Medical data processing method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application disclose a medical data processing method and device, electronic equipment and storage medium; including: obtaining a plurality of state names determined in the current text information; grouping the plurality of state names according to the attribute information corresponding to each state name, obtaining a first number of state groups; scoring each state group according to the state name included in each state group and the state description information, obtaining the score of each state group; from the first number of state groups, obtain the target state group with the highest score, the target state group includes the name of at least one candidate state; according to the pre-set selection condition, the severity of each candidate state and the resource consumption degree of each candidate state, select a set number of target state names from the name of at least one candidate state. The present application can determine the target state name from the plurality of state names more quickly, reducing the consumption of manpower.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent medical treatment, in particular to a medical data processing method and device, electronic equipment and storage medium. BACKGROUND

[0002] In daily work, there is often a need to filter out a target professional and technical name from a plurality of professional and technical names included in text information. The professional and technical name is a name with high technical or professional threshold, such as a crime in law, a disease name in medicine, etc.

[0003] In the prior art, when filtering out a target professional and technical name from a plurality of professional and technical names, a person with relevant professional and technical knowledge often needs to spend a lot of time and effort to filter out the target professional and technical name, which consumes a lot of manpower. SUMMARY

[0004] The embodiments of the present application provide a medical data processing method and device, electronic equipment and storage medium, which can improve the problem of consuming a lot of manpower in the prior art.

[0005] The embodiments of the present application provide a medical data processing method for filtering out a target state name from a plurality of state names included in current text information, the method comprising: obtaining a plurality of determined state names included in the current text information; grouping the plurality of state names according to attribute information corresponding to each state name to obtain a first number of state groups, each state group including at least one state name; scoring each state group according to the state names included in each state group and state description information, to obtain a score of each state group, wherein the state description information is recorded in the current text information, and the score is positively correlated with the severity of the state; obtaining a target state group with the highest score from the first number of state groups, the target state group including the names of at least one candidate state; and filtering out a set number of target state names from the names of the at least one candidate state according to a pre-set selection condition, the severity of each candidate state and the resource consumption degree of each candidate state.

[0006] The embodiments of the present application also provide a medical data processing device for filtering out a target state name from a plurality of state names included in current text information, the device comprising:

[0007] A state name obtaining unit is configured to obtain a plurality of determined state names included in the current text information.

[0008] The state grouping obtaining unit is configured to group the plurality of state names according to attribute information corresponding to each state name, to obtain a first number of state groups, each of which includes at least one state name.

[0009] The state grouping scoring unit is configured to score each state group according to state names included in the state group and state description information recorded in the current text information, to obtain a score of each state group, wherein the score is positively correlated with the severity of the state.

[0010] The high-score obtaining unit is configured to obtain a target state group with the highest score from the first number of state groups, the target state group including names of at least one candidate state.

[0011] The target state screening unit is configured to screen a set number of target state names from the names of the at least one candidate state according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state.

[0012] In some embodiments, the attribute information includes a human body part where the state is located, a repair site corresponding to the state, and a frequent item set where the state is located; and the state grouping obtaining unit includes:

[0013] The preliminary grouping obtaining sub-unit is configured to obtain a human body part where each state name is located, and group state names with the same human body part into the same group, to obtain a second number of preliminary groups.

[0014] The secondary grouping obtaining sub-unit is configured to, for any two preliminary groups in the second number of preliminary groups, combine the two preliminary groups into a secondary group if a state represented by a state name in one preliminary group and a state represented by a state name in the other preliminary group correspond to the same repair site, to obtain a third number of secondary groups.

[0015] The final grouping obtaining sub-unit is configured to, for any two secondary groups in the third number of secondary groups, combine the two secondary groups into a final group if a state represented by a state name in one secondary group and a state represented by a state name in the other secondary group are located in the same frequent item set, to obtain a first number of final groups, the final groups being the state groups.

