Medical scale quality control method based on AIGC

Abstract

The invention discloses a medical scale quality control method based on AIGC, and relates to the technical field of medical information, and the method comprises the steps: collecting medical data, and constructing a dynamically extended medical knowledge graph; embedding the dynamically expanded medical knowledge graph into a generative AI, and training the generative AI to obtain a medical scale generation model; the medical scale generation model generates an initial medical scale by calling a medical knowledge graph interface; performing consistency analysis on the initial medical scale to generate a secondary medical scale; dynamically optimizing the secondary medical scale by using a reinforcement learning algorithm to generate a tertiary medical scale; and calculating and evaluating the reliability and the structural validity of the three-time medical scale, and generating a final medical scale. According to the method, comprehensive and real-time updated medical information support is provided through the dynamically expanded medical knowledge graph, and it is ensured that a high-quality medical scale generation model is trained through the generative AI.

Description

Technical Field

[0001] The invention relates to the field of medical information technology, and in particular to a medical scale quality control method based on AIGC. Background Art

[0002] With the rapid development of artificial intelligence technology, especially the application of generative artificial intelligence (AIGC) in multiple fields, the medical field has gradually benefited from the innovation and development of this technology. Medical scales, as a commonly used clinical diagnostic tool, are widely used in fields such as mental health and chronic disease management to quantitatively assess patients' health status.

[0003] There are some challenges in the development and optimization process of traditional medical scales. Traditional medical scales usually rely on expert experience and manual editing. Their content is updated slowly and is prone to bias. It is difficult to fully consider the individual differences of patients and the latest medical research results. It is difficult to flexibly respond to the differences between different patient groups. In the face of diverse disease manifestations, it is often impossible to adjust the assessment dimensions and item content in the scale in real time. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a medical scale quality control method based on AIGC to solve the problem of difficulty in flexibly responding to the differences among different patient groups.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a medical scale quality control method based on AIGC, which comprises:

[0008] By collecting medical data, we can build a dynamically expanding medical knowledge graph;

[0009] Embed the dynamically expanded medical knowledge graph into generative AI and train the generative AI to obtain a medical scale generation model;

[0010] The medical scale generation model generates an initial medical scale by calling the medical knowledge graph interface;

[0011] The initial medical scale was analyzed for consistency and a secondary medical scale was generated;

[0012] Using reinforcement learning algorithms to dynamically optimize the secondary medical scale to generate a tertiary medical scale;

[0013] The reliability and construct validity of the three medical scales were calculated and evaluated to generate the final medical scale.

[0014] As a preferred solution of the AIGC-based medical scale quality control method of the present invention, wherein: the dynamically expanded medical knowledge graph is constructed based on medical data, and the specific steps are as follows:

[0015] Disease types, symptoms, and diagnostic process records were collected as medical data through hospital medical records;

[0016] After denoising and redundancy detection of the collected medical data, different texts are converted into a unified format;

[0017] Use natural language processing methods to semantically annotate medical data after denoising and redundancy detection, and extract entities and entity relationships;

[0018] According to the entity relationship structure of the medical knowledge graph, a graph database is used to store the medical knowledge graph;

[0019] When the medical scale is generated, entities and entity relationships that are not covered by the original medical knowledge graph appear. External resources are called through external interfaces to expand the entities and entity relationships.

[0020] As a preferred solution of the AIGC-based medical scale quality control method of the present invention, the dynamically expanded medical knowledge graph is embedded in the generative AI, and the generative AI is trained to obtain a medical scale generation model. The specific steps are as follows:

[0021] Construct entity query, entity relationship query and semantic path query interfaces based on the query language of graph database;

[0022] Based on entity query, entity relationship query and semantic path query interfaces, the medical knowledge graph is embedded in the generative AI to obtain the generative AI embedded in the medical knowledge graph;

[0023] The generative AI embedded in the medical knowledge graph is trained to obtain a medical scale generation model.

[0024] As a preferred solution of the AIGC-based medical scale quality control method of the present invention, the generative AI embedded in the medical knowledge graph is trained to obtain a medical scale generation model, and the specific steps are as follows:

[0025] Extract the items, answer options, and scoring methods from the CGI scale as the medical scale generation logic for generative AI embedded in the medical knowledge graph;

[0026] Extract the semantic relationship between disease type, symptoms and treatment process from the medical knowledge graph as the background of the generation task;

[0027] In the first stage of generative AI training embedded in the medical knowledge graph, diseases, evaluation dimensions, and medical semantic information are input according to the generation task context, and items are output;

[0028] The second stage of generative AI training embedded in the medical knowledge graph outputs the corresponding answer options and scoring methods for the items based on the items output in the first stage and the semantic information of the medical knowledge graph;

[0029] Using the cross entropy loss function, we calculated the gap between the output items and the corresponding answer options of the generative AI embedded in the medical knowledge graph, and the gap between the items and the corresponding answer options in the CGI scale.

