Medical follow-up visit suggestion generation method, system, equipment and medium

Abstract

The invention discloses a medical follow-up visit suggestion generation method, system and device and a medium, and relates to the technical field of computers, and the method comprises the steps: obtaining a target inspection report, and extracting a target abnormal value and a corresponding target abnormal value label from the target inspection report; inputting the target abnormal value label into a preset semantic vector model so as to perform vector conversion on the target abnormal value label through the preset semantic vector model to obtain a corresponding target feature vector; performing vector conversion on each abnormal value label in an abnormal value label-follow-up visit suggestion constructed in a key value pair form in a preset follow-up visit database through a preset semantic vector model to obtain a corresponding feature vector to be matched; and calculating the similarity between the target feature vector and each to-be-matched feature vector, determining a follow-up visit suggestion corresponding to the target feature vector based on a target similarity result with the highest similarity, obtaining a medical follow-up visit suggestion corresponding to the target abnormal value, and ensuring that the provided follow-up visit suggestion has pertinence.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, system, device and medium for generating medical follow-up suggestions. Background Art

[0002] With the rapid development of artificial intelligence and big data technologies, data-driven applications in the medical field have gradually become a research hotspot. Large models have performed well in the field of natural language processing, especially in processing massive medical data and complex medical problems. Medical follow-up, as an important part of medical services, is of great significance for patient rehabilitation, chronic disease management, and disease prevention. However, the generation of traditional follow-up recommendations mainly relies on the experience and subjective judgment of doctors, and has problems such as low efficiency and lack of personalization. Especially in the face of complex cases and abnormal situations, traditional methods often find it difficult to provide accurate follow-up recommendations. Therefore, how to use advanced artificial intelligence technology, especially large model technology, to improve the quality and efficiency of follow-up recommendations has become an important topic in current medical research.

[0003] At present, many studies have attempted to apply big models to the medical field, especially in medical image analysis, electronic medical record processing and disease prediction. For example, medical image diagnosis systems based on deep learning can reach or even exceed the level of human experts in certain specific disease detection; natural language processing technology also shows significant advantages in electronic medical record information extraction and structured processing. However, in the specific field of follow-up recommendation generation, related research and applications are still relatively limited. Some existing attempts mainly focus on using machine learning models to analyze and predict follow-up data, but these methods usually rely on specific feature engineering and model design, and it is difficult to fully utilize the powerful capabilities of big models.

[0004] In summary, how to achieve efficient automatic generation of follow-up recommendations for similar abnormalities is a technical problem to be solved in this field. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method, system, device and medium for generating medical follow-up suggestions, which can realize efficient automatic generation of similar abnormal follow-up suggestions. The specific scheme is as follows:

[0006] In a first aspect, the present application discloses a method for generating a medical follow-up recommendation, comprising:

[0007] Obtain a target inspection and testing report, and extract target outliers and corresponding target outlier labels from the target inspection and testing report;

[0008] Inputting the target outlier label into a preset semantic vector model, so as to perform vector conversion on the target outlier label through the preset semantic vector model to obtain a corresponding target feature vector;

[0009] Performing vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in the preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched;

[0010] The similarities between the target feature vector and each of the feature vectors to be matched are calculated respectively, and the follow-up recommendation corresponding to the target feature vector is determined based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up recommendation for the target abnormal value.

[0011] Optionally, before performing vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in the preset follow-up database by using the preset semantic vector model to obtain a corresponding feature vector to be matched, the method further includes:

[0012] Inputting the case text reports of each data party into a plurality of medical large language models respectively, so that each medical large language model performs text cleaning and text structuring processing on the case text reports to obtain structured case texts in a preset format;

[0013] Splitting each of the structured case texts according to the follow-up department and abnormal symptoms to obtain each abnormal value after the split, and marking each of the abnormal values ​​with a corresponding abnormal value label;

[0014] Input the case text reports of each data party into several large medical language models respectively, so that each large medical language model can extract the follow-up suggestions in the case text reports, and merge several follow-up suggestions for the same outlier label extracted by each large medical language model by voting and selecting the best, so as to determine the follow-up suggestion corresponding to the outlier label;

[0015] Each of the outlier labels and the corresponding follow-up suggestions are stored in the form of key-value pairs to obtain outlier labels-follow-up suggestions, and a preset follow-up database is constructed based on the outlier labels-follow-up suggestions.

[0016] Optionally, before inputting the case text reports of each data party into a plurality of large medical language models, the method further includes:

[0017] The preprocessed medical case text data and medical knowledge base content are used to perform outlier extraction training, follow-up suggestion extraction training, and follow-up suggestion prediction training on each initial large language model, so that each initial large language model can learn the correlation between medical terms and outliers, the relationship between case long sentences and abnormal information, the mapping relationship between case text and structured lists, and the correspondence between outlier labels and follow-up suggestions, so as to obtain the trained medical large language models.

[0018] Optionally, the structured case texts are split according to the follow-up departments and abnormal symptoms to obtain the split abnormal values, and the corresponding abnormal value labels are marked for the abnormal values, including:

[0019] Based on medical knowledge, each of the medical big language models is used to determine whether there is abnormal information in the structured medical record text. If so, all the abnormal information in the structured case text is split according to the follow-up department and abnormal symptoms, and an abnormal value list of each follow-up department is generated; wherein the abnormal value list includes the follow-up department, the corresponding abnormal values ​​and the abnormal value label of each abnormal value.

