Diagnostic prediction method and related device, electronic device and storage medium

Abstract

The present application discloses a diagnosis prediction method and related devices, electronic devices and storage media, wherein the diagnosis prediction method includes: obtaining the medical record text of the target object, and obtaining the personal history text of the target object; wherein the personal history text includes the target locations where the target object has lived and the length of residence in each target location; obtaining the regional attribute information and disease rule information related to the personal history text; wherein the regional attribute information includes the first disease prevalent in each target location and its severity, the disease rule information includes the rules for the occurrence of various first diseases, and the occurrence rules include the mapping relationship between the length of residence in the target location where the first disease is prevalent and the possibility of suffering from the first disease; based on the medical record text, the personal history text, the regional attribute information and the disease rule information, the diagnosis prediction is performed to obtain the diagnosis prediction result of the target object. The above scheme can improve the accuracy of the diagnosis prediction result.

Description

Technical Field

[0001] The present application relates to the field of intelligent medical technology, and in particular to a diagnosis and prediction method and related devices, electronic equipment and storage media. Background Art

[0002] At present, common diagnosis recommendation systems mainly recommend diagnoses for patients based on their current medical history and past medical history, which can greatly save doctors' consultation time, assist doctors in making diagnoses quickly, and improve doctors' work efficiency. Generally, the similarity between two cases is calculated, and if the similarity is greater than a threshold, the case is retained and recommended to the user; or, by annotating a large amount of existing data, the medical record is compared with the annotated data, and the weight is assigned based on the pre-set weight, and finally the diagnosis with the highest weight is recommended to the user.

[0003] However, the former method relies too much on threshold judgment, and it is not scientific and rigorous enough to delete some data simply by using the threshold. The latter method only marks the gender, age, symptoms, and test results of each preset diagnosis. The information considered is not comprehensive enough and cannot distinguish the impact of different regions and hospitals on the diagnosis. Summary of the invention

[0004] The main technical problem solved by the present application is to provide a diagnosis prediction method and related devices, electronic devices and storage media, which can improve the accuracy of diagnosis prediction results.

[0005] In order to solve the above-mentioned technical problems, the first aspect of the present application provides a diagnosis prediction method, including: obtaining a medical record text of a target object, and obtaining a personal history text of the target object; wherein the personal history text includes the target locations where the target object has lived and the length of residence at each target location; obtaining regional attribute information and disease rule information related to the personal history text; wherein the regional attribute information includes the first diseases prevalent in each target location and their severity, the disease rule information includes the rules for the occurrence of various first diseases, and the occurrence rules include a mapping relationship between the length of residence at the target location where the first disease is prevalent and the possibility of suffering from the first disease; performing diagnosis prediction based on the medical record text, the personal history text, the regional attribute information and the disease rule information to obtain a diagnosis prediction result for the target object; wherein the diagnosis prediction result includes the probability values ​​of the target object suffering from several diseases, and the several diseases include the first disease.

[0006] In order to solve the above-mentioned technical problems, the second aspect of the present application provides a diagnosis and prediction device, including: a first acquisition module, used to obtain the medical record text of the target object, and to obtain the personal history text of the target object; wherein the personal history text includes the target places where the target object has lived and the length of residence in each target place; a second acquisition module, used to obtain regional attribute information and disease rule information related to the personal history text; wherein the regional attribute information includes the first diseases prevalent in each target place and their severity, the disease rule information includes the rules for suffering from various first diseases, and the suffering rules include the mapping relationship between the length of residence in the target place where the first disease is prevalent and the possibility of suffering from the first disease; a prediction module, used to perform diagnosis and prediction based on the medical record text, the personal history text, the attribute information and the disease rule information, and obtain the diagnosis prediction result of the target object; wherein the diagnosis and prediction result includes the probability value of the target object suffering from several diseases, and the several diseases include the first disease.

[0007] In order to solve the above technical problems, the third aspect of the present application provides an electronic device, including a memory and a processor coupled to each other, the memory storing program instructions, and the processor being used to execute the program instructions to implement the diagnostic prediction method in the above first aspect.

[0008] In order to solve the above technical problem, the fourth aspect of the present application provides a computer-readable storage medium, which stores program instructions that can be executed by a processor, and the program instructions are used to implement the diagnostic prediction method in the above first aspect.

[0009] The above scheme comprehensively considers the medical record text, personal history text, regional attribute information and disease rule information. That is, this application not only considers the gender, age, symptoms, test examination and other information of the target object in the medical record text, but also considers the target location and the length of residence information at the target location in the personal history text of the target object, as well as the regional attribute information and disease rule information related to the personal history text, etc., and generates a diagnosis prediction result on this basis, so that the diagnosis prediction result of the target object is more in line with the real scene and more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flowchart of an embodiment of a diagnostic prediction method provided by the present application;

[0011] Figure 2 This is a flow chart of an embodiment of obtaining a diagnosis prediction result based on rule matching provided by the present application;

[0012] Figure 3 This is a flow chart of another embodiment of obtaining a diagnosis prediction result based on rule matching provided by the present application;

[0013] Figure 4This is a flow chart of another embodiment of obtaining a diagnosis prediction result based on rule matching provided by the present application;

[0014] Figure 5 This is a flow chart of an embodiment of obtaining a diagnosis prediction result based on model prediction provided by the present application;

[0015] Figure 6 It is a structural schematic diagram of an embodiment of a diagnostic prediction model provided by the present application;

[0016] Figure 7 This is a flow chart of another embodiment of obtaining a diagnosis prediction result based on model prediction provided by the present application;

[0017] Figure 8 This is a flow chart of another embodiment of obtaining a diagnosis prediction result based on model prediction provided by the present application;

[0018] Fig. 9 It is a schematic diagram of the framework of an embodiment of a diagnosis and prediction device provided by the present application;

[0019] Fig.10 It is a schematic diagram of a framework of an embodiment of an electronic device provided by the present application;

[0020] Fig.11 It is a schematic diagram of a framework of an embodiment of a computer-readable storage medium provided by the present application. DETAILED DESCRIPTION

[0021] The scheme of the embodiment of the present application is described in detail below in conjunction with the drawings of the specification.

[0022] In the following description, for the purpose of explanation rather than limitation, specific details such as specific system structures, interfaces, and technologies are provided to facilitate a thorough understanding of the present application.

[0023] The terms "system" and "network" are often used interchangeably in this article. The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship. In addition, "many" in this article means two or more than two.

[0024] See also Figure 1 , Figure 1 1 is a flow chart of an embodiment of the diagnosis prediction method provided by the present application. It should be noted that if there are substantially the same results, this embodiment does not Figure 1 The process sequence shown is limited. Figure 1 As shown, this embodiment includes:

[0025] Step S11: Obtain the medical record text of the target object, and obtain the personal history text of the target object.

[0026] The method of this embodiment is used to generate a diagnosis prediction result of the target object based on the medical record text of the target object and the personal history text of the target object. The medical record text described in this article may specifically include but is not limited to: main suit, current illness history, previous history, auxiliary examination, routine checkup, etc., which are not limited here. It should be noted that the main suit is the patient's self-reported condition, such as "paroxysmal right upper abdominal pain with nausea and vomiting for 1 day"; the current medical history is a record of the whole process of the patient's illness, that is, the occurrence, development, evolution and diagnosis and treatment process, such as "paroxysmal right upper abdominal pain, nausea and vomiting for 1 day, no fever, no diarrhea"; the previous history is the patient's previous health status and past diseases asked by the doctor when the patient sees the doctor; auxiliary examinations may include but are not limited to: CT (Computed Tomography, computer tomography), B-ultrasound and other examinations; routine examinations may include but are not limited to: blood pressure, physical examinations and other examinations.

[0027] Among them, the personal history text described in this article includes the target places where the target object has lived and the length of residence in each target place. The target places include birthplace, growth place, residence place and short-term stay place, etc. For example, the personal history text of a patient A involves 4 target places where he has lived, specifically: birthplace (Quanjiao County)-18 years, growth place (Mingguang County)-4 years, residence place (Shangyi County)-2 years, short-term stay place (Tianchang County)-10 days.

[0028] In one embodiment, natural language processing (NLP) techniques such as key information extraction, named entity recognition, and entity association can be used to extract the personal history text of the target object from the description of the target object. In other embodiments, other processing techniques can also be used to extract the personal history text of the target object from the description of the target object, which is not specifically limited here.

[0029] Step S12: Acquire the regional attribute information and disease rule information related to the personal history text.

[0030] In this embodiment, the regional attribute information and disease rule information related to the personal history text are obtained. That is, based on the obtained personal history text of the target object, the regional attribute information and disease rule information related to the personal history text of the target object are obtained.

[0031] The regional attribute information includes the first disease prevalent in each target location and its severity. For example, the personal history text of a patient A involves three target locations where he has lived, namely: Zhangbei County, Shangyi County and Zhuolu County; the first disease prevalent in the three target locations is Keshan disease; the severity of Keshan disease prevalent in each target location is: Zhangbei County-Keshan disease-severe disease area, Shangyi County-Keshan disease-medium disease area, Zhuolu County-Keshan disease-mild disease area.

[0032] The disease rule information includes various first disease-suffering rules, and the suffering rules include a mapping relationship between the residence time at the target location where the first disease is prevalent and the possibility of suffering from the first disease. That is, there is a relationship between the probability of the target object suffering from the first disease and the residence time at the target location where the first disease is prevalent. Among them, the mapping relationship between the residence time at the target location where the first disease is prevalent and the possibility of suffering from the first disease can be that the longer the residence time, the greater the probability of suffering from the first disease; or it can be that the shorter the residence time, the greater the probability of suffering from the first disease. For example, the first disease is Keshan disease, and the mapping relationship between the residence time in Zhangbei County, where Keshan disease is prevalent, and the possibility of suffering from Keshan disease is: the longer the residence time in Zhangbei County, the greater the probability of the target object suffering from Keshan disease.