[0030] Calculate the n-gram overlap rate of the items and corresponding answer options generated by the output of the generative AI embedded in the medical knowledge graph, and the n-gram overlap rate of the items and the corresponding answer options in the CGI scale, and obtain the generation quality score of the output items and the corresponding answer options of the generative AI embedded in the medical knowledge graph;

[0031] Based on the length of the words and sentences of the items and the corresponding answer options in the CGI scale, the cross entropy loss value threshold and the generation quality score threshold of the output items and the corresponding answer options of the generative AI embedded in the medical knowledge graph are set;

[0032] When the cross-entropy loss values ​​of the output items and the corresponding answer options of multiple consecutive training rounds of the generative AI embedded in the medical knowledge graph are less than the cross-entropy loss value threshold and the quality score is greater than the generation quality score threshold, the training of the generative AI embedded in the medical knowledge graph is completed and the medical scale generation model is obtained.

[0033] As a preferred solution of the AIGC-based medical scale quality control method of the present invention, the medical scale generation model generates an initial medical scale by calling the medical knowledge graph interface. The specific steps are as follows:

[0034] According to the symptoms described by the patient, the medical scale generation model calls the entity query, entity relationship query, and semantic path query interfaces of the medical knowledge graph to obtain the relevant entities, entity relationships, and semantic path information of the patient's disease;

[0035] The query results of the knowledge graph on the relevant entities, entity relationships and semantic path information of the patient's disease are integrated and input into the medical scale generation model to output the initial medical scale about the patient's disease.

[0036] As a preferred solution of the AIGC-based medical scale quality control method of the present invention, the following specific steps are performed to analyze the consistency of the initial medical scale and generate the secondary medical scale:

[0037] Each item of the initial medical scale is processed by sentence segmentation to decompose the entity and entity relationship of the item, and the entity and entity relationship of each item are mapped to the medical knowledge graph to verify whether they exist;

[0038] If it does not exist, it means that the project does not conform to the medical knowledge graph, and the medical scale generation model needs to be retrained to regenerate the medical scale. If it does exist, continue with the subsequent analysis;

[0039] For items in the initial medical scale, the text similarity between two items was calculated to determine whether the two items were semantically redundant;

[0040] Based on the existing question-answer data without semantic duplication in the CGI scale, the text similarity between two items is calculated, and the highest text similarity value is used as the text similarity threshold;

[0041] When the text similarity between two items in the initial medical scale is less than the text similarity threshold, it means that the semantics are not repeated and they are two independent items;

[0042] On the contrary, it means that the two items are repeated, and the item with more semantic path information is retained, and the item with less semantic path information is deleted;

[0043] The initial medical scale after medical knowledge graph verification and semantic duplication judgment is output as a secondary medical scale.

[0044] As a preferred solution of the AIGC-based medical scale quality control method of the present invention, the following specific steps are used to dynamically optimize the secondary medical scale using a reinforcement learning algorithm to generate a tertiary medical scale:

[0045] The response time, completion rate and experience score for each item in the process of patients filling out the secondary medical scale are taken as the state vector;

[0046] Adjusting the number of items, adjusting the order of items, and adjusting the description of items are used as action vectors;

[0047] Set a reward function based on the patient's experience score for each item, answering time, and completion rate;

[0048] The state vector collected in real time is input into the reinforcement learning algorithm, and the secondary medical scale is adjusted according to the action vector through the policy network of the reinforcement learning algorithm;

[0049] After executing the action, an optimized quadratic medical scale is generated, and a reward function of the optimized quadratic medical scale is calculated;

[0050] Update the policy network through the deep Q learning algorithm;

[0051] By continuously optimizing the secondary medical scale in a cycle of state, action, and reward, after each round of optimization, the state vector is updated according to the patient's answer time for each item, completion rate, and experience score for each item;

[0052] According to the change of reward value, a reward value change rate threshold is set. When the reward value change rate in multiple rounds of continuous optimization of the reinforcement learning algorithm is less than the reward value change rate threshold, it means that the reinforcement learning algorithm converges and outputs three medical scales;

[0053] On the contrary, when the reward value change rate is greater than or equal to the reward value change rate threshold in multiple consecutive rounds of optimization, it means that the reinforcement learning algorithm has not converged, and the secondary medical scale continues to be optimized.