[0020] Optionally, all abnormal information in the structured case text is split according to the follow-up department and abnormal symptoms, and a list of abnormal values ​​of each follow-up department is generated, including:

[0021] The abnormal information in the structured case long sentence is extracted through the outlier extraction module in each of the medical large language models and based on the correlation between medical terms and abnormal information, and the abnormal information in each structured case long sentence is split according to the follow-up department and abnormal symptoms to obtain an outlier label constructed with the follow-up department and medical terms, and an outlier list constructed with the corresponding outlier values.

[0022] Optionally, determining the follow-up recommendation corresponding to the target feature vector based on the target similarity result with the highest similarity to obtain the corresponding medical follow-up recommendation for the target abnormal value includes:

[0023] If the target similarity result is greater than a preset similarity threshold, determining the follow-up suggestion corresponding to the to-be-matched feature vector in the target similarity result as the follow-up suggestion corresponding to the target feature vector;

[0024] If the target similarity result is less than or equal to the preset similarity threshold, the target outlier value label is input into each medical large language model so as to make corresponding follow-up suggestion recommendations based on each medical large language model, and determine the medical follow-up recommendation corresponding to the target outlier value according to the follow-up suggestion recommendation result.

[0025] Optionally, after determining the follow-up suggestion corresponding to the target feature vector based on the target similarity result with the highest similarity to obtain the medical follow-up suggestion corresponding to the target abnormal value, the method further includes:

[0026] The medical follow-up recommendation is given a textual explanation and precautionary description through a medical big language model to generate a long text of the medical follow-up recommendation.

[0027] In a second aspect, the present application discloses a medical follow-up suggestion generating system, comprising:

[0028] A report extraction module is used to obtain a target inspection and testing report, and extract target outliers and corresponding target outlier labels from the target inspection and testing report;

[0029] A vector conversion module, used for inputting the target outlier label into a preset semantic vector model, so as to perform vector conversion on the target outlier label through the preset semantic vector model to obtain a corresponding target feature vector;

[0030] A matching module, used for performing vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in a preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched;

[0031] The suggestion generation module is used to respectively calculate the similarity between the target feature vector and each feature vector to be matched, and determine the follow-up suggestion corresponding to the target feature vector based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up suggestion for the target abnormal value.

[0032] In a third aspect, the present application discloses an electronic device, comprising:

[0033] Memory, used to store computer programs;

[0034] The processor is used to execute the computer program to implement the steps of the medical follow-up recommendation generating method disclosed above.

[0035] In a fourth aspect, the present application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the steps of the aforementioned disclosed method for generating medical follow-up recommendations are implemented.

[0036] It can be seen that the present application discloses a method for generating medical follow-up suggestions, including: obtaining a target inspection and test report, and extracting a target abnormal value and a corresponding target abnormal value label from the target inspection and test report; inputting the target abnormal value label into a preset semantic vector model, so as to perform vector conversion on the target abnormal value label through the preset semantic vector model to obtain a corresponding target feature vector; performing vector conversion on each abnormal value label in the abnormal value label-follow-up suggestion constructed in the form of a key-value pair in a preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched; respectively calculating the similarity between the target feature vector and each of the feature vectors to be matched, and determining the follow-up suggestion corresponding to the target feature vector based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up suggestion of the target abnormal value. It can be seen that extracting the target abnormal value from the target inspection and test report can avoid errors in human judgment and accurately identify the abnormal condition of the patient through a professional extraction method. Using a preset semantic vector model to perform vector conversion on the target abnormal value label can convert complex medical information into a computable numerical form. This method can more accurately capture the essential characteristics of outliers and avoid the ambiguity and uncertainty of traditional text matching. By calculating the similarity between the target feature vector and the feature vector to be matched, the existing abnormal conditions and follow-up recommendations that are most similar to the target outliers can be found. This similarity-based matching method can make full use of historical data to provide patients with more accurate medical follow-up recommendations. The outlier label-follow-up recommendations in the preset follow-up database are constructed in the form of key-value pairs, and each outlier label corresponds to a specific follow-up recommendation. This one-to-one correspondence ensures that the follow-up recommendations provided to patients are targeted and avoids the inapplicability of general recommendations. In this way, the entire generation process is highly automated and does not require manual intervention. From outlier extraction to vector conversion, to similarity calculation and follow-up recommendation output, all are automatically completed by the computer, greatly improving work efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for generating medical follow-up recommendations disclosed in this application;

[0039] Figure 2 A medical system workflow diagram of a follow-up recommendation generation system disclosed in this application;

[0040] Figure 3 A diagram of a text structuring and outlier label extraction process based on a medical language large model disclosed in this application;

[0041] Figure 4 A flowchart for voting and selecting the best results for storage disclosed in this application;

[0042] Figure 5 A flow chart of a method for recommending follow-up suggestions based on vector similarity disclosed in this application;

[0043] Figure 6 A schematic diagram of the structure of a medical follow-up suggestion generating system disclosed in this application;

[0044] Figure 7 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. 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 work are within the scope of protection of the present invention.

[0046] With the rapid development of artificial intelligence and big data technologies, data-driven applications in the medical field have gradually become a research hotspot. Large models have performed well in the field of natural language processing, especially in processing massive medical data and complex medical problems. Medical follow-up, as an important part of medical services, is of great significance for patient rehabilitation, chronic disease management, and disease prevention. However, the generation of traditional follow-up recommendations mainly relies on the experience and subjective judgment of doctors, and has problems such as low efficiency and lack of personalization. Especially in the face of complex cases and abnormal situations, traditional methods often find it difficult to provide accurate follow-up recommendations. Therefore, how to use advanced artificial intelligence technology, especially large model technology, to improve the quality and efficiency of follow-up recommendations has become an important topic in current medical research.