[0033] Step S13: Perform diagnosis prediction based on the medical record text, personal history text, regional attribute information and disease rule information to obtain the diagnosis prediction result of the target object.

[0034] Due to the unique geology, landforms, water quality, climate, food, living conditions, etc. in different regions, different regions may have their own unique epidemic diseases, which are divided into two categories: chemical epidemic diseases and biological epidemic diseases. For example, iodine deficiency can cause endemic goiter or cretinism, and places with excessive fluorine distribution can cause endemic fluorosis, and fluorine-deficient areas can cause dental caries and increased osteoporosis in the elderly; the latter include schistosomiasis, malaria, plague, etc. Therefore, considering only the target object's gender, age, symptoms, test results and other information is not scientific and comprehensive enough, and the diagnostic prediction results obtained are not in line with the real scene and have low accuracy.

[0035] In this embodiment, a diagnosis prediction is performed based on the medical record text, personal history text, regional attribute information and disease rules to obtain a diagnosis prediction result of the target object. That is to say, this application not only considers the gender, age, symptoms, test examination and other information of the target object in the medical record text, but also considers the target location and the residence time of the target location in the personal history text of the target object, as well as the regional attribute information and disease rule information related to the personal history text, that is, by comprehensively considering the medical record text, personal history text, regional attribute information and disease rule information to generate a diagnosis prediction result, the obtained diagnosis prediction result of the target object is more in line with the real scene and more accurate.

[0036] The diagnosis prediction result includes the probability values ​​of the target object suffering from several diseases respectively, and the several diseases include the first disease. For example, the diagnosis prediction results include: "septic arthritis", "Keshan disease", "acute cholecystitis", "bile duct stones", "upper respiratory tract infection" and "tonsillitis", and the probability value of the target object suffering from "septic arthritis" is 0.4, the probability value of suffering from "Keshan disease" is 0.3, the probability value of suffering from "acute cholecystitis" is 0.2, the probability value of suffering from "bile duct stones" is 0.5, the probability value of suffering from "upper respiratory tract infection" is 0.4, and the probability value of suffering from "tonsillitis" is 0.8.

[0037] In one embodiment, the disease corresponding to the largest probability value can be used as the diagnosis prediction result for user reference. In other embodiments, the probability values ​​can be sorted in descending order, and several diseases ranked before the preset order can be selected as the diagnosis prediction result for user reference. Among them, the preset order can be set according to the actual application needs. For example, it can be set to 3, 4, 5, etc., which is not limited here.

[0038] In one embodiment, when generating the diagnosis prediction results of the target object, the information of the disease's high-incidence season can also be considered at the same time. Considering the high-incidence time of different diseases, for example, hand, foot and mouth disease is more likely to break out in summer and autumn, and chronic obstructive pulmonary disease usually breaks out in winter and spring, so considering the information of the disease's high-incidence season at the same time can dynamically adjust the ranking of various diseases in the diagnosis prediction results, thereby improving the accuracy of the diagnosis prediction results.

[0039] In other embodiments, when generating the diagnosis prediction results of the target object, the incidence probability rule information of hospitals of different levels can also be considered at the same time. There are differences in the frequency of diseases in hospitals of different levels. For example, tertiary hospitals are more likely to have critical and severe diseases and rare diseases, while primary hospitals are more likely to have common diseases. For example, cough, sputum, and asthma may correspond to bronchitis, acute exacerbation of chronic obstructive pulmonary disease, etc. The probability of acute exacerbation of chronic obstructive pulmonary disease in tertiary hospitals is much higher than that of bronchitis, while the opposite is true for primary hospitals. Therefore, taking into account the incidence probability rule information of hospitals of different levels can make the diagnosis prediction results more fully in line with the real scene and improve the accuracy of the diagnosis prediction results.

[0040] In one embodiment, the diagnosis prediction result may be obtained based on rule matching, that is, based on the rule matching method, the medical record text, personal history text, regional attribute information and disease rule information are processed to obtain the diagnosis prediction result of the target object. It is understandable that in other embodiments, in order to improve the generalization ability of generating the diagnosis prediction result of the target object, the diagnosis prediction result may also be obtained based on model prediction, that is, based on the model prediction method, the medical record text, personal history text, regional attribute information and disease rule information are processed to obtain the diagnosis prediction result of the target object. It is understandable that in other embodiments, the diagnosis prediction result obtained based on rule matching and the diagnosis prediction result obtained based on model prediction may also be fused to obtain the final diagnosis prediction result.

[0041] The above scheme comprehensively considers the medical record text, personal history text, regional attribute information and disease rule information. That is, this application not only considers the gender, age, symptoms, test examination and other information of the target object in the medical record text, but also considers the target location and the length of residence information at the target location in the personal history text of the target object, as well as the regional attribute information and disease rule information related to the personal history text, etc., and generates a diagnosis prediction result on this basis, so that the diagnosis prediction result of the target object is more in line with the real scene and more accurate.

[0042] See also Figure 2 , Figure 2 1 is a flow chart of an embodiment of obtaining a diagnosis prediction result based on rule matching provided by the present application. It should be noted that if there are substantially the same results, this embodiment does not use Figure 2 The process sequence shown is limited. Figure 2 As shown, in this embodiment, the diagnosis prediction result is obtained based on rule matching, specifically including:

[0043] Step S21: based on the severity of the first disease prevalent in the target location, a first weight coefficient of the first disease is obtained, and based on the affliction rule, a second weight coefficient of the first disease is obtained.

[0044] In this embodiment, a first weight coefficient of the first disease is obtained according to the severity of the first disease prevalent in the target location; and a second weight coefficient of the first disease is obtained according to the rule of suffering from the first disease. Among them, the relationship between the severity of the first disease prevalent in the target location and the first weight coefficient of the first disease, and the relationship between the rule of suffering from the first disease and the second weight coefficient of the first disease are not specifically limited. For example, the severity of the first disease prevalent in the target location and the first weight coefficient of the first disease are positively correlated, that is, the higher the severity of the first disease prevalent in the target location, the larger the first weight coefficient; the length of residence in the target location where the first disease is prevalent and the second weight coefficient of the first disease are positively correlated, that is, the longer the length of residence in the target location where the first disease is prevalent, the larger the second weight coefficient.

[0045] Step S22: The first weight coefficient and the second weight coefficient are integrated to obtain a first probability value that the target object suffers from the first disease.

[0046] In this embodiment, the first weight coefficient of the first disease and the second weight coefficient of the first disease are fused to obtain a first probability value that the target object suffers from the first disease.

[0047] In one embodiment, the first weight coefficient and the second weight coefficient may be weighted and summed to obtain a first probability value of the target object suffering from the first disease. The weights of the first weight coefficient and the second weight coefficient are not limited and may be specifically set according to actual use needs. For example, the first weight coefficient is 0.4, the weight corresponding to the first weight coefficient is 0.3, the second weight coefficient is 0.6, and the weight corresponding to the second weight coefficient is 0.7, so the first probability value of the target object suffering from the first disease = 0.4 × 0.3 + 0.6 × 0.7 = 0.54.

[0048] It is understandable that in other implementations, the first weight coefficient and the second weight coefficient may also be summed to obtain a first probability value that the target object suffers from the first disease.

[0049] Step S23: Acquire diagnosis rule information of a plurality of second diseases.

[0050] In this embodiment, the diagnosis rule information includes several rule texts of the diagnosis text. The diagnosis text can be specifically set according to the actual application needs. For example, for community hospitals, clinics and other primary medical institutions, the diagnosis text can include common diagnosis texts, such as upper respiratory tract infection, tonsillitis, etc.; or, for pediatrics, orthopedics and other specialized hospitals, the diagnosis text can include diagnosis texts related to medical specialties, such as for orthopedic hospitals, the diagnosis text can specifically include: suppurative arthritis, osteomalacia, etc.; or, for general hospitals, the diagnosis text can include diagnosis texts related to each medical specialty, such as: diagnosis texts related to orthopedics such as the above-mentioned suppurative arthritis and osteomalacia, diagnosis texts related to general surgery such as acute cholecystitis and bile duct stones, and diagnosis texts related to medical disciplines such as cardiology, neurology, burns, andrology, and gynecology, etc., are not given one by one here.

[0051] It should be noted that, in the embodiment of the present disclosure, each diagnosis text is provided with a corresponding number of rule texts.

[0052] In a specific implementation scenario, in order to improve the accuracy of subsequent diagnosis recommendations, the plurality of rule texts may specifically include: a first text related to at least one of a sufficient condition rule and a necessary condition rule, and a second text related to a negative condition rule. When the sufficient condition rule of the diagnosis text is met, it can be inferred with a high probability that the medical record text matches the diagnosis text; when the necessary condition rule of the diagnosis text is not met, it can be inferred that the medical record text does not match the diagnosis text; when the negative condition rule of the diagnosis text is met, it can be inferred that the medical record text does not match the diagnosis text.

[0053] In another specific implementation scenario, the rule text may specifically include several rule words. In order to improve the accuracy of subsequent diagnosis recommendations, the rule words may also be marked with their corresponding word types. The word types may include but are not limited to: causes, symptoms, signs, etc., which are not limited here.