[0054] As a preferred solution of the AIGC-based medical scale quality control method of the present invention, the steps of calculating and evaluating the reliability and structural validity of the three medical scales to generate the final medical scale are as follows:

[0055] The response results of the patients filling out the medical scale three times were organized into a response data matrix;

[0056] Based on the response data matrix, the reliability of the three-item medical scale was calculated using Cronbach's Alpha formula;

[0057] Collect the response results of the CGI scale, calculate the reliability of the CGI scale, and set the reliability of the CGI scale as the reliability threshold;

[0058] Based on the response data matrix, the Pearson correlation coefficient between the scores of the corresponding response options of the two items is calculated to obtain a correlation matrix;

[0059] Based on the correlation matrix, the average of all off-diagonal correlation values ​​was calculated to obtain the average correlation value of the three medical scales;

[0060] Based on the response data matrix, the total scores of the corresponding response options for all items used in the medical scale filled out three times by each patient were summed up;

[0061] The Pearson correlation coefficient between the patient's score on each item and the total score was calculated to obtain the overall correlation value of the three-item medical scale;

[0062] Calculate the average correlation value and the overall correlation value of the CGI scale response records as the average correlation threshold and the overall correlation threshold;

[0063] When the reliability, average correlation value and overall correlation value of the three-time medical scale are simultaneously less than the reliability threshold, average correlation threshold and overall correlation threshold, respectively, the reinforcement learning algorithm is used to re-optimize;

[0064] Otherwise, the final medical scale is output.

[0065] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the medical scale quality control method based on AIGC as described in the first aspect of the present invention is implemented.

[0066] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the medical scale quality control method based on AIGC as described in the first aspect of the present invention is implemented.

[0067] The beneficial effects of the present invention are as follows: the present invention provides comprehensive and real-time updated medical information support through a dynamically expanded medical knowledge graph, ensuring that the generative AI trains a high-quality medical scale generation model. After the model generates the initial medical scale, it removes duplicate and inconsistent items through consistency analysis and optimizes it into a secondary medical scale. The reinforcement learning algorithm is used to further adjust the item description and order according to the user experience, improve the filling efficiency and satisfaction, and form a tertiary medical scale. By evaluating the reliability and structural validity of the tertiary scale, the scientificity and reliability of the scale are ensured, thereby outputting the final medical scale. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0069] Figure 1 This is a flow chart of the medical scale quality control method based on AIGC in Example 1.

[0070] Figure 2 This is a flow chart of dynamically optimizing the secondary medical scale using the reinforcement learning algorithm in Example 1. DETAILED DESCRIPTION

[0071] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0072] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0073] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0074] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a medical scale quality control method based on AIGC, comprising the following steps:

[0075] S1. Build a dynamically expanding medical knowledge graph based on medical data.

[0076] The types of diseases, symptoms, and treatment process records (treatment methods, drugs used, etc.) were collected as medical data through hospital medical records;

[0077] After denoising and redundancy detection of the collected medical data, the different texts are converted into a unified format (such as JSON or CSV);

[0078] Specifically, remove non-medical related content, such as the time and location of patient visits;

[0079] Use text similarity algorithms (such as cosine similarity) to detect duplicate and highly similar content, and retain the most semantically complete and informative records;

[0080] Using natural language processing (NLP) methods, the medical data after denoising and redundancy detection are semantically annotated, and entities and entity relationships are extracted;

[0081] Specifically, the domain-adapted BioBERT model is used to extract entities from the text, annotate medical entities such as diseases (such as diabetes), symptoms (such as polydipsia), drugs (such as insulin), and treatment methods (such as blood sugar monitoring), and extract semantic relationships between entities through dependency parsing. For example, the relationship diabetes → need → blood sugar monitoring is extracted from diabetes requiring blood sugar monitoring.

[0082] According to the entity-relationship structure (entity, relationship, entity) of the medical knowledge graph, a graph database (such as Neo4j) is used to store the medical knowledge graph to support fast query and dynamic expansion;

[0083] When the medical scale is generated, entities and entity relationships that are not covered by the original medical knowledge graph appear. External resources (such as hospital medical records) are called through external interfaces to expand the entities and entity relationships.

[0084] S2. Embed the dynamically expanded medical knowledge graph into the generative AI and train the generative AI to obtain a medical scale generation model.

[0085] Build entity query, entity relationship query and semantic path query interfaces based on the query language (Cypher) of graph databases (such as Neo4j);

[0086] Based on entity query, entity relationship query and semantic path query interfaces, the medical knowledge graph is embedded in the generative AI to obtain the generative AI embedded in the medical knowledge graph;

[0087] Specifically, input the name of a disease or symptom in the entity query interface, and return the related entities and their attributes (such as definition and incidence rate, etc.); input an entity in the entity relationship query interface, and return the relationship between the entity and other entities and the target entity; input two entities in the semantic path query interface, and return the multi-hop path between them, which is used for complex semantic derivation, so that generative AI can access entities and entity relationships in the knowledge graph in real time;

[0088] In the input of generative AI, the query results of the medical knowledge graph are added to provide semantic context to ensure that the generated results have medical logic and contextual consistency. During the generation process, the semantic information provided by the medical knowledge graph is used to limit the content generated by the generative AI. For example, it is forbidden to generate items that are not related to the target disease. When generating items and answer options, it is ensured that the answer options conform to the contextual semantics. If entities or entity relationships that do not exist in the medical knowledge graph are detected during the generative AI generation process, new entities and entity relationships are supplemented through real-time expansion of the medical knowledge graph.