[0047] At present, many studies have attempted to apply big models to the medical field, especially in medical image analysis, electronic medical record processing and disease prediction. For example, medical image diagnosis systems based on deep learning can reach or even exceed the level of human experts in certain specific disease detection; natural language processing technology also shows significant advantages in electronic medical record information extraction and structured processing. However, in the specific field of follow-up recommendation generation, related research and applications are still relatively limited. Some existing attempts mainly focus on using machine learning models to analyze and predict follow-up data, but these methods usually rely on specific feature engineering and model design, and it is difficult to fully utilize the powerful capabilities of big models.

[0048] To this end, the present invention provides a medical follow-up suggestion generation solution, which can realize efficient automatic generation of similar abnormal follow-up suggestions.

[0049] Reference Figure 1 As shown, an embodiment of the present invention discloses a method for generating a medical follow-up suggestion, comprising:

[0050] Step S11: Obtain a target inspection and testing report, and extract target outliers and corresponding target outlier labels from the target inspection and testing report.

[0051] In this embodiment, relevant medical case record text data is obtained from the clinical environment and used as a target inspection and test report, wherein the current target inspection and test report is an inspection report that does not carry corresponding medical follow-up recommendations. For new medical records that require follow-up recommendations, these texts are unstructured. In order to avoid interfering with the key medical information input into the large model, these medical record data must first be cleaned and organized. The present invention takes some common prompt sample texts as examples (the text before ":" is a fixed template, and the text after ":" is the medical record text to be processed), and the instructions for text structuring are shown in Table 1:

[0052] Table 1

[0053]

[0054] Furthermore, these structured medical record data need to be judged by the medical language model using medical knowledge to determine whether there are abnormalities. If there are abnormal values, the large model will split all abnormalities in each report to form a list of abnormal values ​​for the corresponding items. The present invention takes some common prompt sample texts as examples (the text before ":" is a fixed template, and the text after ":" is the medical record text to be processed) to show the instruction structure of abnormal value extraction, as well as the input and output texts, as shown in Table 2:

[0055] Table 2

[0056]

[0057] It can be seen that if multiple outliers appear in the output sample, the medical big language model will automatically assign them to a follow-up recommendation, and multiple outliers will form an outlier list.

[0058] In a specific embodiment, according to the data cube where the target inspection and test report is located, the medical language big model of the corresponding data cube is selected from several medical language big models after training, and the current target inspection and test report is input into the selected medical language big model, so that the medical language big model can extract the target abnormal value in the target inspection and test report. Specifically, the target abnormal value may include but is not limited to: abnormal values ​​of physiological indicators, abnormal values ​​of imaging examinations, abnormal values ​​related to disease history, abnormal values ​​of symptom manifestations, etc., wherein the abnormal values ​​of physiological indicators may include but are not limited to: abnormal blood test indicators, such as excessive or too low white blood cell counts in blood routine tests may indicate infection or immune dysfunction; abnormal red blood cell counts and hemoglobin content may reflect anemia and other conditions; abnormal platelet counts may be related to coagulation function. Abnormal biochemical test indicators, such as elevated alanine aminotransferase and aspartate aminotransferase in liver function tests, may indicate liver damage; elevated creatinine and urea nitrogen in renal function tests may indicate renal dysfunction. Abnormal endocrine test indicators, such as abnormal thyroid stimulating hormone and thyroxine in thyroid function tests, may indicate thyroid disease; abnormally elevated or decreased blood sugar may be a manifestation of diabetes or hypoglycemia. Abnormal values ​​of imaging examinations may include: abnormalities found in medical imaging examinations, such as X-rays, CT, and MRI. For example, chest CT shows linear shadows in the middle lobe of the right lung and the lower lobe of the left lung, which may indicate lesions such as lung inflammation and fibrosis; abdominal ultrasound shows cystic foci in the liver, which may indicate liver cysts, and cystic foci in the left kidney, which may indicate kidney cysts. Abnormalities in bone imaging examinations, such as fractures, bone hyperplasia, and narrowing of joint spaces. For example, old rib fractures of the 3rd to 5th ribs on the right. Abnormal values ​​related to disease history may include: patients have had certain specific diseases in the past, such as a history of heart disease, diabetes, and hypertension. These disease histories may be regarded as abnormal values ​​during follow-up and require special attention and management. Abnormal factors related to the current disease that may exist in the family history of genetic diseases, such as a family history of certain genetic diseases that may increase the risk of patients suffering from specific diseases. Abnormal values ​​of symptom manifestations may include: abnormal symptoms currently occurring in patients, such as chest pain, dyspnea, cough, abdominal pain, diarrhea, etc. These symptoms may be manifestations of multiple diseases and need to be comprehensively judged in combination with other examination results. Abnormal skin symptoms, such as rash, itching, redness and swelling, may indicate skin manifestations of allergies, skin diseases or other systemic diseases. Abnormal neurological symptoms, such as headache, dizziness, limb numbness, weakness, etc., may be related to cerebrovascular diseases, neurological diseases, etc.

[0059] In another specific implementation, if it is impossible to determine the medical big language model corresponding to the data party where the current target inspection and test report is located, the target inspection and test report is directly input into several medical big language models after training, so as to count all the outlier data results output by each medical big language model, remove duplicate outlier data results, and obtain the target outlier.