[0054] Taking the diagnosis text "acute cholecystitis" as an example, it may include the following first texts related to the sufficient condition rule: "right upper abdominal pain 6|right upper abdominal distension 6|right upper abdominal discomfort 6|shoulder and back radiating pain 6, tenderness in the gallbladder area 7|right upper abdominal rebound pain 7|Murphy's disease positive 7|nausea 6|vomiting 6, ^jaundice 6, ^skin and sclera yellowing 7, ^digestive tract perforation 9", "eating greasy food 5|drinking 5, right upper abdominal pain 6|right upper abdominal tenderness 7, ^right upper abdominal pain 6, ^jaundice 6, ^skin and sclera yellowing 7, ^improper diet 5", "right upper abdominal pain 6, nausea 6|vomiting 6|shoulder and back radiating pain 6, ^gallbladder stones 3, ^improper diet 5, ^digestive tract perforation 9", etc., which are not limited here. In addition, it may also include the following first texts related to the necessary condition rule: "right upper abdominal pain 6|right upper abdominal distension 6|upper abdominal discomfort 6|shoulder and back radiating pain 6", etc., which are not limited here. In addition, it may also include the following second text related to the negative conditional rule: "Lower abdominal pain 6 | Lower abdominal distension 6 | Left lower abdominal pain 6 | Left lower abdominal distension 6", "Kidney area percussion pain 7 | McBurney's point tenderness 7", etc., which are not limited here.

[0055] In the above rule texts, the numbers "5", "6", and "7" all represent the word types corresponding to the rule words, where the number "5" represents the cause, the number "6" represents the symptom, and the number "7" represents the sign. Using the above numbers to mark the word types is only one possible way in practical applications. In other implementation scenarios, other numbers can also be used to mark the word types, or other characters such as letters can also be used to mark the word types, which are not limited here.

[0056] In addition, in the above rule text, '|' represents logical OR, ',' represents logical AND, and '^' represents logical NOT. For example, for the above rule text "lower abdominal pain 6|lower abdominal distension 6|left lower abdominal pain 6|left lower abdominal distension 6", its specific meaning is "lower abdominal pain or lower abdominal distension or left lower abdominal pain or left lower abdominal distension", and other rule texts can be deduced by analogy, and no examples are given here one by one.

[0057] In addition, it should be noted that in order to improve the accuracy of diagnostic recommendations, the rule base of diagnostic texts can be updated during use. For example, when a doctor finds that a rule text is no longer applicable, it can be modified or deleted, without limitation; or, when a doctor finds a more accurate rule text, the rule text can be added to the rule base, or the rule text can replace an old rule text in the rule base, without limitation.

[0058] Step S24: based on the semantic similarities between the medical record text and the rule texts of various second diseases, obtain second probability values ​​of the target object suffering from various second diseases.

[0059] In this embodiment, the second probability value of the target object suffering from various second diseases is obtained according to the semantic similarity between the medical record text and the rule text of various second diseases. That is to say, the semantic similarity between the medical record text and the rule text of various second diseases can be obtained by using the semantic representation of the medical record and the rule semantic representation of the rule text of several second diseases. On this basis, the semantic similarity between the medical record text and various second diseases can be obtained based on the semantic similarity between the medical record text and the rule text of various second diseases, that is, the second probability value of the target object suffering from various second diseases can be obtained.

[0060] In one implementation, semantic extraction may be performed directly on the medical record text to obtain a semantic representation of the medical record.

[0061] In a specific implementation scenario, the medical record text can be segmented to obtain a number of medical record words, and the medical record words can be vector-mapped to obtain word vectors of the medical record words, so that the word vectors can be semantically extracted to obtain word semantic representations of the medical record words, and then the combination of word semantic representations of the medical record words can be used as the medical record semantic representation. For example, the medical record text contains |R| medical record words, and the word semantic representation of each medical record word is a d (e.g., 256)-dimensional vector, then the medical record semantic representation can be represented as a vector with a dimension of |R|*d.

[0062] In another specific implementation scenario, in order to improve the efficiency of semantic extraction, a diagnosis recommendation model can be pre-trained, and the diagnosis recommendation model includes a medical record semantic extraction network, so that the word vector of the above medical record words can be input into the medical record semantic extraction network to obtain the medical record semantic representation. The medical record semantic extraction network can include but is not limited to: BERT (Bidirectional Encoder Representation from Transformers), EMLO (Embedding from Language Model) model, etc., which are not limited here.

[0063] In another implementation scenario, in order to improve the accuracy of the semantic representation of the medical record, the key words in the medical record text can also be identified, and the medical record text and the key words are spliced ​​to update the medical record text, so that the updated medical record text can be semantically extracted to obtain the medical record semantic representation. Specifically, the above key words may include but are not limited to: symptoms, inducements, medical history, etc., which are not limited here. The above method, by identifying the key words in the medical record text, splicing the medical record text with the key words to update the medical record text, and then extracting the updated medical record text semantically to obtain the medical record semantic representation, can be beneficial to enhance the medical record text, improve the richness of the medical record text, and improve the accuracy of the medical record semantic representation. In addition, in the subsequent calculation of the correlation process with the rule semantic representation of the rule text, not only the semantic information of the medical record itself can be considered, but also the semantic information of the extracted key words can be considered, which can be beneficial to improve the accuracy of the correlation. The key words may be specifically obtained by identifying the medical record text using NER (Named-Entity Recognition) tools such as HanLP, Jieba, and LTP, and are not limited here.

[0064] In one embodiment, as mentioned above, the rule text of the second disease contains several rule words, on this basis, the semantic representation of several rule words of the second disease can be obtained, so that the rule semantic representation of the rule text of the second disease can be obtained based on the combination of the semantic representations of several rule words of the second disease. In a specific implementation scenario, the semantic representation of the rule words can be obtained by pre-training. For example, the semantic representation of each word in the vocabulary obtained by pre-training of the BERT model can be obtained, and the semantic representation of the words in the vocabulary corresponding to the rule words can be used as the semantic representation of the corresponding rule words. In another specific implementation scenario, in order to facilitate the subsequent calculation of relevance, the combination of the semantic representations of several rule words can also be subjected to dimensionality reduction processing to obtain the rule semantic representation of the rule text. Specifically, the combination of the semantic representations of the above-mentioned rule words can be subjected to dimensionality reduction processing using a convolutional neural network (CNN).

[0065] In another embodiment, as mentioned above, the rule text contains several rule words, on this basis, the first semantic representation of the rule words can be obtained, and the second semantic representation of the rule type of the rule text can be obtained, so that the first semantic representations of several rule words and the second semantic representations of the rule type can be fused to obtain the rule semantic representation of the rule text. Specifically, the rule type may include but is not limited to: sufficient condition rule, necessary condition rule, negative condition rule, the specific meaning of the above rule types can refer to the above related description, and will not be repeated here. The above method, by obtaining the first semantic representation of the rule words, and obtaining the second semantic representation of the rule type of the rule text, so that the first semantic representations of several rule words and the second semantic representations of the rule type can be fused to obtain the rule semantic representation of the rule text, can make the rule semantic representation contain not only the semantic information of the rule text itself, but also the semantic information of the rule type, which can help improve the accuracy of the rule semantic representation of the rule text, and help improve the accuracy of subsequent diagnosis recommendations.

[0066] In a specific implementation, the second semantic representation of the rule type may be a random vector, that is, the second semantic representation may be obtained by random initialization.

[0067] In another specific embodiment, the rule semantic representation can be specifically represented as a vector of a preset dimension d (e.g., 256). In addition, for ease of processing, a maximum number M of rule texts can be pre-set, such as 58. Then, when the number of rule texts contained in the rule base of the diagnostic text is less than the above maximum number, the number of rule texts can be supplemented to the maximum number, so that the rule semantic representation of the several rule texts corresponding to the diagnostic text can be finally represented as a vector matrix with a dimension of M*d, such as a vector matrix with a dimension of 58*256. For example, the several rule texts corresponding to the diagnostic text contain rule texts related to 3 sufficient condition rules, 1 necessary condition rule and 2 negative condition rules. The rule semantic representation of the diagnostic text can be represented as a vector matrix with a dimension of 6*d, and the insufficient 52*d can be supplemented with 0 elements. In addition, for ease of description, the vector matrix can be recorded as.

[0068] In another embodiment, in order to improve the extraction efficiency and accuracy of the semantic representation of the rule, before extracting the semantic representation of the rule, it is also possible to detect whether the rule text contains a preset character for representing a logical OR (such as the aforementioned '|'), and if the rule text contains the preset character, split the rule text into at least two sub-rule texts based on the preset character, so that semantic extraction can be performed on the split sub-rule texts separately later. The above method can help improve the efficiency and accuracy of extracting the semantic representation of the rule by detecting whether the rule text contains a preset character for representing a logical OR before respectively extracting the semantic representation of the rule texts of several rule texts in the rule base, and if the rule text contains the preset character, split the rule text into at least two sub-rule texts based on the preset character.

[0069] In a specific implementation, the rule text can be split according to preset characters, and in the splitting process, the rule words related to logical AND and logical non-relevant are retained. Taking the rule text "dizziness|headache, ^abdominal pain" as an example, it can be split according to the preset character '|', and the rule word "abdominal pain" related to logical AND and logical non-relevant is retained, and finally two sub-rule texts "dizziness, ^abdominal pain" and "headache, ^abdominal pain" can be split. Other situations can be deduced by analogy, and no examples are given here one by one.

[0070] Step S25: combining the first probability values ​​of the target object suffering from various first diseases and the second probability values ​​of the target object suffering from various second diseases to obtain the probability values ​​of the target object suffering from various diseases.