[0089] Train the generative AI embedded in the medical knowledge graph to obtain a medical scale generation model;

[0090] Extract the items (referring to the specific items in the CGI scale), answer options and scoring methods from the CGI scale as the medical scale generation logic of the generative AI embedded in the medical knowledge graph;

[0091] For example, the item is “Did you feel depressed in the past week?”, the response options and scores are “never = 0; occasionally = 1; often = 2; always = 3”, and the scoring method is “the total score is 0 to 3 points, and the higher the score, the more serious the depression”;

[0092] Extract the semantic relationship between disease type, symptoms and treatment process from the medical knowledge graph as the background of the generation task;

[0093] For example, the knowledge relationship is diabetes → psychological impact → anxiety;

[0094] In the first stage of generative AI training embedded in the medical knowledge graph, diseases, evaluation dimensions, and medical semantic information are input according to the generation task context, and items are output;

[0095] For example, the input disease is "diabetes", the assessment dimension is "mental health", the medical semantic information is "diabetes → psychological impact → anxiety", and the output item is "In the past week, have you felt anxious because of blood sugar fluctuations?";

[0096] The second stage of generative AI training embedded in the medical knowledge graph outputs the corresponding answer options and scoring methods for the items based on the items output in the first stage and the semantic information of the medical knowledge graph;

[0097] For example, the input item is "Did you feel anxious because of blood sugar fluctuations in the past week?", the semantic information is "diabetes → psychological impact → anxiety", the output response options are "never, score 0; occasionally score 1; often score 2; always score 3", and the scoring method is "the total score is 0 to 3 points, and the higher the score, the more severe the depression";

[0098] The cross entropy loss function is used to calculate the gap between the output items and the corresponding answer options of the generative AI embedded in the medical knowledge graph, and the gap between the items and the corresponding answer options in the CGI scale. The expression is as follows:

[0099] L = ∑(y i ×log(p i ));

[0100] Among them, L is the cross entropy loss value, y i is the probability distribution of items and corresponding response options in the CGI scale, p i Output items of generative AI embedded in medical knowledge graph and probability distribution of corresponding answer options;

[0101] Although the input and output of the generative AI embedded in the medical knowledge graph are both text data, the text data is usually converted into serialized discrete symbols (tokens) inside the generative AI embedded in the medical knowledge graph. For example, the input text is decomposed into words, characters or characters by a tokenizer and mapped to the corresponding index number. The output of the generative AI embedded in the medical knowledge graph is also a serialized token probability distribution, so the project becomes a sequence generation task, and the cross entropy loss function is widely used in classification tasks and sequence generation tasks. Therefore, it is suitable for the gap between the probability distribution of items and corresponding answer options generated by the generative AI embedded in the medical knowledge graph and the probability distribution of items and corresponding answer options in the CGI scale;

[0102] Calculate the n-gram overlap rate of the output items of the generative AI embedded in the medical knowledge graph and the corresponding answer options of the items, and the n-gram overlap rate of the items and the corresponding answer options of the items in the CGI scale, and obtain the quality score of the output items of the generative AI embedded in the medical knowledge graph and the corresponding answer options of the items;

[0103] Specifically, n-gram refers to a phrase consisting of n consecutive words. The overlap rates of n-grams of different lengths (such as 1-gram, 2-gram, and 3-gram) are calculated, and these overlap rates are combined to form the generation quality score of the generative AI embedded in the medical knowledge graph;

[0104] For example, the short sentence "diabetes can easily cause anxiety" in the CGI scale is used as a reference sentence, and the short sentence "diabetes may cause anxiety" output by the generative AI embedded in the medical knowledge graph is used as a candidate sentence, that is, the overlapping parts of the 1-gram are "diabetes" and "anxiety", and the overlap rate is the number of overlapping 1-grams divided by the total number of 1-grams in "diabetes may cause anxiety", and the overlap rate of 1-gram is 0.5. Similarly, the overlap rates of 2-gram and 3-gram are calculated respectively;

[0105] BLEU penalizes candidate sentences that are too short because short sentences are more likely to match reference sentences. The expression is as follows:

[0106]

[0107] Among them, BP is the penalty coefficient, c is the number of words in the candidate sentence, and v is the number of words in the reference sentence;