[0016] In some embodiments, the state description information includes a fourth number of state description types and a fifth number of state repair types; and the state grouping scoring unit includes:

[0017] a first vector subunit, configured to perform encoding processing on a sixth number of state names included in the first state group to obtain a first vector corresponding to the first state group, the first state group being any one of the first number of state groups;

[0018] a type sub-vector subunit, configured to calculate, for each state description type, a type sub-vector corresponding to the state description type according to text data corresponding to the state description type and the first vector;

[0019] a second vector subunit, configured to concatenate a fourth number of type sub-vectors to obtain a second vector corresponding to the first state group;

[0020] a conditional probability subunit, configured to obtain, according to historical text information, a conditional probability corresponding to each state repair type when the sixth number of state names are respectively taken as target states;

[0021] a third vector subunit, configured to concatenate a fifth number of conditional probabilities to obtain a third vector corresponding to the first state group;

[0022] a feature vector subunit, configured to concatenate the first vector, the second vector, and the third vector to obtain a feature vector of the first state group;

[0023] a score obtaining subunit, configured to input the feature vector of the first state group into a full connection layer to obtain a score of the first state group.

[0024] In some embodiments, the first vector subunit includes:

[0025] a first encoding sub-subunit, configured to perform first encoding processing on each state name to obtain a state name vector corresponding to each state name;

[0026] a state group matrix sub-subunit, configured to combine the sixth number of state name vectors into a state group matrix;

[0027] a first fusion sub-subunit, configured to fuse, by using an attention mechanism, the sixth number of state name vectors and a first trainable parameter to obtain a first weight value vector, the first weight value vector including a weight value of each state name vector;

[0028] a first matrix operation sub-subunit, configured to perform matrix operation on the first weight value vector and the state group matrix to obtain the first vector.

[0029] In some embodiments, the first encoding subunit is specifically configured to encode a first state name in the sixth number of state names, and obtain m character vectors by encoding m characters included in the first state name, where the first state name is any state name in the sixth number of state names, and m is a positive integer; calculate a sum of the m character vectors to obtain a sum vector; and splice the sum vector and a comparison vector corresponding to the first state name to obtain a state name vector corresponding to the first state name.

[0030] In some embodiments, the type sub-vector subunit comprises:

[0031] The sentence vector subunit is configured to encode t sentences included in the text data corresponding to the first state description type to obtain t sentence vectors, where the first state description type is any state description type in the fourth number of state description types, and t is a positive integer.

[0032] The description matrix subunit is configured to combine the t sentence vectors into a first description matrix.

[0033] The second fusion subunit is configured to fuse the second trainable parameter, the first vector, and the first description matrix by using an attention mechanism to obtain a second weight value vector.

[0034] The second matrix operation subunit is configured to perform matrix operation on the second weight value vector and the first description matrix to obtain a type sub-vector corresponding to the first state description type.

[0035] In some embodiments, the conditional probability subunit comprises:

[0036] The entity acquisition subunit is configured to acquire r first state repair entities corresponding to the first state repair type from the current text information, where the first state repair type is any state repair type in the fifth number of state repair types, and r is a positive integer.

[0037] The conditional probability subunit is configured to calculate, for each first state repair entity, a conditional probability that the first state repair entity is implemented when each state name in the sixth number of state names is a target state.

[0038] The probability sum subunit is configured to calculate a sum of the conditional probabilities that each first state repair entity is implemented to obtain a conditional probability sum corresponding to each first state repair entity.

[0039] The adding subunit is configured to add the conditional probabilities corresponding to the r first state repair entities respectively to obtain an addition of the conditional probabilities of the first state repair type.

[0040] In some embodiments, the target state screening unit comprises:

[0041] The target state name screening subunit is configured to screen a seventh number of target state names from the target state groups according to the pre-set selection condition.

[0042] The first comparison subunit is configured to screen the set number of target state names from the seventh number of target state names according to the severity of each candidate state and the resource consumption degree of each candidate state when the seventh number is not less than the set number.

[0043] In some embodiments, the target state screening unit further comprises:

[0044] The second comparison subunit is configured to calculate a difference between the set number and the seventh number when the seventh number is less than the set number.

[0045] The remaining screening subunit is configured to screen a number of target state names corresponding to the difference from the remaining candidate states of the target state groups according to the severity of each candidate state and the resource consumption degree of each candidate state.

[0046] In the medical data processing method provided by the embodiments, a plurality of state names included in the current text information can be obtained; and the plurality of state names are grouped according to attribute information corresponding to each state name in the plurality of state names to obtain a first number of state groups, wherein each state group has one or more state names corresponding thereto. Then, each state group is scored according to the state names and state description information included in each state group, and a state group with the highest score is taken as a target state group. The one or more state names included in the target state group are candidate state names. Finally, the target state names are screened from the one or more candidate state names according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state, and the number of the target state names can be a set number.

