback, including:

[0159] The preprocessing module preprocesses TCM time-series diagnosis and treatment data to construct a sequence of consultation pairs containing multiple visits of the patient. Each consultation pair includes the consultation status and efficacy feedback.

[0160] The state transition tuple generation module maps symptoms and prescriptions into structured vectors based on TCM theory and generates state transition tuples.

[0161] The rule extraction module, combined with efficacy feedback information, calculates the dynamic evolution pattern of the composite state of the state transition tuple through the state transition probability formula; uses a sequential rule mining algorithm to extract the rules of composite state vector mapping, and combines it with an interpretable decision tree model to generate rules that conform to the dialectical thinking of traditional Chinese medicine.

[0162] This method deeply integrates dynamic efficacy feedback with TCM sequential diagnosis and treatment logic. Based on structured vector mapping and state transition probability calculation, it transforms dispersed diagnosis and treatment records into a dynamically evolving "symptom-prescription" knowledge network. This method accurately quantifies the driving weight of efficacy feedback on syndrome differentiation and decision-making, overcoming the limitations of traditional static correlation methods. Through sequential rule mining and interpretable decision tree models, it extracts the high-confidence rule that "if the effect is not improved, the prescription should be changed; if the effect is poor, the prescription should be changed." This system constructs a decision path library for the multi-dimensional linkage of "efficacy-symptom-prescription," providing clinicians with real-time, evidence-based decision support.

[0163] Based on the same inventive concept, the present invention also provides a computer device, comprising: a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the above-mentioned dynamic efficacy feedback traditional Chinese medicine sequential diagnosis and treatment logic mining method.

[0164] The processing method of the computer device can refer to the description of the above method and will not be repeated here.

[0165] An embodiment of the present application further provides a non-transitory machine-readable storage medium having an executable program stored thereon. When the executable program is executed by a microprocessor, the processor executes the method provided in the above embodiment.

[0166] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the described method.

[0167] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute the described method.

[0168] The embodiments described above are merely illustrative, wherein the modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected based on actual needs to achieve the objectives of the embodiments. Persons of ordinary skill in the art will be able to understand and implement the embodiments without inventive effort.

[0169] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus the necessary general hardware platform, or of course, by means of hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0170] Finally, it should be noted that the embodiments disclosed in the present invention are only preferred embodiments of the present invention, which are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback, characterized by: include: Preprocess TCM time series diagnosis and treatment data to construct a sequence of consultation pairs containing multiple visits of the patient. Each consultation pair includes the consultation status and efficacy feedback. Based on TCM theory, symptoms and prescriptions are mapped into structured vectors to generate state transition tuples. Combined with efficacy feedback information, the dynamic evolution pattern of the composite state of the state transition tuple is calculated through the state transition probability formula; the sequential rule mining algorithm is used to extract the rules of composite state vector mapping, and combined with the interpretable decision tree model to generate rules that conform to the dialectical thinking of traditional Chinese medicine.

2. The method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback according to claim 1 is characterized in that: Combined with the therapeutic effect feedback information, the dynamic evolution pattern of the composite state is calculated through the state transition probability formula, including: Based on the diagnosis pair sequence, the state transition probability under different efficacy feedback is calculated to generate the state transition probability matrix; The rules of previous state-efficacy-next state are extracted through the frequent state sequence mining algorithm, and strong association rules that meet the support and confidence thresholds are screened; An interpretable decision tree model was constructed to predict subsequent prescription strategies based on efficacy feedback and previous status, and to extract rules that conform to TCM dialectical thinking.

3. The method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback according to claim 2 is characterized in that: The state transition probability under different efficacy feedback is calculated by the following formula: P(State{i+1}|State i,Efficacy{i+1})=Count(State i,Efficacy{i+1},State{i+1}) / Sum{all S'}Count(State i,Efficacy{i+1},S') Where P is the state transition probability, State i is the i-th state, Efficacy{i+1} is the i+1-th structured evaluation, and S' is the set of all possible next states under the current state State i and efficacy{i+1}.