[0108] Combining the penalty coefficient and the overlap rate of n-grams, the generation quality score of the items and the corresponding answer options in the generative AI embedded in the medical knowledge graph is calculated. The expression is as follows:

[0109] BLEU=BP×(∑ω n ×log(q n ));

[0110] Among them, BLEU is the quality score of the items and the corresponding answer options in the generative AI embedded in the medical knowledge graph, ω n is the weight of the overlap rate of n consecutive words, which is evenly divided for all n-grams. For example, if there are 4 words in the candidate sentence, the weights of 1-gram, 2-gram, 3-gram and 4-gram are all 0.25, and q n is the overlap rate of n consecutive words, where n is the number of consecutive words, such as 3-gram, which means three consecutive words;

[0111] Based on the length of the words and sentences of the items and the corresponding answer options in the CGI scale, the cross entropy loss value threshold and the generation quality score threshold of the output items and the corresponding answer options of the generative AI embedded in the medical knowledge graph are set;

[0112] When the cross entropy loss values ​​of the output items and the corresponding answer options of the generative AI embedded in the medical knowledge graph for multiple (10) consecutive training rounds are less than the cross entropy loss value threshold and the generation quality score is greater than the generation quality score threshold, the training of the generative AI embedded in the medical knowledge graph is completed and the medical scale generation model is obtained. Otherwise, it means that the training continues;

[0113] Among them, the second stage of training can only start after the first stage of training is completed.

[0114] S3. The medical scale generation model generates an initial medical scale by calling the medical knowledge graph interface.

[0115] According to the symptoms described by the patient, the medical scale generation model calls the entity query, entity relationship query, and semantic path query interfaces of the medical knowledge graph to obtain the relevant entities, entity relationships, and semantic path information of the patient's disease;

[0116] For example, the patient's disease is diabetes, and the assessment dimension is mental health. The medical scale generation model calls the entity query interface to query entity information related to the target disease (e.g., diabetes is defined as a metabolic disease, and common symptoms are polydipsia and polyuria, etc.), calls the entity relationship query interface to query the direct relationship between the target disease and other entities (e.g., diabetes requires blood sugar monitoring, and mental health problems may cause anxiety, etc.), and calls the semantic path query interface to query multiple paths between the target disease and the target assessment dimension (e.g., "mental health") to supplement complex semantic derivation (e.g., diabetes → psychological impact → anxiety → accompanying → insomnia) to provide medical semantic context for the medical scale generation model, ensuring that the generated results have medical logic and context consistency;

[0117] The query results of the knowledge graph on the relevant entities, entity relationships and semantic path information of the patient's disease are integrated and input into the medical scale generation model to output the initial medical scale (items, response options and scoring methods) about the patient's disease.

[0118] S4. Perform consistency analysis on the initial medical scale and generate a secondary medical scale.

[0119] Each item of the initial medical scale is processed by sentence segmentation to decompose the entity and entity relationship of the item, and the entity and entity relationship of each item are mapped to the medical knowledge graph to verify whether they exist;

[0120] If it does not exist, it means that the project does not conform to the medical knowledge graph, and the medical scale generation model needs to be retrained to regenerate the medical scale. If it does exist, continue with the subsequent analysis;

[0121] For the items in the initial medical scale, the text similarity (cosine similarity) between two items was calculated to determine whether the two items were semantically repeated;

[0122] Based on the existing items without semantic duplication in the CGI scale, the text similarity between two items is calculated, and the highest text similarity value is used as the text similarity threshold;

[0123] When the text similarity between two items in the initial medical scale is less than the text similarity threshold, it means that the semantics are not repeated and they are two independent items;

[0124] On the contrary, it means that the two items are repeated, and the item with more semantic path information is retained, and the item with less semantic path information is deleted;

[0125] The initial medical scale after medical knowledge graph verification and semantic duplication judgment is output as a secondary medical scale.

[0126] S5. Use reinforcement learning algorithm to dynamically optimize the secondary medical scale and generate a tertiary medical scale.

[0127] The response time, completion rate and experience score of each item in the process of filling out the secondary medical scale (the patient's score of the ease of use and understanding of each item in the medical scale, 1 to 5 points) were used as the state vector;

[0128] For example, a timer is used to record the length of time it takes a patient to fill out each item on a medical questionnaire, the completion rate is calculated based on the number of questions completed by the patient and the total number of questions, and the patient rates their experience with each item after completing the questionnaire;

[0129] Take adjusting the number of items (deleting redundant items or adding supplementary items), adjusting the order of items (reordering items to optimize the filling experience), and adjusting the wording of items (optimizing the language of items to make them easier to understand) as action vectors;

[0130] The reward function is set based on the patient's experience score for each item, answering time and completion rate. The expression is as follows:

[0131]