[0071] In this embodiment, the first probability values ​​of the target object suffering from various first diseases and the second probability values ​​of the target object suffering from various second diseases are combined to obtain the probability values ​​of the target object suffering from various diseases.

[0072] In one embodiment, there is no identical disease among the first disease and the second disease, then the probability values ​​of the target object suffering from the various diseases are directly obtained according to the first probability values ​​of the target object suffering from the various first diseases and the second probability values ​​of the target object suffering from the various second diseases. For example, the first probability values ​​of the target object suffering from the various first diseases are: the first probability value of "cretinism" is 0.1, the first probability value of "schistosomiasis" is 0.4, the first probability value of "plague" is 0.6, and the first probability value of "osteoporosis in the elderly" is 0.5; the second probability values ​​of the target object suffering from the various second diseases are: the second probability value of "septic arthritis" is 0.4, the second probability value of "Kashin-Beck disease" is 0.3, the second probability value of "acute cholecystitis" is 0.2, the second probability value of "bile duct stones" is 0.5, the second probability value of "upper respiratory tract infection" is 0.4, and the second probability value of "tonsillitis" is 0. The second probability value of " is 0.8; since there is no identical disease between various first diseases and various second diseases, the probability values ​​of the target object suffering from various diseases are: the probability value of "cretinism" is 0.1, the probability value of "schistosomiasis" is 0.4, the probability value of "plague" is 0.6, the probability value of "senile osteoporosis" is 0.5, the probability value of "septic arthritis" is 0.4, the probability value of "Kashin-Beck disease" is 0.3, the probability value of "acute cholecystitis" is 0.2, the probability value of "bile duct stones" is 0.5, the probability value of "upper respiratory tract infection" is 0.4, and the probability value of "tonsillitis" is 0.8.

[0073] In other embodiments, there are some common diseases in the various diseases of the first disease and the second disease, so it is necessary to fuse the first probability value and the second probability value of the disease to obtain the final probability value corresponding to the disease. In a specific embodiment, the first probability value and the second probability value of the disease can be directly added to obtain the probability value of the disease. In other specific embodiments, the product of the first preset coefficient and the first probability value can also be added on the basis of the second probability value to obtain the probability value of the disease. Among them, the size of the first preset coefficient is not limited, and can be specifically set according to actual use needs. For example, the first probability value of the first disease "cretinism" is 0.3, the second probability value of the second disease "cretinism" is 0.5, and the first preset coefficient is 0.2, so the probability value of the disease "cretinism" = 0.5 + 0.3 × 0.2 = 0.56.

[0074] See also Figure 3 , Figure 3 FIG. 1 is a flow chart of another embodiment of the present application for obtaining a diagnosis prediction result based on rule matching. It should be noted that if there are substantially the same results, this embodiment does not use Figure 3 The process sequence shown is limited. Figure 3As shown, in this embodiment, the information of the season when the disease is prone to occur is taken into account at the same time, that is, the medical record text, personal history text, regional attribute information, disease rule information and disease prone season information are taken into account at the same time to obtain the probability values ​​of the target object suffering from several diseases respectively, specifically including:

[0075] Step S31: based on the medical record text of the target object, the onset time of the target object is obtained, and based on the prevalence rule information, the prevalence time of various diseases is obtained.

[0076] In this embodiment, the onset time of the target object is obtained based on the medical record text of the target object, and the prevalence time of various diseases is obtained based on the disease prevalence rule information.

[0077] For example, the target patient's visit date is June 28, 2021, and the duration of the symptoms in the main complaint extracted from the target patient's medical record is 3 days. Therefore, according to the visit date and the duration of the symptoms in the medical record, the onset time of the target patient is June 25, 2021. For another example, hand, foot and mouth disease is more likely to break out in summer and autumn, and chronic obstructive pulmonary disease usually breaks out in winter and spring. The seasons are converted into corresponding months, such as March, April and May in the northern hemisphere and September, October and November in the southern hemisphere.

[0078] Step S32: In response to the presence of a disease whose prevalence time coincides with the onset time, the probability value of the disease is increased.

[0079] In this embodiment, in response to the presence of a disease whose predisposition time coincides with the onset time, the probability value of the disease is increased. That is to say, among the various diseases that the target object may suffer from obtained in the above manner, if there is a disease whose predisposition time coincides with the onset time of the target object, the probability value of the disease is increased accordingly. Among them, the size of the probability value adjustment is not limited, and can be specifically set according to actual use needs, for example, increasing by 0.1, 0.15, 0.2, etc. At the same time, considering the predisposition time information of different diseases, such as hand, foot and mouth disease is more likely to break out in summer and autumn, and chronic obstructive pulmonary disease usually occurs in winter and spring, the probability value of each disease in the diagnosis prediction results can be adjusted, thereby dynamically adjusting the ranking of various diseases, thereby improving the accuracy of the diagnosis prediction results.

[0080] For example, the onset time of the target object is April 25, 2021, the peak period of "hand, foot and mouth disease" that the target object may suffer from obtained through the above method is March-May, and the probability value of "hand, foot and mouth disease" is 0.3. Therefore, the peak period of "hand, foot and mouth disease" coincides with the onset time of the target object. Therefore, the probability value of "hand, foot and mouth disease" is increased by 0.1, that is, 0.4.

[0081] See also Figure 4 , Figure 4 FIG. 1 is a flow chart of another embodiment of the present application for obtaining a diagnosis prediction result based on rule matching. It should be noted that if there are substantially the same results, this embodiment does not use Figure 4 The process sequence shown is limited. Figure 4 As shown, in this embodiment, the information of the incidence probability rules of hospitals of different levels is considered at the same time, that is, the medical record text, personal history text, regional attribute information, disease rule information and the incidence probability rule information of hospitals of different levels are considered at the same time to obtain the probability values ​​of the target object suffering from several diseases respectively, specifically including:

[0082] Step S41: Based on the medical record text of the target object, obtain the medical institution where the target object has been treated.

[0083] In this embodiment, the target object's medical institution is obtained based on the target object's medical record text. For example, the target object's medical institution is a tertiary hospital, a primary hospital, or the like.

[0084] Step S42: Obtain the diagnosis statistics of the medical institution.

[0085] In this embodiment, based on the target object's medical institution obtained from the target object's medical record text, the diagnostic statistical information is retrieved from the target object's medical institution, thereby obtaining the diagnostic statistical information of the medical institution. The diagnostic statistical information of the medical institution includes the proportion of each disease diagnosed in the total number of patients received by the medical institution.

[0086] Step S43: Adjust the probability value of the disease based on the proportion corresponding to the disease.

[0087] In this embodiment, the probability value of the disease is adjusted according to the proportion of the disease corresponding to the target object's medical institution. Among them, the proportion of the disease corresponding to the medical institution is positively correlated with the adjustment range of the probability value of the disease, that is, the greater the proportion of a certain disease that the target object may suffer from in the target object's medical institution, the greater the adjustment range of the probability value of the disease. Among them, the adjustment range of the probability value corresponding to each proportion is not limited, and can be specifically set according to actual use needs. There are differences in the frequency of diseases in hospitals of different levels. For example, tertiary hospitals are more likely to have critical and severe diseases and rare diseases, and primary hospitals are more likely to have common diseases; for example, coughing, sputum, and asthma may correspond to bronchitis, acute exacerbation of chronic obstructive pulmonary disease, etc. The probability of acute exacerbation of chronic obstructive pulmonary disease in tertiary hospitals is much higher than that of bronchitis, while the opposite is true in primary hospitals. Therefore, the probability values ​​of the target object suffering from several diseases obtained by taking into account the probability rule information of different levels of hospitals can be more fully in line with the real scene and can improve the accuracy of the diagnosis prediction results.

[0088] For example, the target object may suffer from "hand, foot and mouth disease" obtained by the above method, and the proportion of the target object's visiting hospitals is 30%, and the probability value of "hand, foot and mouth disease" is 0.5, and the proportion of "bronchitis" in the target object's visiting hospitals is 10%, and the probability value of "bronchitis" is 0.2. Since the larger the proportion, the greater the adjustment of the disease probability value, the probability value of "hand, foot and mouth disease" is adjusted to 0.7, and the probability value of "bronchitis" is adjusted to 0.3.

[0089] It can be understood that in other embodiments, the disease-prone season information and the incidence probability rule information of hospitals of different levels can also be considered simultaneously, that is, the medical record text, personal history text, regional attribute information, disease rule information, incidence probability rule information of hospitals of different levels and the disease-prone season information can be considered simultaneously to obtain the probability values ​​of the target object suffering from several diseases respectively. In this case, steps S31-S32 and steps S41-S43 need to be executed.

[0090] Please refer to Figure 5-Figure 6 , Figure 5 is a flow chart of an embodiment of obtaining a diagnosis prediction result based on model prediction provided by the present application, Figure 6 is a schematic diagram of the structure of an embodiment of the diagnosis prediction model provided by the present application. It should be noted that if there are substantially the same results, this embodiment does not Figure 5 The process sequence shown is limited.

[0091] like Figure 5 As shown, in this embodiment, the diagnosis prediction result is obtained based on the model prediction, specifically including:

[0092] Step S51: Use the encoding network of the diagnosis prediction model to semantically encode the medical record text, personal history text, regional attribute information and disease rule information respectively to obtain a first semantic representation of the medical record text, a second semantic representation of the personal history text, a third semantic representation of the regional attribute information and a fourth semantic representation of the disease rule information.