[0060] In this embodiment, when the target outlier value is obtained, the target outlier value label corresponding to the target outlier value is outputted simultaneously through the current medical big language model, wherein the target outlier value label is a medical term text used to describe the target outlier value. Since the target outlier value label is a concise expression obtained by summarizing and refining the abnormal conditions in the medical records, it is usually a word or phrase that can clearly point out the specific type of abnormality, has a clear semantic meaning, and is convenient for subsequent processing and analysis.

[0061] For example, if the target abnormal value is found in the medical record as "blood sugar value is 18mmol / L, higher than the normal range", the corresponding target abnormal value label can be "abnormal high blood sugar"; if the target abnormal value is that the chest CT shows "shadow in the right middle lobe of the lung", the target abnormal value label can be "abnormal shadow in the right middle lobe of the lung".

[0062] Step S12: inputting the target outlier value label into a preset semantic vector model, so as to perform vector conversion on the target outlier value label through the preset semantic vector model to obtain a corresponding target feature vector.

[0063] In this embodiment, since the anomaly-follow-up is stored in the follow-up library in pairs, a concise and efficient method is needed to quickly extract the follow-up suggestions to be generated. The present invention uses high-dimensional feature vectors to represent texts to distinguish their subtle semantic differences. At the same time, the similarity between different feature vectors can be calculated to obtain the best follow-up suggestion recommendation. Specifically, the semantic vector model is a method for converting discrete symbols or text data into continuous vector space representation. These vectors are usually called embeddings, which capture the semantic and syntactic information of symbols or texts in high-dimensional space. That is, the semantic vector model can effectively convert complex text data into a numerical form that can be processed by a machine learning model. The present invention uses a semantic vector model to convert each outlier label that needs to be recommended into a high-dimensional feature vector. Similarly, the outlier labels and follow-up suggestions in the follow-up library are also vectorized, thereby achieving a text processing effect that can calculate similarity.

[0064] Specifically, the target outlier label is input into the semantic vector model, and then the semantic vector model extracts features from the target outlier label of the target outlier, including vocabulary analysis, semantic understanding, context analysis and feature encoding. Among them, vocabulary analysis: analyze the input target outlier label at the vocabulary level to identify the key words and phrases therein. For example, for medical terms, specific disease names, symptom descriptions, examination items and other words will be identified. Semantic understanding: use deep learning algorithms, such as neural networks, to semantically understand the target outlier label. By learning a large amount of text data, the model can understand the meaning of each word and its changes in different contexts. Context analysis: consider the context information of the input target outlier, including the vocabulary relationship between the context and the sentence structure. This helps to understand the overall meaning of the text more accurately and avoid processing each word in isolation. Feature encoding: convert the analyzed and understood target outliers into high-dimensional feature vectors. This process uses a distributed representation method to map each word or text fragment to a vector position in a high-dimensional space. The numerical features of these vectors reflect the semantic and syntactic information of the word or text. Then we enter the vector generation stage, which specifically includes fusion features, optimization vectors and output vectors. Fusion features: fuse features extracted from different levels to form a comprehensive feature vector. This process includes weighted averaging, splicing and other operations to ensure that the importance of different features is reasonably reflected. Optimization vector: optimize the generated feature vector by continuously adjusting the parameters of the model. This can be achieved through optimization methods such as back propagation algorithms to minimize the loss function of the model on the training data. Output vector: The high-dimensional feature vector finally generated is used as the output of the semantic vector model. This vector can capture the rich semantic information of the input text, and has a certain distance relationship with the vectors of other similar texts in the vector space, reflecting the semantic similarity between them. The semantic vector model is a model that converts the symbolic representation of the text into a vector representation in the semantic space.

[0065] Step S13: performing vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in the preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched.

[0066] In this embodiment, as described above, each outlier label in the preset follow-up database is vector-converted to obtain a corresponding feature vector to be matched, wherein each outlier label in the library corresponds to a corresponding follow-up suggestion, and the outlier label-follow-up suggestion is stored in the preset follow-up database in the form of a key-value pair.

[0067] In this embodiment, before the preset semantic vector model is used to perform vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in the preset follow-up database to obtain the corresponding feature vector to be matched, it also includes: inputting the case text report of each data party into several medical large language models respectively, so that each medical large language model performs text cleaning and text structuring processing on the case text report to obtain a structured case text in a preset format; splitting each structured case text according to the follow-up department and abnormal symptoms to obtain each outlier after splitting, and marking each outlier with a corresponding outlier label; inputting the case text report of each data party into several medical large language models respectively, so that each medical large language model extracts the follow-up suggestions in the case text report, and merges several follow-up suggestions for the same outlier label extracted by each medical large language model by voting and selecting the best, so as to determine the follow-up suggestion corresponding to the outlier label; storing each outlier label and the corresponding follow-up suggestion in the form of key-value pairs to obtain outlier label-follow-up suggestion, and constructing a preset follow-up database based on the outlier label-follow-up suggestion. Specifically, each of the structured case texts is split according to the follow-up department and abnormal symptoms to obtain the split abnormal values, and a corresponding abnormal value label is marked for each of the abnormal values, including: judging whether there is abnormal information in the structured medical record text based on medical knowledge through each of the medical large language models; if so, all the abnormal information in the structured case text is split according to the follow-up department and abnormal symptoms, and generating an abnormal value list for each follow-up department; wherein the abnormal value list includes the follow-up department, the corresponding abnormal values ​​and the abnormal value label of each abnormal value.