[0132] Among them, R is the reward value, x is the patient's experience score for each item, b is the patient's answering time for each item, and d is the patient's completion rate;

[0133] The state vector collected in real time is input into the reinforcement learning algorithm, and the secondary medical scale is adjusted according to the action vector through the policy network of the reinforcement learning algorithm;

[0134] Specifically, the ε-greedy strategy is used to select the optimal action in the action vector, a random action is selected with probability ε, and the optimal action is selected with probability 1-ε;

[0135] After executing the action, an optimized quadratic medical scale is generated, and a reward function of the optimized quadratic medical scale is calculated;

[0136] The policy network is updated through the deep Q learning algorithm, and the expression is as follows:

[0137] Q(s,a)=Q(s,a)+α×[r+γ×max(Q(s , ,a , ))-Q(s,a)];

[0138] Among them, Q(s,a) is the Q value of executing action a in state s, α is the learning rate, r is the reward value after executing action a in state s, γ is the discount factor, Q(s , a , ) is the next state s , Next, perform action a , of,

[0139] Q value; s is the state vector, a is the action vector, [R+γ×max(Q(s , a , ))-Q(s,a)] is the TD error,

[0141] Indicates the gap between the current Q value and the target Q value;

[0142] Based on the Q value of the current state s and action a, using the reward value and the next state s , Next, perform action a ,The target Q value is calculated based on the Q value of the TD error, and the current Q value is gradually updated to make it close to the target Q value;

[0143] By continuously optimizing the secondary medical scale in a cycle of state, action, and reward, after each round of optimization, the state vector is updated according to the patient's answer time for each item, completion rate, and experience score for each item;

[0144] The range of reward values ​​was calculated based on the patient's experience rating for each item, answering time, and completion rate when filling out the secondary medical scale;

[0145] According to the range of reward values, the minimum change rate of reward values ​​is calculated, and a reward value change rate threshold is set. When the reward value change rate is less than the reward value change rate threshold in multiple rounds (10 rounds) of optimization of the reinforcement learning algorithm, it indicates that the reinforcement learning algorithm converges and outputs three medical scales;

[0146] On the contrary, when the reward value change rate is greater than or equal to the reward value change rate threshold in multiple consecutive rounds of optimization, it means that the reinforcement learning algorithm has not converged, and the secondary medical scale continues to be optimized.

[0147] S6. Calculate and evaluate the reliability and construct validity of the three medical scales to generate the final medical scale.

[0148] The results of the three medical scales filled in by the patients are organized into a response data matrix, where each row represents the scores of a patient for all the items in the three medical scales corresponding to the response options, and each column represents the scores of different patients for the same item corresponding to the response options, for example

[0149] Based on the response data matrix, the reliability of the three-time medical scale was calculated using the Cronbach's Alpha formula, as shown below:

[0150]

[0151] Where f is the reliability of the three-time medical scale, k is the number of items in the three-time medical scale, and w i is the variance of the score corresponding to the response option of the ith item in the three-dimensional medical scale, and W is the variance of the sum of the scores corresponding to the response options of all items in the three-dimensional medical scale;

[0152] The response results of the CGI scale were collected through hospital medical records, the reliability of the CGI scale was calculated, and the reliability of the CGI scale was set as the reliability threshold;

[0153] Based on the response data matrix, the Pearson correlation coefficient between the scores of the corresponding response options of two items is calculated, and the expression is as follows:

[0154]

[0155] Where m is the Pearson correlation coefficient between the scores of the corresponding answer options of the two items, ranging from [-1,1], and g k and h k are the scores of the kth patient's response options for items g and h, respectively. and are the mean scores of all patients for item g and item h, respectively, and sqrt represents the Take the square root to standardize The role of;

[0156] The Pearson correlation coefficients between all items in the three-dimensional medical scale were summarized into a matrix form to obtain a correlation matrix, e.g.

[0157] Among them, each correlation value in the correlation matrix represents the correlation between two items. According to the example, it can be seen that the correlation value on the diagonal of the correlation matrix is ​​always equal to 1, indicating that the correlation between any item and itself is completely positively correlated. When the correlation value in the correlation matrix is ​​greater than 0, it means that the two items are positively correlated. When the correlation value in the correlation matrix is ​​equal to 0, it means that there is no correlation between the two items. When the correlation value in the correlation matrix is ​​less than 0, it means that the two items are negatively correlated.

[0158] Based on the correlation matrix, the average of all off-diagonal correlation values ​​was calculated to obtain the average correlation value of the three medical scales;

[0159] Based on the response data matrix, the total scores of the corresponding response options for all items used in the medical scale filled out three times by each patient were summed up;

[0160] The Pearson correlation coefficient between the patient's score on each item and the total score was calculated to obtain the overall correlation value of the three-item medical scale;

[0161] Calculate the average correlation value and the overall correlation value of the CGI scale response records as the average correlation threshold and the overall correlation threshold;

[0162] When the reliability, average correlation value and overall correlation value of the three-time medical scale are simultaneously less than the reliability threshold, average correlation threshold and overall correlation threshold, respectively, the reinforcement learning algorithm is used to re-optimize;

[0163] Otherwise, the final medical scale is output.