4. The method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback according to claim 2 is characterized in that: Construct an interpretable decision tree model to predict subsequent prescription strategies based on efficacy feedback and previous status, and extract the rules of TCM dialectical thinking, including: Extract the defined features and target variables from the diagnosis-to-transformation dataset T and construct a feature matrix; Use the C4.5 algorithm to calculate information gain and split information, use the information gain rate as the splitting criterion, and prioritize nodes with high gain rates as decision nodes to build a decision tree; The pessimistic pruning algorithm is used to prune overfitting branches and retain the decision path that conforms to clinical experience; Based on the decision tree model, rules that conform to traditional Chinese medicine dialectical thinking are extracted.

5. The method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback according to claim 4 is characterized in that: The pessimistic pruning algorithm removes overfitting branches and retains decision paths that are consistent with clinical experience, including: Calculate the pessimistic error estimate of the subtree, the adjusted error number of the subtree, and the error number of leaf nodes after pruning; Compare the adjusted error count of the subtree and the error count of the leaf node after pruning. If the error count of the leaf node after pruning is not greater than the adjusted error count of the subtree, perform the pruning operation.

6. The method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback according to claim 3 is characterized in that: Extract rules that conform to TCM dialectical thinking, including: Determine a therapeutic effect split point in the decision tree, and analyze a first branch and a second branch corresponding to the therapeutic effect split point; the first branch is a branch with good therapeutic effect, and the second branch is a branch with poor therapeutic effect; Compare the differences between the first and second branches, count the proportion of prescription-maintaining rules in the first branch and the proportion of prescription-changing rules in the second branch, and quantify the driving weight of efficacy on syndrome differentiation decision-making; Through pessimistic pruning, branches with support below the preset threshold are removed to extract the IF-THEN rule set that conforms to the dialectical thinking of traditional Chinese medicine.

7. The method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback according to claim 6 is characterized in that: Extract the rules of previous state-efficacy-next state through frequent state sequence mining, and screen strong association rules that meet the support and confidence thresholds, including: Count the frequency of each state in all state sequences from the state sequence database D, calculate the support of a single state, filter out states with support greater than the minimum support, and generate a frequent 1-sequence set L1; Iteratively generate candidate k-sequence set Lk, remove infrequent subsequences through pruning operation, calculate the support of candidate sequences, filter candidate sequences with support greater than the minimum support, and form frequent k-sequence set Lk; Based on the frequent k-sequence set Lk and the state sequence database D, the rule support and confidence are calculated, and strong association rules are generated in combination with efficacy feedback.

8. The method for mining logic of sequential diagnosis and treatment in traditional Chinese medicine with dynamic efficacy feedback according to claim 7 is characterized in that: Also includes: Visualization rules are generated with composite states as nodes and therapeutic effect feedback as edge weights; the visualization rules include state transition diagrams or flow charts; the composite states include symptom patterns and prescription strategies.

9. A TCM sequential diagnosis and treatment logic mining system with dynamic efficacy feedback, characterized by: include: The preprocessing module preprocesses TCM time-series diagnosis and treatment data to construct a sequence of consultation pairs containing multiple visits of the patient. Each consultation pair includes the consultation status and efficacy feedback. The state transition tuple generation module maps symptoms and prescriptions into structured vectors based on TCM theory and generates state transition tuples. The rule extraction module, combined with efficacy feedback information, calculates the dynamic evolution pattern of the composite state of the state transition tuple through the state transition probability formula; uses a sequential rule mining algorithm to extract the rules of composite state vector mapping, and combines it with an interpretable decision tree model to generate rules that conform to the dialectical thinking of traditional Chinese medicine.

10. A computer storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the traditional Chinese medicine sequential diagnosis and treatment logic mining method with dynamic efficacy feedback as claimed in any one of claims 1 to 8 are implemented.

Images