[0068] like Figure 2As shown in the figure, the medical language model is first used to extract outlier labels. Specifically, in the process of deepening the capabilities of the medical model, in order to strengthen its ability to understand and process medical texts, the input and output pairs for extracting outlier labels are first constructed to ensure that the model can accurately identify and process complex information in the medical field. By providing examples with clear instructions, the model is guided to learn how to accurately identify outliers in unstructured medical texts. Therefore, it is necessary to first clean and process unstructured data, clean the data written by the original doctor, and remove useless information. Secondly, the examinations, tests and medical histories in the medical records are selected as inputs to find the abnormal-follow-up data pairs in the report to form a structured and easy-to-query follow-up database. The medical language model is used to remove invalid information and structure the text. In the process of extracting abnormal-follow-up pairs, invalid information such as non-medical terms and duplicate data will be eliminated to ensure the purity of the data set. These data are converted into JSON format readable by the big model for easy model understanding and processing. Table 3 shows the prompt information input to the large model in detail. You only need to replace the {input_data} part of the prompt information with the text to be processed, and the large model will output the processed text.

[0069] Table 3

[0070]

[0071] Furthermore, the abnormal information in the structured case long sentence is extracted through the abnormal value extraction module in each of the medical large language models based on the association between medical terms and abnormal information, and the abnormal information in each structured case long sentence is split according to the follow-up department and abnormal symptoms to obtain an abnormal value label constructed by the follow-up department and medical terms, and an abnormal value list constructed by the corresponding abnormal value. Figure 3 As shown, the structured text is not perfect. For example, when a doctor writes a follow-up recommendation, he or she may write multiple follow-up departments together, such as "Follow-up is recommended in the respiratory department and hepatobiliary surgery department." Secondly, in most cases, a follow-up department in the structured text contains multiple abnormal contents. These multiple abnormalities form a short sentence. In this case, follow-up recommendations cannot be provided independently for each abnormality. In this step, the text after the structured text will be processed to make the parsed text structure clearer. In Table 4, the prompt information input to the large model in this step is shown in detail, and the large model will output the abnormal extraction text.

[0072] Table 4

[0073]

[0074] like Figure 3 As shown, each element of the outlier list is an outlier label that can give an independent follow-up recommendation.

[0075] It should be noted that due to the different adaptability of the current large model on different data, the extraction effect from the follow-up advice text written by the doctor is also different. The present invention selects multiple open source large models to perform the corresponding label extraction and parsing process on the text. The initial large language model can specifically include ChatGLM3, Qianwen and Llama3, etc. Finally, a medical record written by a doctor corresponds to the structured extraction results of multiple models.

[0076] In this embodiment, the follow-up recommendations extracted by multiple large medical language models and the structured abnormality-follow-up text pairs will be put into the follow-up library. This process is not a decision made by a single model, but a voting method of multiple large models to fuse the prediction results of multiple models to achieve higher decision quality. Figure 4 As shown in the figure, in the process of using a large model to process medical record text, the large model first needs to output the text according to the preset prompt template format (that is, it has structuring capabilities), and then compare the effects of outliers and follow-up pairs. If a model cannot output in the template format, its subsequent processing will not be able to proceed, that is, it cannot participate in the voting process of multiple models. Finally, each large model will generate its corresponding results, that is, there will be a situation where one anomaly corresponds to multiple follow-up suggestions. At this time, the extraction results of different models for the same text will be displayed, and on this basis, the same results of the majority of models will be taken as the final results of the follow-up library (selecting the majority), and this process becomes a voting and selection process for multiple large models.

[0077] It should be noted that when the model extracts follow-up recommendations, the content of the medical knowledge base is also introduced as an additional input. This strategy aims to improve the decision-making quality and interpretability of the model. The medical knowledge base contains widely recognized guidelines, standards, and best practices in the medical field. This information can provide the model with more comprehensive background knowledge and help the model consider more dimensional factors when generating follow-up recommendations, such as the patient's specific condition, previous treatment history, and the latest medical research results. The follow-up library includes not only abnormal examination data, but also detailed follow-up records and doctor's advice. By structured processing and annotation of these data, a high-quality follow-up library containing rich follow-up data is formed to provide data support for the subsequent generation of follow-up recommendations. Secondly, the follow-up library will be regularly updated through process-based outlier extraction and voting for the best follow-up recommendations, incorporating new abnormal data and the latest medical research results to ensure the timeliness and effectiveness of follow-up recommendations.