[0164] This embodiment also provides a computer device, which is applicable to the medical scale quality control method based on AIGC, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to implement the medical scale quality control method based on AIGC proposed in the above embodiment.

[0165] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0166] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the medical scale quality control method based on AIGC as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0167] In summary, the present invention provides comprehensive and real-time updated medical information support through: dynamically expanded medical knowledge graphs, ensuring that generative AI trains a high-quality medical scale generation model. After the model generates the initial medical scale, it removes duplicate and inconsistent items through consistency analysis and optimizes it into a secondary medical scale. The reinforcement learning algorithm is used to further adjust the item description and order according to the user experience to improve the filling efficiency and satisfaction, and form a tertiary medical scale. By evaluating the reliability and structural validity of the tertiary scale, the scientificity and reliability of the scale are ensured, thereby outputting the final medical scale.

[0168] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A medical scale quality control method based on AIGC, characterized by: include, By collecting medical data, we can build a dynamically expanding medical knowledge graph; Embed the dynamically expanded medical knowledge graph into generative AI and train the generative AI to obtain a medical scale generation model; The medical scale generation model generates an initial medical scale by calling the medical knowledge graph interface; The initial medical scale was analyzed for consistency and a secondary medical scale was generated; Using reinforcement learning algorithms to dynamically optimize the secondary medical scale to generate a tertiary medical scale; The reliability and construct validity of the three medical scales were calculated and evaluated to generate the final medical scale.

2. The AIGC-based medical scale quality control method according to claim 1, characterized in that: The specific steps of constructing a dynamically expanded medical knowledge graph based on medical data are as follows: Disease types, symptoms, and diagnostic process records were collected as medical data through hospital medical records; After denoising and redundancy detection of the collected medical data, different texts are converted into a unified format; Use natural language processing methods to semantically annotate medical data after denoising and redundancy detection, and extract entities and entity relationships; According to the entity relationship structure of the medical knowledge graph, a graph database is used to store the medical knowledge graph; When the medical scale is generated, entities and entity relationships that are not covered by the original medical knowledge graph appear. External resources are called through external interfaces to expand the entities and entity relationships.

3. The AIGC-based medical scale quality control method according to claim 2, characterized in that: The dynamically expanded medical knowledge graph is embedded into the generative AI, and the generative AI is trained to obtain a medical scale generation model. The specific steps are as follows: Construct entity query, entity relationship query and semantic path query interfaces based on the query language of graph database; Based on entity query, entity relationship query and semantic path query interfaces, the medical knowledge graph is embedded in the generative AI to obtain the generative AI embedded in the medical knowledge graph; The generative AI embedded in the medical knowledge graph is trained to obtain a medical scale generation model.

4. The AIGC-based medical scale quality control method according to claim 3, characterized in that: The generative AI embedded in the medical knowledge graph is trained to obtain a medical scale generation model. The specific steps are as follows: Extract the items, answer options, and scoring methods from the CGI scale as the medical scale generation logic for generative AI embedded in the medical knowledge graph; Extract the semantic relationship between disease type, symptoms and treatment process from the medical knowledge graph as the background of the generation task; In the first stage of generative AI training embedded in the medical knowledge graph, diseases, evaluation dimensions, and medical semantic information are input according to the generation task context, and items are output; The second stage of generative AI training embedded in the medical knowledge graph outputs the corresponding answer options and scoring methods for the items based on the items output in the first stage and the semantic information of the medical knowledge graph; Using the cross entropy loss function, we calculated the gap between the output items and the corresponding answer options of the generative AI embedded in the medical knowledge graph, and the gap between the items and the corresponding answer options in the CGI scale. Calculate the n-gram overlap rate of the output items of the generative AI embedded in the medical knowledge graph and the corresponding answer options of the items, and the n-gram overlap rate of the items and the corresponding answer options in the CGI scale, and obtain the generation quality score of the output items of the generative AI embedded in the medical knowledge graph and the corresponding answer options of the items; Based on the length of the words and sentences of the items and the corresponding answer options in the CGI scale, the cross entropy loss value threshold and the generation quality score threshold of the output items and the corresponding answer options of the generative AI embedded in the medical knowledge graph are set; When the cross-entropy loss values ​​of the output items and the corresponding answer options of multiple consecutive training rounds of the generative AI embedded in the medical knowledge graph are less than the cross-entropy loss value threshold and the quality score is greater than the generation quality score threshold, the training of the generative AI embedded in the medical knowledge graph is completed and the medical scale generation model is obtained.