[0047] In this application, multiple state names can first be grouped according to attribute information: those with the same attribute information are grouped together. Then, each group is scored, and the group with the highest score is obtained. Subsequently, the multiple state names included in the group with the highest score are filtered to obtain the target state name. Compared with the prior art, this application can quickly determine the target state name from multiple state names included in the current text information, reducing manpower consumption and improving filtering efficiency. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of a scenario for the medical data processing method provided in an embodiment of this application;

[0050] Figure 2 This is a schematic flowchart of a medical data processing method provided in an embodiment of this application;

[0051] Figure 3 A schematic diagram of a specific implementation of the data processing model in the application stage is shown;

[0052] Figure 4 It shows Figure 3 A schematic diagram illustrating a specific implementation of the text understanding layer in the application phase;

[0053] Figure 5 This is a schematic diagram of a medical data processing device provided in one embodiment of this application;

[0054] Figure 6 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0055] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] This application provides a medical data processing method, apparatus, electronic device, and storage medium.

[0057] The medical data processing apparatus can be integrated in an electronic device, which can be a terminal, a server, or the like. The terminal can be a mobile phone, a tablet computer, a smart Bluetooth device, a notebook computer, a personal computer (PC), or the like. The server can be a single server or a server cluster composed of multiple servers.

[0058] In some embodiments, the medical data processing apparatus can also be integrated in multiple electronic devices, for example, the medical data processing apparatus can be integrated in multiple servers to implement the medical data processing method of the present application.

[0059] In some embodiments, the server can also be implemented in the form of a terminal.

[0060] For example, the electronic device described above can perform the following method: obtaining a plurality of determined state names included in current text information; grouping the plurality of state names according to attribute information corresponding to each state name, to obtain a first number of state groups, each state group including at least one state name; scoring each state group according to the state names included in each state group and state description information, to obtain a score of each state group, wherein the state description information is recorded in the current text information, and the score is positively correlated with the severity of the state; obtaining a target state group with the highest score from the first number of state groups, the target state group including the names of at least one candidate state; and selecting a set number of target state names from the names of the at least one candidate state according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state.

[0061] For details, please refer to Figure 1 x state names are obtained from the current text information, and each of the x state names is included in the current text information. Then, the x state names are grouped according to the attribute information corresponding to each state name, to obtain y state groups, each state group including at least one state name. Each state group is scored according to the state names included in each state group and state description information, and a target state group with the highest score is obtained, which is state group 2 for example. All state names included in state group 2 are names of candidate states. Finally, a target state name is selected from the names of the candidate states according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state.

[0062] The medical data processing method provided in the embodiments of the present application can be used for screening names with high technical or professional threshold, for example, can be used for screening crimes that a criminal suspect may commit, can be used for screening disease names that a patient who is discharged from hospital or a patient who dies of illness has suffered from, and can be used for screening medical terms corresponding to sub-health states of ordinary people.

[0063] The following will be described in detail. It should be noted that the serial numbers of the following embodiments are not regarded as a limitation on the preferred order of the embodiments.

[0064] In the present embodiment, a medical data processing method is provided, as shown in Figure 2 The medical data processing method is applied to a terminal, and the specific process of the method can include the following steps 110 to 150.

[0065] 110. Obtain a plurality of state names determined in current text information.

[0066] The current text information is text information including a plurality of professional names, wherein the professional name is a name with a high technical or professional understanding threshold, and the professional name can be the state name. The state name is the name of a natural person in a certain determined state.

[0067] Alternatively, the natural person here can be a criminal suspect, and the current text information can be a judgment document text issued by a court for the criminal suspect; accordingly, the state name can be the crime name that the criminal suspect is sentenced to. For example, if the criminal suspect is sentenced to A crime and B crime, the criminal suspect is in the state of being sentenced to A crime and B crime, and A crime and B crime are the state names.

[0068] Alternatively, the natural person here can also be a sick patient, and the current text information can be a medical record text made by a hospital for the sick patient; accordingly, the state name can be the disease name diagnosed for the sick patient. For example, if the sick patient is diagnosed with X disease and Y disease, the sick patient is in the state of being diagnosed with X disease and Y disease, and X disease and Y disease are the state names.