[0093] In this embodiment, the encoding network of the diagnosis prediction model is used to semantically encode the medical record text, personal history text, regional attribute information and disease rule information respectively, to obtain a first semantic representation of the medical record text, a second semantic representation of the personal history text, a third semantic representation of the regional attribute information and a fourth semantic representation of the disease rule. The semantic encoding network may include, but is not limited to: BERT (Bidirectional Encoder Representation from Transformers), EMLO (Embedding from Language Model) model, etc., which is not limited here and can be specifically set according to actual use needs.

[0094] Specifically, Figure 6 As shown, first, the medical record text, personal history text, regional attribute information and disease rule information are used as the input layer of the diagnosis prediction model. The main input information includes the target object's gender, age, symptoms, examination and test results, etc. extracted from the medical record text, the target locations where the target object has lived and the length of residence in each target location, etc. extracted from the personal history text, the first disease prevalent in each target location and its severity, etc. extracted from the regional attribute information, and the various first disease's suffering rule information, etc. extracted from the disease rule information. Secondly, the encoding layer of the diagnosis prediction model quantizes the input information texts and inputs them into the BERT model to obtain the first semantic representation of the corresponding medical record text, the second semantic representation of the corresponding personal history text, the third semantic representation of the corresponding regional attribute information, and the fourth semantic representation of the corresponding disease rule information.

[0095] Step S52: using the interactive network of the diagnosis prediction model to perform information exchange among the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain the first importance, the second importance, the third importance and the fourth importance.

[0096] In this embodiment, the interactive network of the diagnosis prediction model is used to interact with the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation, thereby obtaining a first importance representing the importance of the personal history text to the medical record text, a second importance representing the importance of the regional attribute information to the medical record text, a third importance representing the importance of the disease rule information to the medical record text, and a fourth importance representing the importance of the regional attribute information to the disease rule information.

[0097] Specifically, Figure 6As shown, according to the correlation between each piece of information, an attention mechanism is adopted to interact the first semantic representation of the medical record text, the second semantic representation of the personal history text, the third semantic representation of the regional attribute information and the fourth semantic representation of the disease rule information, thereby obtaining a first importance representing the importance of the personal history text to the medical record text, a second importance representing the importance of the regional attribute information to the medical record text, a third importance representing the importance of the disease rule information to the medical record text and a fourth importance representing the importance of the regional attribute information to the disease rule information.

[0098] Step S53: Based on the first importance, the second importance and the third importance, the second semantic representation, the third semantic representation and the fourth semantic representation are weighted to obtain a first fused representation, and based on the fourth importance, the third semantic representation is weighted to obtain a second fused representation.

[0099] In this embodiment, the second semantic representation, the third semantic representation, and the fourth semantic representation are weighted according to the first importance indicating the importance of personal history text to medical record text, the second importance indicating the importance of regional attribute information to medical record text, and the third importance indicating the importance of disease rule information to medical record text to obtain a first fused representation, and the third semantic representation is weighted according to the fourth importance indicating the importance of regional attribute information to disease rule information to obtain a second fused representation.

[0100] Step S54: using the prediction network of the diagnosis prediction model to perform diagnosis prediction on the first fusion representation and the second fusion representation to obtain a diagnosis prediction result.

[0101] In this embodiment, the prediction network of the diagnosis prediction model is used to perform diagnosis prediction on the first fusion representation and the second fusion representation to obtain a diagnosis prediction result. Specifically, Figure 6 As shown, the final expression of the degree of matching between the target object and various diseases in the input information, namely the first fusion representation and the second fusion representation, are passed through a fully connected layer to obtain the probability value of the target object suffering from the relevant disease (e.g., endemic disease) in the input information, i.e., the diagnosis prediction result.

[0102] In this embodiment, a diagnosis prediction is performed based on the medical record text, personal history text, regional attribute information and disease rules to obtain a diagnosis prediction result of the target object. That is to say, this application not only considers the target object's gender, age, symptoms, test examinations and other information in the medical record text, but also considers the target location and the residence time of the target location in the target object's personal history text, as well as the regional attribute information and disease rule information related to the personal history text, that is, by comprehensively considering the medical record text, personal history text, regional attribute information and disease rule information to generate a diagnosis prediction result, the obtained diagnosis prediction result of the target object is more in line with the real scene and more accurate.

[0103] The prediction network of the diagnostic prediction model uses the idea of ​​cross entropy LOSS calculation to train model parameters during the training process. During the training process, the probability value is directly used as the final output probability value, and the scores of different diseases are compared. The disease with the largest probability value is taken as the final diagnostic prediction result. The cross entropy LOSS formula is:

[0104]

[0105] Among them, y i represents the labeling result (0 / 1) of the ith training example; p represents the model prediction score of the ith training example, and n represents the number of training categories.

[0106] See also Figure 7 , Figure 7 FIG. 1 is a flow chart of another embodiment of the present application for obtaining a diagnosis prediction result based on a model prediction. It should be noted that if there are substantially the same results, this embodiment does not use Figure 7 The process sequence shown is limited. Figure 7 As shown, in this embodiment, the information of the season when the disease is prone to occur is taken into account at the same time, that is, the medical record text, personal history text, regional attribute information, disease rule information and disease prone season information are taken into account at the same time to obtain the probability values ​​of the target object suffering from several diseases respectively, specifically including:

[0107] Step S71: Use the encoding network of the diagnosis prediction model to semantically encode the medical record text, personal history text, regional attribute information and disease rule information respectively to obtain a first semantic representation of the medical record text, a second semantic representation of the personal history text, a third semantic representation of the regional attribute information and a fourth semantic representation of the disease rule information.

[0108] Step S71 is similar to step S51 and will not be described again here.

[0109] Step S72: using the encoding network to perform semantic encoding on the favorable sending rule information to obtain a fifth semantic representation of the favorable sending rule information.

[0110] In this embodiment, the coding network of the diagnosis prediction model is used to semantically encode the prone rule information to obtain the fifth semantic representation of the prone rule information. Among them, the prone rule information includes the prone time of the disease. The semantic coding network can include but is not limited to: BERT (Bidirectional Encoder Representation from Transformers, i.e., Encoder of bidirectional Transformer), EMLO (Embedding from Language Model) model, etc., which are not limited here and can be specifically set according to actual use needs.

[0111] Specifically, Figure 6 As shown in the figure, first, the prevalence rule information is used as the input layer of the diagnosis prediction model, and the main input information includes the prevalence time of different diseases. Secondly, the encoding layer of the diagnosis prediction model quantizes the input information text and inputs it into the BERT model to obtain the fifth semantic representation of the corresponding prevalence rule information.

[0112] Step S73: using the interactive network to exchange information among the first semantic representation, the second semantic representation, the third semantic representation, the fourth semantic representation and the fifth semantic representation to obtain the first importance, the second importance, the third importance, the fourth importance and the fifth importance, and the sixth importance.

[0113] In this embodiment, the interactive network of the diagnosis prediction model is used to interact with the first semantic representation, the second semantic representation, the third semantic representation, the fourth semantic representation and the fifth semantic representation to obtain a first importance indicating the importance of the personal history text to the medical record text, a second importance indicating the importance of the regional attribute information to the medical record text, a third importance indicating the importance of the disease rule information to the medical record text, a fourth importance indicating the importance of the regional attribute information to the disease rule information, a fifth importance indicating the importance of the prone rule information to the medical record text and a sixth importance indicating the importance of the medical record text to the prone rule information.

[0114] Specifically, Figure 6As shown, according to the correlation between each piece of information, the attention mechanism is adopted to interact the first semantic representation of the medical record text, the second semantic representation of the personal history text, the third semantic representation of the regional attribute information, the fourth semantic representation of the disease rule information, and the fifth semantic representation of the prone rule information, thereby obtaining a first importance representing the importance of the personal history text to the medical record text, a second importance representing the importance of the regional attribute information to the medical record text, a third importance representing the importance of the disease rule information to the medical record text, a fourth importance representing the importance of the regional attribute information to the disease rule information, a fifth importance representing the importance of the prone rule information to the medical record text, and a sixth importance representing the importance of the medical record text to the prone rule information.

[0115] Step S74: Based on the first importance, the second importance, the third importance and the fifth importance, the second semantic representation, the third semantic representation, the fourth semantic representation and the fifth semantic representation are weighted to obtain a first fused representation, and based on the fourth importance, the third semantic representation is weighted to obtain a second fused representation, and based on the sixth importance, the first semantic representation is weighted to obtain a third fused representation.

[0116] In this embodiment, the second semantic representation, the third semantic representation, the fourth semantic representation and the fifth semantic representation are weighted according to the first importance indicating the importance of personal history text to medical record text, the second importance indicating the importance of regional attribute information to medical record text, the third importance indicating the importance of disease rule information to medical record text and the fifth importance indicating the importance of prone rule information to medical record text to obtain a first fused representation; and the third semantic representation is weighted according to the fourth importance indicating the importance of regional attribute information to disease rule information to obtain a second fused representation; and the sixth importance indicating the importance of medical record text to prone rule information is weighted to obtain a third fused representation.

[0117] Step S75: Use the prediction network to perform diagnostic prediction on the first fusion representation, the second fusion representation, and the third fusion representation to obtain a diagnostic prediction result.

[0118] In this embodiment, the prediction network of the diagnosis prediction model is used to perform diagnosis prediction on the first fusion representation, the second fusion representation and the third fusion representation to obtain a diagnosis prediction result. Specifically, Figure 6 As shown, the final expressions of the degree of matching between the target object and various diseases in the input information, namely the first fusion representation, the second fusion representation and the third fusion representation, are passed through a fully connected layer to obtain the probability value of the target object suffering from the relevant diseases in the input information (for example, endemic diseases, diseases that coincide with the onset time, etc.), that is, the diagnosis prediction result.