5. The AIGC-based medical scale quality control method according to claim 4, characterized in that: The medical scale generation model generates an initial medical scale by calling the medical knowledge graph interface. The specific steps are as follows: According to the symptoms described by the patient, the medical scale generation model calls the entity query, entity relationship query, and semantic path query interfaces of the medical knowledge graph to obtain the relevant entities, entity relationships, and semantic path information of the patient's disease; The query results of the knowledge graph on the relevant entities, entity relationships and semantic path information of the patient's disease are integrated and input into the medical scale generation model to output the initial medical scale about the patient's disease.

6. The AIGC-based medical scale quality control method according to claim 5, characterized in that: The consistency analysis of the initial medical scale is performed to generate the secondary medical scale. The specific steps are as follows: Each item of the initial medical scale is processed by sentence segmentation to decompose the entity and entity relationship of the item, and the entity and entity relationship of each item are mapped to the medical knowledge graph to verify whether they exist; If it does not exist, it means that the project does not conform to the medical knowledge graph, and the medical scale generation model needs to be retrained to regenerate the medical scale. If it does exist, continue with the subsequent analysis; For items in the initial medical scale, the text similarity between two items was calculated to determine whether the two items were semantically redundant; Based on the existing items without semantic duplication in the CGI scale, the text similarity between two items is calculated, and the highest text similarity value is used as the text similarity threshold; When the text similarity between two items in the initial medical scale is less than the text similarity threshold, it means that the semantics are not repeated and they are two independent items; On the contrary, it means that the two items are repeated, and the item with more semantic path information is retained, and the item with less semantic path information is deleted; The initial medical scale after medical knowledge graph verification and semantic duplication judgment is output as a secondary medical scale.

7. The AIGC-based medical scale quality control method according to claim 6, characterized in that: The reinforcement learning algorithm is used to dynamically optimize the secondary medical scale to generate the tertiary medical scale. The specific steps are as follows: The response time, completion rate and experience score for each item in the process of patients filling out the secondary medical scale are taken as the state vector; Adjusting the number of items, adjusting the order of items, and adjusting the description of items are used as action vectors; Set a reward function based on the patient's experience score for each item, answering time, and completion rate; The state vector collected in real time is input into the reinforcement learning algorithm, and the secondary medical scale is adjusted according to the action vector through the policy network of the reinforcement learning algorithm; After executing the action, an optimized quadratic medical scale is generated, and a reward function of the optimized quadratic medical scale is calculated; Update the policy network through the deep Q learning algorithm; By continuously optimizing the secondary medical scale in a cycle of state, action, and reward, after each round of optimization, the state vector is updated according to the patient's answer time for each item, completion rate, and experience score for each item; According to the change of reward value, a reward value change rate threshold is set. When the reward value change rate in multiple rounds of continuous optimization of the reinforcement learning algorithm is less than the reward value change rate threshold, it means that the reinforcement learning algorithm converges and outputs three medical scales; On the contrary, when the reward value change rate is greater than or equal to the reward value change rate threshold in multiple consecutive rounds of optimization, it means that the reinforcement learning algorithm has not converged, and the secondary medical scale continues to be optimized.

8. The AIGC-based medical scale quality control method according to claim 7, characterized in that: The reliability and structural validity of the three medical scales are calculated and evaluated to generate the final medical scale. The specific steps are as follows: The response results of the patients filling out the medical scale three times were organized into a response data matrix; Based on the response data matrix, the reliability of the three-item medical scale was calculated using Cronbach's Alpha formula; Collect the response results of the CGI scale, calculate the reliability of the CGI scale, and set the reliability of the CGI scale as the reliability threshold; Based on the response data matrix, the Pearson correlation coefficient between the scores of the corresponding response options of the two items is calculated to obtain a correlation matrix; Based on the correlation matrix, the average of all off-diagonal correlation values ​​was calculated to obtain the average correlation value of the three medical scales; Based on the response data matrix, the total scores of the corresponding response options for all items used in the medical scale filled out three times by each patient were summed up; The Pearson correlation coefficient between the patient's score on each item and the total score was calculated to obtain the overall correlation value of the three-item medical scale; Calculate the average correlation value and the overall correlation value of the CGI scale response records as the average correlation threshold and the overall correlation threshold; When the reliability, average correlation value and overall correlation value of the three-time medical scale are simultaneously less than the reliability threshold, average correlation threshold and overall correlation threshold, respectively, the reinforcement learning algorithm is used to re-optimize; Otherwise, the final medical scale is output.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the medical scale quality control method based on AIGC according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the AIGC-based medical scale quality control method according to any one of claims 1 to 8 are implemented.

Images