[0078] In this embodiment, before the case text reports of each data party are input into several medical large language models, it also includes: using the pre-processed medical case text data and the content of the medical knowledge base to perform outlier extraction training, follow-up suggestion extraction training, and follow-up suggestion prediction training on each initial large language model, so that each initial large language model learns the correlation between medical terms and outliers, the relationship between case long sentences and abnormal information, the mapping relationship between case text and structured lists, and the corresponding relationship between outlier labels and follow-up suggestions, and obtains each trained medical large language model. It can be understood that for the outlier extraction training task, the mapping relationship or association relationship learned by each initial large language model: the association between medical terms and abnormal indicators: the large model needs to learn to identify which medical terms are related to abnormal indicators. For example, it can establish a connection between medical terms such as "elevated white blood cell count" and "low red blood cell count" and abnormal states. The relationship between long sentence structure and abnormal description: understand the grammatical structure and semantics of long sentences, and accurately extract abnormal descriptions from them. For example, it can distinguish between background information and abnormal information in sentences, and organize complex long sentences into a list of outlier labels. Mapping of unstructured text to structured lists: Convert unstructured medical records into a list structure that can be parsed by computers, with each element corresponding to an abnormal label, to achieve the conversion from natural language description to structured data. For the follow-up advice extraction training task, the mapping relationship or association relationship learned by each initial large language model: The correspondence between abnormal value labels and follow-up advice: The large model needs to learn what kind of follow-up advice should correspond to different abnormal value labels. For example, when the label "abnormal hypertension" appears, follow-up advice such as "control blood pressure, measure regularly, and adjust diet" can be extracted. The relationship between doctor-written text and structured follow-up advice: Extract key information from the unstructured follow-up advice text written by doctors and perform structured processing to establish a mapping between the original text and the structured follow-up advice. Rules for fusion of multi-model results: Since multiple large models are used to extract follow-up advice, the large model needs to learn how to fuse the results of multiple models through voting and other methods to determine the final follow-up advice. For the follow-up recommendation prediction training task, each initial large language model learns the mapping relationship or association relationship: Similarity association between new medical record features and existing follow-up recommendations: For new medical record data, the large model needs to learn how to compare its features with the existing data in the follow-up library to find similar historical cases. For example, the outlier labels of the new medical record are converted into high-dimensional feature vectors through the semantic vector model, and the similarity is calculated with the vectors in the follow-up library to find the most similar abnormal conditions and their follow-up recommendations. The relationship between outlier similarity and follow-up recommendation adjustment: When the outliers of the new medical record do not completely match the outliers in the follow-up library, the large model needs to learn how to adjust the follow-up recommendations based on the degree of similarity. If the similarity is high, the follow-up recommendations for similar abnormalities can be directly retrieved; if the similarity is low, it may be necessary to use the large model to generate new follow-up recommendations.Relationship between low-similarity outliers and large model resource allocation: For low-similarity outliers, the large model needs to learn how to reasonably allocate resources for follow-up suggestion generation to reduce resource consumption and deployment requirements while ensuring the accuracy and flexibility of the suggestions.

[0079] Step S14: respectively calculating the similarity between the target feature vector and each feature vector to be matched, and determining the follow-up recommendation corresponding to the target feature vector based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up recommendation for the target abnormal value.

[0080] In this embodiment, in order to give the most appropriate follow-up recommendation to the target outlier value to be recommended, the similarity between the target feature vector and each feature vector to be matched is calculated to obtain each target similarity result. If the target similarity result is greater than a preset similarity threshold, the follow-up recommendation corresponding to the feature vector to be matched in the target similarity result is determined as the follow-up recommendation corresponding to the target feature vector; if the target similarity result is less than or equal to the preset similarity threshold, the target outlier value label is input into each medical large language model so that corresponding follow-up recommendation can be made based on each medical large language model, and the medical follow-up recommendation corresponding to the target outlier value is determined according to the follow-up recommendation result. It can be understood that if the target similarity result is 100%, it indicates a complete match, and the corresponding follow-up suggestions in the follow-up library are directly retrieved as medical follow-up suggestions. If the target similarity result is between the preset similarity threshold and 100%, it indicates that there is no completely matching follow-up suggestion, but there are similar follow-up suggestions. Then the follow-up suggestion corresponding to the to-be-matched feature vector of the target similarity result with the highest similarity is selected as the medical follow-up suggestion. If the target similarity result is less than or equal to the preset similarity threshold, it indicates that the outlier has low similarity, and it is very likely that a few difficult and complicated diseases have been encountered. In the case of this embodiment, these target outliers will be put into the trained medical large language model for follow-up suggestion recommendations.

[0081] In this way, the follow-up suggestions given by the large model will be more flexible when the outliers have only low similarity with the outliers in the follow-up library. The resource capacity required to directly call the large model for each anomaly is high, and only generating low-similarity outliers for the large model will greatly reduce the resource capacity used for text processing and reduce the deployment requirements of the large model.

[0082] like Figure 5As shown in the figure, part of the content of the follow-up library and the processing process of a query example are shown. The follow-up library stores the follow-up of different departments and their corresponding abnormalities in the form of key-value pairs. For example, "Orthopedic follow-up" corresponds to "Old rib fracture of the 3rd to 5th ribs on the right side", and "Respiratory follow-up" corresponds to "Local enlargement and adhesion of bilateral pleura". The query content given is "Micro nodules under the pleura of the left lower lobe of the right lung". Through vector similarity calculation, the query is compared with each abnormality in the library. Among them, the similarity with "Old rib fracture of the 3rd to 5th ribs on the right side" is 0.4797, the similarity with "Local enlargement and adhesion of bilateral pleura" is 0.6680, and the similarity with "Micro hypodensity of the left thyroid gland" and "Low-density shadow of the upper segment of the right posterior lobe of the liver" are 0.6216 and 0.6035 respectively. Finally, it is determined that the one with the highest similarity to the query content is "Local enlargement and adhesion of bilateral pleura", which is recommended to be sent to "Respiratory follow-up". This process uses a semantic vector model to convert outlier labels into feature vectors, and determines the most suitable follow-up recommendations by calculating the similarity between the feature vector of the query content and the feature vector of each outlier label in the follow-up library, thereby providing patients with personalized medical follow-up recommendations.

[0083] In this embodiment, the medical follow-up suggestion is explained and explained with cautions through the medical big language model to generate a long text of the medical follow-up suggestion. It can be understood that after obtaining accurate medical follow-up suggestions from the follow-up library, since the follow-up suggestions in the follow-up library only include the follow-up departments, the medical big language model as a text generator can polish the text given to the user according to the key information of the follow-up department, and generate a long text of the medical follow-up suggestion that is easy for the user to understand by inputting the follow-up suggestion short sentence corresponding to the outlier label, so that the user has a better reading experience.