[0119] In this embodiment, the information on the time when different diseases are prone to occur is taken into consideration at the same time. For example, hand, foot and mouth disease is more likely to break out in summer and autumn, and chronic obstructive pulmonary disease usually occurs in winter and spring. Therefore, by considering the information on the seasons when diseases are prone to occur at the same time, the probability value of each disease in the diagnosis prediction results can be adjusted, thereby dynamically adjusting the ranking of various diseases, thereby improving the accuracy of the diagnosis prediction results.

[0120] See also Figure 8 , Figure 8 FIG. 1 is a flow chart of another embodiment of the present application for obtaining a diagnosis prediction result based on a model prediction. It should be noted that if there are substantially the same results, this embodiment does not use Figure 8 The process sequence shown is limited. Figure 8 As shown, in this embodiment, the information of the incidence probability rules of hospitals of different levels is considered at the same time, that is, the medical record text, personal history text, regional attribute information, disease rule information and the incidence probability rule information of hospitals of different levels are considered at the same time to obtain the probability values ​​of the target object suffering from several diseases respectively, specifically including:

[0121] Step S81: Use the encoding network of the diagnosis prediction model to semantically encode the medical record text, personal history text, regional attribute information and disease rule information respectively to obtain a first semantic representation of the medical record text, a second semantic representation of the personal history text, a third semantic representation of the regional attribute information and a fourth semantic representation of the disease rule information.

[0122] Step S81 is similar to step S51 and will not be described again here.

[0123] Step S82: Based on the medical record text of the target object, obtain the medical institution where the target object has been treated.

[0124] Step S82 is similar to step S41 and will not be described again here.

[0125] Step S83: Obtain the diagnosis statistics of the medical institution.

[0126] Step S83 is similar to step S42 and will not be described again here.

[0127] Step S84: using the encoding network to perform semantic encoding on the diagnostic statistical information to obtain a sixth semantic representation of the diagnostic statistical information.

[0128] In this implementation, the diagnostic statistical information is semantically encoded using the encoding network of the diagnostic prediction model to obtain the sixth semantic representation of the diagnostic statistical information. The semantic encoding network may include, but is not limited to, BERT (Bidirectional Encoder Representation from Transformers), EMLO (Embedding from Language Model) models, etc., which are not limited here and can be specifically set according to actual use needs.

[0129] Specifically, Figure 6 As shown, first, the diagnostic statistical information is used as the input layer of the diagnostic prediction model. The main input information includes the proportion of each disease diagnosed in the total number of patients received by the medical institution. Secondly, the encoding layer of the diagnostic prediction model quantizes the input information text and inputs it into the BERT model to obtain the sixth semantic representation of the corresponding diagnostic statistical information.

[0130] Step S85: Use the interactive network to exchange information among the first semantic representation, the second semantic representation, the third semantic representation, the fourth semantic representation and the sixth semantic representation to obtain the first importance, the second importance, the third importance, the fourth importance and the seventh importance and the eighth importance.

[0131] In this embodiment, the interactive network of the diagnosis prediction model is used to interact with the first semantic representation, the second semantic representation, the third semantic representation, the fourth semantic representation and the sixth semantic representation to obtain a first importance representing the importance of personal history text to medical record text, a second importance representing the importance of regional attribute information to medical record text, a third importance representing the importance of disease rule information to medical record text, a fourth importance representing the importance of regional attribute information to disease rule information, a seventh importance representing the importance of diagnostic statistical information to medical record text, and an eighth importance representing the importance of personal history text to diagnostic statistical information.

[0132] Specifically, Figure 6As shown, according to the correlation between each piece of information, an attention mechanism is adopted to interact the first semantic representation of the medical record text, the second semantic representation of the personal history text, the third semantic representation of the regional attribute information, the fourth semantic representation of the disease rule information, and the sixth semantic representation of the diagnosis statistical information, thereby obtaining a first importance representing the importance of the personal history text to the medical record text, a second importance representing the importance of the regional attribute information to the medical record text, a third importance representing the importance of the disease rule information to the medical record text, a fourth importance representing the importance of the regional attribute information to the disease rule information, a seventh importance representing the importance of the diagnosis statistical information to the medical record text, and an eighth importance representing the importance of the personal history text to the diagnosis statistical information.

[0133] Step S86: Based on the first importance, the second importance, the third importance and the seventh importance, the second semantic representation, the third semantic representation, the fourth semantic representation and the sixth semantic representation are weighted to obtain a first fused representation, and based on the fourth importance, the third semantic representation is weighted to obtain a second fused representation, and based on the eighth importance, the second semantic representation is weighted to obtain a fourth fused representation.

[0134] In this embodiment, the second semantic representation, the third semantic representation, the fourth semantic representation and the sixth semantic representation are weighted according to the first importance indicating the importance of personal history text to medical record text, the second importance indicating the importance of regional attribute information to medical record text, the third importance indicating the importance of disease rule information to medical record text and the seventh importance indicating the importance of diagnosis statistical information to medical record text to obtain a first fused representation; and the third semantic representation is weighted according to the fourth importance indicating the importance of regional attribute information to disease rule information to obtain a second fused representation; and the eighth importance indicating the importance of personal history text to diagnosis statistical information is weighted to obtain a fourth fused representation.

[0135] Step S87: Use the prediction network to perform diagnostic prediction on the first fusion representation, the second fusion representation, and the fourth fusion representation to obtain a diagnostic prediction result.

[0136] In this embodiment, the prediction network of the diagnosis prediction model is used to perform diagnosis prediction on the first fusion representation, the second fusion representation and the fourth fusion representation to obtain a diagnosis prediction result. Specifically, Figure 6 As shown, the final expressions of the degree of matching between the target object and various diseases in the input information, namely the first fusion representation, the second fusion representation and the fourth fusion representation, are passed through a fully connected layer to obtain the probability value of the target object suffering from the relevant diseases in the input information (for example, endemic diseases, diseases related to the hospital visited, etc.), that is, the diagnosis prediction result.

[0137] In this embodiment, the incidence probability rule information of hospitals of different levels is considered at the same time. There are differences in the frequency of diseases in hospitals of different levels. For example, critical and severe diseases and rare diseases are more common in tertiary hospitals, while common diseases are more common in primary hospitals. For example, cough, sputum, and asthma may correspond to bronchitis, acute exacerbation of chronic obstructive pulmonary disease, etc. The probability of acute exacerbation of chronic obstructive pulmonary disease in tertiary hospitals is much higher than that of bronchitis, while the opposite is true in primary hospitals. Therefore, the probability values ​​of the target object suffering from several diseases obtained by taking into account the incidence probability rule information of hospitals of different levels can be more fully in line with the real scene and can improve the accuracy of the diagnosis prediction results.

[0138] It is understandable that in other embodiments, the disease-prone season information and the incidence probability rule information of hospitals of different levels can also be considered simultaneously to obtain the probability values ​​of the target object suffering from several diseases respectively. In this case, steps S71-S75 and steps S81-S87 need to be executed.

[0139] In other embodiments, the diagnosis prediction result is obtained based on rule matching and model prediction, and the diagnosis prediction result is a fusion of the diagnosis prediction result obtained based on rule matching and the diagnosis prediction result obtained based on model prediction.

[0140] In a specific embodiment, if there are no identical diseases in the diagnosis prediction results obtained based on rule matching and the diagnosis prediction results obtained based on model prediction, then the diagnosis prediction results are the diagnosis prediction results obtained based on rule matching and the diagnosis prediction results obtained based on model prediction. For example, the diagnosis prediction results obtained based on rule matching are: the probability value of "cretinism" is 0.1, the probability value of "schistosomiasis" is 0.4, the probability value of "plague" is 0.6, and the probability value of "senile osteoporosis" is 0.5; the diagnosis prediction results obtained based on model prediction are: the probability value of "septic arthritis" is 0.4, the probability value of "Kashin-Beck disease" is 0.3, the probability value of "acute cholecystitis" is 0.2, the probability value of "bile duct stones" is 0.5, the probability value of "upper respiratory tract infection" is 0.4, and the probability value of "tonsillitis" is 0.8; due to the diagnosis prediction results obtained based on rule matching There are no identical diseases in the diagnosis prediction results and the diagnosis prediction results based on the model prediction, so the final diagnosis prediction results are: the probability value of "cretinism" is 0.1, the probability value of "schistosomiasis" is 0.4, the probability value of "plague" is 0.6, the probability value of "senile osteoporosis" is 0.5, the probability value of "septic arthritis" is 0.4, the probability value of "Kashin-Beck disease" is 0.3, the probability value of "acute cholecystitis" is 0.2, the probability value of "bile duct stones" is 0.5, the probability value of "upper respiratory tract infection" is 0.4, and the probability value of "tonsillitis" is 0.8.

[0141] In another specific embodiment, there are some common diseases in the diagnosis prediction results obtained based on rule matching and the diagnosis prediction results obtained based on model prediction, so it is necessary to fuse the probability value of the disease obtained based on rule matching and the probability value obtained based on model prediction to obtain the final probability value corresponding to the disease. The fusion processing includes but is not limited to taking an average value, weighted summation, etc. For example, taking the fusion processing as an average value as an example, the diagnostic prediction results based on rule matching are: the probability value of "cretinism" is 0.1, and the probability value of "schistosomiasis" is 0.4; the diagnostic prediction results based on model prediction are: the probability value of "septic arthritis" is 0.4, and the probability value of "cretinism" is 0.2; since the various diseases in the diagnostic prediction results based on rule matching and the diagnostic prediction results based on model prediction are the same, the probability value of "cretinism" based on rule matching and the probability value of 0.1 and the probability value of 0.2 obtained based on model prediction are averaged to obtain the final probability value of "cretinism" of 0.15; therefore, the final diagnostic prediction results are: the probability value of "cretinism" is 0.15, the probability value of "schistosomiasis" is 0.4, and the probability value of "septic arthritis" is 0.4.

[0142] See also Fig. 9 , Fig. 9 It is a schematic diagram of the framework of an embodiment of a diagnosis prediction device provided by the present application. The diagnosis prediction device 90 includes: a first acquisition module 91, a second acquisition module 92 and a prediction module 93. The first acquisition module 91 is used to obtain the medical record text of the target object and the personal history text of the target object; wherein the personal history text includes the target locations where the target object has lived and the length of residence in each target location; the second acquisition module 92 is used to obtain the regional attribute information and disease rule information related to the personal history text; wherein the regional attribute information includes the first disease prevalent in each target location and its severity, the disease rule information includes the rules for the occurrence of various first diseases, and the occurrence rules include the mapping relationship between the length of residence in the target location where the first disease is prevalent and the possibility of suffering from the first disease; the prediction module 93 is used to perform diagnosis prediction based on the medical record text, the personal history text, the attribute information and the disease rule information to obtain the diagnosis prediction result of the target object; wherein the diagnosis prediction result includes the probability value of the target object suffering from several diseases, and the several diseases include the first disease.

[0143] Among them, the above-mentioned diagnostic prediction results are obtained based on rule matching, and the above-mentioned rule matching steps include: the prediction module 93 is used to obtain a first weight coefficient of the first disease based on the severity of the first disease prevalent in the target location, and obtain a second weight coefficient of the first disease based on the suffering rule; the first weight coefficient and the second weight coefficient are integrated to obtain a first probability value of the target object suffering from the first disease.

[0144] Among them, the diagnosis prediction device 90 also includes a combination module 94, which is used to obtain diagnosis rule information of several second diseases; wherein the diagnosis rule information includes the rule text of each second disease; based on the semantic similarity between the medical record text and the rule text of various second diseases, the second probability value of the target object suffering from various second diseases is obtained; the first probability value of the target object suffering from various first diseases and the second probability value of the target object suffering from various second diseases are combined to obtain the probability value of the target object suffering from various diseases.

[0145] Among them, the diagnosis and prediction device 90 also includes a first adjustment module 95, which is used to obtain the onset time of the target object based on the medical record text of the target object, and obtain the prevalence time of various diseases based on the prevalence rule information; in response to the existence of a disease whose prevalence time coincides with the onset time, the probability value of the disease is increased.

[0146] Among them, the diagnosis prediction device 90 also includes a second adjustment module 96, which is used to obtain the target object's medical institution based on the target object's medical record text; obtain the diagnosis statistical information of the medical institution; wherein the diagnosis statistical information includes the proportion of various diseases diagnosed in the total number of patients received by the medical institution; based on the proportion corresponding to the disease, adjust the probability value of the disease; wherein, the larger the proportion corresponding to the disease, the more positively correlated with the adjustment range of the probability value of the disease.

[0147] Among them, the above-mentioned diagnosis prediction result is obtained based on model prediction, and the steps of the above-mentioned model prediction include: the prediction module 93 is used to use the encoding network of the diagnosis prediction model to semantically encode the medical record text, personal history text, regional attribute information and disease rule information respectively, to obtain a first semantic representation of the medical record text, a second semantic representation of the personal history text, a third semantic representation of the regional attribute information and a fourth semantic representation of the disease rule information; the interactive network of the diagnosis prediction model is used to interact the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain a first importance, a second importance, a third importance and a fourth importance. Importance; wherein the first importance, the second importance and the third importance respectively represent the importance of the personal history text, the regional attribute information and the disease rule information to the medical record text, and the fourth importance represents the importance of the regional attribute information to the disease rule information; based on the first importance, the second importance and the third importance, the second semantic representation, the third semantic representation and the fourth semantic representation are weighted to obtain the first fusion representation, and based on the fourth importance, the third semantic representation is weighted to obtain the second fusion representation; the prediction network of the diagnosis prediction model is used to perform diagnostic prediction on the first fusion representation and the second fusion representation to obtain the diagnosis prediction result.

[0148] Among them, the prediction module 93 is used to use the interactive network of the diagnosis prediction model to exchange information between the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain the first importance, the second importance, the third importance and the fourth importance, and specifically also includes: using the encoding network to semantically encode the prone rule information to obtain the fifth semantic representation of the prone rule information; wherein the prone rule information includes the prone time of the disease; the prediction module 93 uses the interactive network of the diagnosis prediction model to exchange information between the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain the first importance, the second importance, the third importance and the fourth importance, and specifically includes: using the interactive network to exchange information between the first semantic representation, the second semantic representation, the third semantic representation, the fourth semantic representation and the fifth semantic representation to obtain the first importance, the second importance, the third importance, the fourth importance and the fifth importance, the sixth importance; wherein the fifth importance The first importance degree indicates the importance of the prone rule information to the medical record text, and the sixth importance degree indicates the importance of the medical record text to the prone rule information; before the prediction module 93 uses the prediction network of the diagnosis prediction model to perform diagnosis prediction on the first fusion representation and the second fusion representation to obtain the diagnosis prediction result, it specifically also includes: based on the first importance degree, the second importance degree, the third importance degree and the fifth importance degree, weighting the second semantic representation, the third semantic representation, the fourth semantic representation and the fifth semantic representation to obtain the first fusion representation, and based on the fourth importance degree, weighting the third semantic representation to obtain the second fusion representation, and weighting the first semantic representation based on the sixth importance degree to obtain the third fusion representation; the prediction module 93 uses the prediction network of the diagnosis prediction model to perform diagnosis prediction on the first fusion representation and the second fusion representation to obtain the diagnosis prediction result, which specifically includes: using the prediction network to perform diagnosis prediction on the first fusion representation, the second fusion representation and the third fusion representation to obtain the diagnosis prediction result.

[0149] Among them, the prediction module 93 is used to use the interactive network of the diagnosis prediction model to exchange information between the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain the first importance, the second importance, the third importance and the fourth importance, and specifically also includes: based on the medical record text of the target object, obtaining the medical institution of the target object; obtaining the diagnostic statistical information of the medical institution; wherein the diagnostic statistical information includes the proportion of each disease diagnosed in the total number of patients received by the medical institution; using the encoding network to semantically encode the diagnostic statistical information to obtain the sixth semantic representation of the diagnostic statistical information; the prediction module 93 is used to use the interactive network of the diagnosis prediction model to exchange information between the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain the first importance, the second importance, the third importance and the fourth importance, and specifically includes: using the interactive network to exchange information between the first semantic representation, the second semantic representation, the third semantic representation, the fourth semantic representation and the sixth semantic representation to obtain the first importance, the second importance, the third importance, The fourth importance, the seventh importance, and the eighth importance; wherein the seventh importance indicates the importance of diagnostic statistical information to the medical record text, and the eighth importance indicates the importance of the personal history text to the diagnostic statistical information; the prediction module 93 is used to perform diagnostic prediction on the first fusion representation and the second fusion representation using the prediction network of the diagnosis prediction model to obtain the diagnosis prediction result, and specifically also includes: based on the first importance, the second importance, the third importance and the seventh importance, the second semantic representation, the third semantic representation, the fourth semantic representation and the sixth semantic representation are weighted to obtain the first fusion representation, and based on the fourth importance, the third semantic representation is weighted to obtain the second fusion representation, and based on the eighth importance, the second semantic representation is weighted to obtain the third fusion representation; the prediction module 93 is used to perform diagnostic prediction on the first fusion representation and the second fusion representation using the prediction network of the diagnosis prediction model to obtain the diagnosis prediction result, and specifically includes: using the prediction network to perform diagnostic prediction on the first fusion representation, the second fusion representation and the fourth fusion representation to obtain the diagnosis prediction result.

[0150] Among them, the above-mentioned diagnosis prediction results are obtained based on rule matching and model prediction, and the diagnosis prediction results are obtained by integrating the diagnosis prediction results obtained based on rule matching and the diagnosis prediction results obtained based on model prediction.

[0151] See also Fig.10 , Fig.101 is a schematic diagram of a framework of an embodiment of an electronic device provided by the present application. The electronic device 100 includes a memory 101 and a processor 102 coupled to each other, the memory 101 stores program instructions, and the processor 102 is used to execute the program instructions to implement the steps in any of the above-mentioned diagnostic prediction method embodiments. Specifically, the electronic device 100 may include but is not limited to: a desktop computer, a laptop computer, a server, a mobile phone, a tablet computer, etc., which are not limited here.

[0152] Specifically, the processor 102 is used to control itself and the memory 101 to implement the steps in any of the above-mentioned diagnosis and prediction method embodiments. The processor 102 can also be called a CPU (Central Processing Unit). The processor 102 may be an integrated circuit chip with signal processing capabilities. The processor 102 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 102 can be implemented by an integrated circuit chip.

[0153] See also Fig.11 , Fig.11 The computer-readable storage medium 110 stores program instructions 111 that can be executed by a processor, and the program instructions 111 are used to implement the steps in any of the above-mentioned diagnosis and prediction method embodiments.