[0084] It can be seen that the present application discloses a method for generating medical follow-up suggestions, including: obtaining a target inspection and test report, and extracting a target abnormal value and a corresponding target abnormal value label from the target inspection and test report; inputting the target abnormal value label into a preset semantic vector model, so as to perform vector conversion on the target abnormal value label through the preset semantic vector model to obtain a corresponding target feature vector; performing vector conversion on each abnormal value label in the abnormal value label-follow-up suggestion constructed in the form of a key-value pair in a preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched; respectively calculating the similarity between the target feature vector and each of the feature vectors to be matched, and determining the follow-up suggestion corresponding to the target feature vector based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up suggestion of the target abnormal value. It can be seen that extracting the target abnormal value from the target inspection and test report can avoid errors in human judgment and accurately identify the abnormal condition of the patient through a professional extraction method. Using a preset semantic vector model to perform vector conversion on the target abnormal value label can convert complex medical information into a computable numerical form. This method can more accurately capture the essential characteristics of outliers and avoid the ambiguity and uncertainty of traditional text matching. By calculating the similarity between the target feature vector and the feature vector to be matched, the existing abnormal conditions and follow-up recommendations that are most similar to the target outliers can be found. This similarity-based matching method can make full use of historical data to provide patients with more accurate medical follow-up recommendations. The outlier label-follow-up recommendations in the preset follow-up database are constructed in the form of key-value pairs, and each outlier label corresponds to a specific follow-up recommendation. This one-to-one correspondence ensures that the follow-up recommendations provided to patients are targeted and avoids the inapplicability of general recommendations. In this way, the entire generation process is highly automated and does not require manual intervention. From outlier extraction to vector conversion, to similarity calculation and follow-up recommendation output, all are automatically completed by the computer, greatly improving work efficiency.

[0085] Reference Figure 6 As shown, the present invention also discloses a medical follow-up suggestion generating system, including:

[0086] A report extraction module 11 is used to obtain a target inspection and testing report, and extract target outliers and corresponding target outlier labels from the target inspection and testing report;

[0087] A vector conversion module 12, used for inputting the target outlier value label into a preset semantic vector model, so as to perform vector conversion on the target outlier value label through the preset semantic vector model to obtain a corresponding target feature vector;

[0088] A matching module 13 is used to perform vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in the preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched;

[0089] The suggestion generating module 14 is used to respectively calculate the similarity between the target feature vector and each feature vector to be matched, and determine the follow-up suggestion corresponding to the target feature vector based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up suggestion for the target abnormal value.

[0090] It can be seen that the present application discloses obtaining a target inspection and testing report, and extracting a target abnormal value and a corresponding target abnormal value label from the target inspection and testing report; inputting the target abnormal value label into a preset semantic vector model so as to perform vector conversion on the target abnormal value label through the preset semantic vector model to obtain a corresponding target feature vector; performing vector conversion on each abnormal value label in the abnormal value label-follow-up suggestion constructed in the form of a key-value pair in a preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched; respectively calculating the similarity between the target feature vector and each of the feature vectors to be matched, and determining the follow-up suggestion corresponding to the target feature vector based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up suggestion for the target abnormal value. It can be seen that extracting the target abnormal value from the target inspection and testing report can avoid errors in human judgment and accurately identify the abnormal condition of the patient through a professional extraction method. Using a preset semantic vector model to perform vector conversion on the target abnormal value label can convert complex medical information into a computable numerical form. This method can more accurately capture the essential characteristics of abnormal values ​​and avoid the ambiguity and uncertainty of traditional text matching. By calculating the similarity between the target feature vector and the feature vector to be matched, the existing abnormal situation and its follow-up recommendations that are most similar to the target abnormal value can be found. This similarity-based matching method can make full use of historical data to provide patients with more accurate medical follow-up recommendations. The abnormal value label-follow-up recommendations in the preset follow-up database are constructed in the form of key-value pairs, and each abnormal value label corresponds to a specific follow-up recommendation. This one-to-one correspondence ensures that the follow-up recommendations provided to patients are targeted and avoids the inapplicability of general recommendations. In this way, the entire generation process is highly automated and does not require manual intervention. From abnormal value extraction to vector conversion, to similarity calculation and follow-up recommendation output, all are automatically completed by the computer, greatly improving work efficiency.

[0091] Furthermore, the present application also discloses an electronic device. Figure 7This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be regarded as any limitation on the scope of use of the present application.

[0092] Figure 7 The present invention provides a schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the method for generating medical follow-up recommendations disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0093] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0094] Among them, the processor 21 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 21 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0095] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0096] Among them, the operating system 221 is used to manage and control the hardware devices and computer programs 222 on the electronic device 20, so as to realize the operation and processing of the massive data 223 in the memory 22 by the processor 21, which can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the medical follow-up suggestion generation method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks. In addition to data transmitted from an external device received by the electronic device, the data 223 can also include data collected by its own input and output interface 25.

[0097] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned disclosed method for generating medical follow-up recommendations is implemented. The specific steps of the method can be referred to the corresponding contents disclosed in the aforementioned embodiments, and will not be repeated here.

[0098] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0099] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application. The steps of the method or algorithm described in conjunction with the embodiments disclosed herein can be implemented directly with hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in a random access memory RAM (Random Access Memory), memory, read-only memory ROM (Read Only Memory), electrically programmable EPROM (Electrically Programmable Read Only Memory), electrically erasable programmable EEPROM (ElectricErasable Programmable Read Only Memory), register, hard disk, removable disk, CD-ROM (CompactDisc-Read Only Memory), or any other form of storage medium known in the technical field.