[0154] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0155] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0156] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0157] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0158] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0159] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0160] The above description is only an implementation method of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A diagnostic prediction method, characterized in that: include: Obtaining a medical record text of a target object and obtaining a personal history text of the target object; wherein the personal history text includes target locations where the target object has lived and the length of residence at each of the target locations; Acquire regional attribute information, disease rule information, prone rule information and diagnostic statistics of the target object's medical institution related to the personal history text; wherein the regional attribute information includes the first diseases prevalent in each of the target locations and their severity, the disease rule information includes the rules for suffering from various first diseases, and the suffering rules include a mapping relationship between the length of residence in the target location where the first disease is prevalent and the possibility of suffering from the first disease, the prone rule information includes the prone time of each of the first diseases, and the diagnostic statistics of the medical institution include the proportion of patients diagnosed with each of the first diseases in the total number of patients received by the medical institution; Based on the medical record text, the personal history text, the regional attribute information, the disease rule information, the predisposition rule information and the diagnostic statistical information of the target object's medical institution, a diagnostic prediction is performed to obtain a diagnostic prediction result of the target object; wherein the diagnostic prediction result includes probability values ​​of the target object suffering from several diseases respectively, and the several diseases include the first disease; The diagnosis prediction result is obtained based on model prediction, and the model prediction step includes: Using the encoding network of the diagnosis prediction model, the medical record text, the personal history text, the regional attribute information, and the disease rule information are semantically encoded respectively to obtain a first semantic representation of the medical record text, a second semantic representation of the personal history text, a third semantic representation of the regional attribute information, and a fourth semantic representation of the disease rule information; The interactive network of the diagnosis prediction model is used to interact the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain a first importance, a second importance, a third importance and a fourth importance; wherein the first importance, the second importance and the third importance respectively represent the importance of the personal history text, the regional attribute information and the disease rule information to the medical record text, and the fourth importance represents the importance of the regional attribute information to the disease rule information; Based on the first importance, the second importance and the third importance, weighting the second semantic representation, the third semantic representation and the fourth semantic representation to obtain a first fused representation, and based on the fourth importance, weighting the third semantic representation to obtain a second fused representation; The prediction network of the diagnosis prediction model is used to perform diagnosis prediction on the first fusion representation and the second fusion representation to obtain the diagnosis prediction result.

2. The method according to claim 1, characterized in that The diagnosis prediction result is obtained based on rule matching, and the rule matching step includes: Based on the severity of the first disease prevalent in the target location, a first weight coefficient of the first disease is obtained, and based on the affliction rule, a second weight coefficient of the first disease is obtained; The first weight coefficient and the second weight coefficient are integrated to obtain a first probability value that the target object suffers from the first disease.

3. The method according to claim 2, characterized in that The method further comprises: Acquire diagnosis rule information of a plurality of second diseases; wherein the diagnosis rule information includes rule texts of each of the second diseases; Based on the semantic similarities between the medical record text and the rule texts of various second diseases, obtaining second probability values ​​of the target object suffering from various second diseases; Combining the first probability values ​​of the target object suffering from the first diseases and the second probability values ​​of the target object suffering from the second diseases, obtaining the probability values ​​of the target object suffering from the diseases; Wherein, in response to the presence of a disease whose prevalence time coincides with the onset time, the probability value of the disease is increased; and, adjusting the probability value of the disease based on the proportion corresponding to the disease; Among them, the proportion corresponding to the disease is positively correlated with the adjustment range of the probability value of the disease.

4. The method according to claim 1, characterized in that: Before the interactive network using the diagnosis prediction model interacts the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain the first importance, the second importance, the third importance and the fourth importance, the method further includes: Using the encoding network to semantically encode the onset rule information, to obtain a fifth semantic representation of the onset rule information; wherein the onset rule information includes the onset time of the disease; The interactive network using the diagnosis prediction model performs information interaction on the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain a first importance, a second importance, a third importance and a fourth importance, including: The first semantic representation, the second semantic representation, the third semantic representation, the fourth semantic representation and the fifth semantic representation are interacted with each other using the interactive network to obtain the first importance, the second importance, the third importance, the fourth importance, the fifth importance and the sixth importance; wherein the fifth importance represents the importance of the favorable rule information to the medical record text, and the sixth importance represents the importance of the medical record text to the favorable rule information; Before performing diagnosis prediction on the first fused representation and the second fused representation using the prediction network of the diagnosis prediction model to obtain the diagnosis prediction result, the method further includes: Based on the first importance, the second importance, the third importance, and the fifth importance, the second semantic representation, the third semantic representation, the fourth semantic representation, and the fifth semantic representation are weighted to obtain the first fused representation, and based on the fourth importance, the third semantic representation is weighted to obtain the second fused representation, and based on the sixth importance, the first semantic representation is weighted to obtain a third fused representation; The using the prediction network of the diagnosis prediction model to perform diagnosis prediction on the first fusion representation and the second fusion representation to obtain the diagnosis prediction result includes: The prediction network is used to perform diagnosis prediction on the first fusion representation, the second fusion representation and the third fusion representation to obtain the diagnosis prediction result.

5. The method according to claim 1, characterized in that Before the interactive network using the diagnosis prediction model interacts the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain the first importance, the second importance, the third importance and the fourth importance, the method further includes: Based on the medical record text of the target object, obtaining the medical institution where the target object has been treated; Obtaining the diagnostic statistical information of the medical institution; wherein the diagnostic statistical information includes the proportion of patients diagnosed with each of the diseases in the total number of patients received by the medical institution; Using the encoding network to semantically encode the diagnostic statistical information to obtain a sixth semantic representation of the diagnostic statistical information; The interactive network using the diagnosis prediction model performs information interaction on the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain a first importance, a second importance, a third importance and a fourth importance, including: The first semantic representation, the second semantic representation, the third semantic representation, the fourth semantic representation and the sixth semantic representation are subjected to information exchange by using the interactive network to obtain the first importance, the second importance, the third importance, the fourth importance and the seventh importance and the eighth importance; wherein the seventh importance represents the importance of the diagnostic statistical information to the medical record text, and the eighth importance represents the importance of the personal history text to the diagnostic statistical information; Before performing diagnosis prediction on the first fused representation and the second fused representation using the prediction network of the diagnosis prediction model to obtain the diagnosis prediction result, the method further includes: Based on the first importance, the second importance, the third importance and the seventh importance, the second semantic representation, the third semantic representation, the fourth semantic representation and the sixth semantic representation are weighted to obtain the first fused representation, and based on the fourth importance, the third semantic representation is weighted to obtain the second fused representation, and based on the eighth importance, the second semantic representation is weighted to obtain a fourth fused representation; The using the prediction network of the diagnosis prediction model to perform diagnosis prediction on the first fusion representation and the second fusion representation to obtain the diagnosis prediction result includes: The prediction network is used to perform diagnosis prediction on the first fusion representation, the second fusion representation and the fourth fusion representation to obtain the diagnosis prediction result.

6. The method according to claim 1, characterized in that The diagnosis prediction result is obtained based on rule matching and model prediction, and the diagnosis prediction result is obtained by fusing the diagnosis prediction result obtained based on the rule matching and the diagnosis prediction result obtained based on the model prediction.

7. A diagnostic prediction device, characterized in that: include: A first acquisition module is used to acquire the medical record text of the target object and acquire the personal history text of the target object; wherein the personal history text includes the target locations where the target object has lived and the length of residence at each of the target locations; A second acquisition module is used to obtain regional attribute information, disease rule information, prone rule information and diagnostic statistics of the target object's medical institutions related to the personal history text; wherein the regional attribute information includes the first diseases prevalent in each of the target locations and their severity, the disease rule information includes the rules for suffering from various first diseases, and the suffering rules include a mapping relationship between the length of residence in the target location where the first disease is prevalent and the possibility of suffering from the first disease, the prone rule information includes the prone time of each of the first diseases, and the diagnostic statistics of the medical institutions include the proportion of patients diagnosed with each of the first diseases in the total number of patients received by the medical institutions; A prediction module, used for performing diagnosis prediction based on the medical record text, the personal history text, the regional attribute information, the disease rule information, the predisposition rule information and the diagnosis statistics information of the treatment institution of the target object, to obtain the diagnosis prediction result of the target object; wherein the diagnosis prediction result includes the probability values ​​of the target object suffering from several diseases respectively, and the several diseases include the first disease; The diagnosis prediction result is obtained based on model prediction, and the model prediction step includes: Using the encoding network of the diagnosis prediction model, the medical record text, the personal history text, the regional attribute information, and the disease rule information are semantically encoded respectively to obtain a first semantic representation of the medical record text, a second semantic representation of the personal history text, a third semantic representation of the regional attribute information, and a fourth semantic representation of the disease rule information; The interactive network of the diagnosis prediction model is used to interact the first semantic representation, the second semantic representation, the third semantic representation and the fourth semantic representation to obtain a first importance, a second importance, a third importance and a fourth importance; wherein the first importance, the second importance and the third importance respectively represent the importance of the personal history text, the regional attribute information and the disease rule information to the medical record text, and the fourth importance represents the importance of the regional attribute information to the disease rule information; Based on the first importance, the second importance and the third importance, weighting the second semantic representation, the third semantic representation and the fourth semantic representation to obtain a first fused representation, and based on the fourth importance, weighting the third semantic representation to obtain a second fused representation; The prediction network of the diagnosis prediction model is used to perform diagnosis prediction on the first fusion representation and the second fusion representation to obtain the diagnosis prediction result.

8. An electronic device, characterized in that: It comprises a memory and a processor coupled to each other, wherein the memory stores program instructions, and the processor is used to execute the program instructions to implement the diagnosis and prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: Program instructions that can be executed by a processor are stored, and the program instructions are used to implement the diagnosis and prediction method according to any one of claims 1 to 6.

Images