[0100] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0101] The scheme provided by the present invention is introduced in detail above. Specific examples are used in this article to illustrate the principle and implementation mode of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation mode and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for generating medical follow-up recommendations, characterized in that: include: Obtain a target inspection and testing report, and extract target outliers and corresponding target outlier labels from the target inspection and testing report; Inputting the target outlier label into a preset semantic vector model, so as to perform vector conversion on the target outlier label through the preset semantic vector model to obtain a corresponding target feature vector; Performing vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in the preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched; The similarities between the target feature vector and each of the feature vectors to be matched are calculated respectively, and the follow-up recommendation corresponding to the target feature vector is determined based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up recommendation for the target abnormal value.

2. The method for generating medical follow-up recommendations according to claim 1, characterized in that: Before performing vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in the preset follow-up database by using the preset semantic vector model to obtain the corresponding feature vector to be matched, the method further includes: Inputting the case text reports of each data party into a plurality of medical large language models respectively, so that each medical large language model performs text cleaning and text structuring processing on the case text reports to obtain structured case texts in a preset format; Splitting each of the structured case texts according to the follow-up department and abnormal symptoms to obtain each abnormal value after the split, and marking each of the abnormal values ​​with a corresponding abnormal value label; Input the case text reports of each data party into several large medical language models respectively, so that each large medical language model can extract the follow-up suggestions in the case text reports, and merge several follow-up suggestions for the same outlier label extracted by each large medical language model by voting and selecting the best, so as to determine the follow-up suggestion corresponding to the outlier label; Each of the outlier labels and the corresponding follow-up suggestions are stored in the form of key-value pairs to obtain outlier labels-follow-up suggestions, and a preset follow-up database is constructed based on the outlier labels-follow-up suggestions.

3. The method for generating medical follow-up recommendations according to claim 2, characterized in that: Before inputting the case text reports of each data party into a plurality of large medical language models, the method further includes: The preprocessed medical case text data and medical knowledge base content are used to perform outlier extraction training, follow-up suggestion extraction training, and follow-up suggestion prediction training on each initial large language model, so that each initial large language model can learn the correlation between medical terms and outliers, the relationship between case long sentences and abnormal information, the mapping relationship between case text and structured lists, and the correspondence between outlier labels and follow-up suggestions, so as to obtain the trained medical large language models.

4. The method for generating medical follow-up recommendations according to claim 2, characterized in that: The structured case texts are split according to the follow-up departments and abnormal symptoms to obtain the split abnormal values, and the corresponding abnormal value labels are marked for each abnormal value, including: Based on medical knowledge, each of the medical big language models is used to determine whether there is abnormal information in the structured medical record text. If so, all the abnormal information in the structured case text is split according to the follow-up department and abnormal symptoms, and an abnormal value list of each follow-up department is generated; wherein the abnormal value list includes the follow-up department, the corresponding abnormal values ​​and the abnormal value label of each abnormal value.

5. The method for generating medical follow-up recommendations according to claim 4, characterized in that: All abnormal information in the structured case text is split according to the follow-up department and abnormal symptoms, and a list of abnormal values ​​of each follow-up department is generated, including: The abnormal information in the structured case long sentence is extracted through the outlier extraction module in each of the medical large language models and based on the correlation between medical terms and abnormal information, and the abnormal information in each structured case long sentence is split according to the follow-up department and abnormal symptoms to obtain an outlier label constructed with the follow-up department and medical terms, and an outlier list constructed with the corresponding outlier values.

6. The method for generating medical follow-up recommendations according to claim 1, characterized in that: The determining the follow-up recommendation corresponding to the target feature vector based on the target similarity result with the highest similarity to obtain the corresponding medical follow-up recommendation of the target abnormal value includes: If the target similarity result is greater than a preset similarity threshold, determining the follow-up suggestion corresponding to the to-be-matched feature vector in the target similarity result as the follow-up suggestion corresponding to the target feature vector; If the target similarity result is less than or equal to the preset similarity threshold, the target outlier value label is input into each medical large language model so as to make corresponding follow-up suggestion recommendations based on each medical large language model, and determine the medical follow-up recommendation corresponding to the target outlier value according to the follow-up suggestion recommendation result.

7. The method for generating medical follow-up recommendations according to any one of claims 1 to 6, characterized in that: After determining the follow-up suggestion corresponding to the target feature vector based on the target similarity result with the highest similarity to obtain the medical follow-up suggestion corresponding to the target abnormal value, the method further includes: The medical follow-up recommendation is given a textual explanation and precautionary description through a medical big language model to generate a long text of the medical follow-up recommendation.

8. A medical follow-up recommendation generation system, characterized in that: include: A report extraction module is used to obtain a target inspection and testing report, and extract target outliers and corresponding target outlier labels from the target inspection and testing report; A vector conversion module, used for inputting the target outlier label into a preset semantic vector model, so as to perform vector conversion on the target outlier label through the preset semantic vector model to obtain a corresponding target feature vector; A matching module, used for performing vector conversion on each outlier label in the outlier label-follow-up suggestion constructed in the form of key-value pairs in a preset follow-up database through the preset semantic vector model to obtain a corresponding feature vector to be matched; The suggestion generation module is used to respectively calculate the similarity between the target feature vector and each feature vector to be matched, and determine the follow-up suggestion corresponding to the target feature vector based on the target similarity result with the highest similarity, so as to obtain the corresponding medical follow-up suggestion for the target abnormal value.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the method for generating medical follow-up recommendations as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program; wherein, when the computer program is executed by a processor, the steps of the method for generating a medical follow-up recommendation as described in any one of claims 1 to 7 are implemented.

Images