[0069] 120. Group the plurality of state names according to attribute information corresponding to each state name, to obtain a first number of state groups, and each state group includes at least one state name.

[0070] The attribute information is information that can be extracted from the state name and can have the same point as other state names. If the state name is a crime name, the attribute information can be a classification of the crime according to the nature of the legal interest violated by the crime. If the state name is a disease name, the attribute information can be a body part of the disease corresponding to the disease name, a diagnosis and treatment department corresponding to the disease, and the like. The first number of number values can be a pre-set number of values, and the specific number of values should not be understood as a limitation of the present application.

[0071] For the convenience of description, the following will take the state name as a disease name as an example for further description.

[0072] Optionally, in a specific embodiment, the attribute information includes a body part of the state, a repair site corresponding to the state, and a frequent item set in which the state is located. Accordingly, step 120 can specifically include steps 121 to 123 as follows:

[0073] 121. Obtain the body part of the state corresponding to each of the state names, and group the state names with the same body part of the state into the same group to obtain a second number of initial selected groups.

[0074] The second number of number values can be a pre-set number of values, and the specific number of values should not be understood as a limitation of the present application.

[0075] Continuing the example above, the state corresponding to the state name can be a disease corresponding to the disease name, i.e., the disease names can be grouped according to the part of the human body where the disease occurs: the disease names corresponding to the diseases occurring in the same part of the human body are grouped into the same group, thereby obtaining a second number of initial selected groups.

[0076] 122. For any two of the second number of initial selected groups, if a state represented by a state name in one initial selected group and a state represented by a state name in another initial selected group correspond to the same repair site, the above two initial selected groups are combined into a secondary selected group to obtain a third number of secondary selected groups.

[0077] The third number of number values can also be a pre-set number of values, and the specific number of values should not be understood as a limitation of the present application.

[0078] Continuing the example above, the repair site corresponding to the state can be a diagnosis and treatment department corresponding to the disease. For the second number of initial selected groups, if there are two initial selected groups in which the diagnosis and treatment department corresponding to any state name in one initial selected group is the same as the diagnosis and treatment department corresponding to any state name in another initial selected group, the two initial selected groups can be combined into one group, which can be denoted as a secondary selected group.

[0079] For example, suppose that two initial groups are initial group a and initial group b, where initial group a includes state name a1, state name a2, state name a3, and state name a4, and initial group b includes state name b1, state name b2, state name b3, and state name b4. Suppose that the department corresponding to state name a2 is the same as the department corresponding to state name b4, then initial group a and initial group b can be combined into secondary group ab, which includes state name a1, state name a2, state name a3, state name a4, state name b1, state name b2, state name b3, and state name b4.

[0080] The processing of all initial groups can be completed by repeatedly performing step 122, so that the third number of secondary groups can be obtained. It should be understood that if there is an initial group, all state names in the initial group do not have the same department as any state name in any other initial group, then the initial group can be directly used as a secondary group.

[0081] 123、For any two secondary groups in the third number of secondary groups, if a state represented by a state name in one secondary group is in the same frequent item set as a state represented by a state name in another secondary group, then the two secondary groups are combined into a final group, obtaining the first number of final groups, and the final group is the state group.

[0082] A set of items is called an item set, and an item set including k items is called a k-item set. When determining whether an item set is a frequent item set, the support of the item set can be compared with the minimum support set in advance, and if the support exceeds the minimum support, the item set is a frequent item set.

[0083] The above example is continued to illustrate that the mining of frequent item sets can be performed using a plurality of historical medical record text information: the plurality of disease names diagnosed by each historical medical record text information is used as a piece of data, and the number of data pieces is the same as the number of historical medical record text information. Then, using the Apriori algorithm or FP-Growth, the frequent item set corresponding to each disease name in all disease names included in the plurality of historical medical record text information is determined.

[0084] For example, suppose that two sub-selected groups are sub-selected group c and sub-selected group d, wherein sub-selected group c includes state name cl, state name c2, state name c3, and state name c4; and sub-selected group d includes state name dl, state name d2, state name d3, and state name d4. Suppose that state name c2 and state name d3 correspond to the same frequent item set, then sub-selected group c and sub-selected group d can be merged into sub-selected group cd, which includes state name cl, state name c2, state name c3, state name c4, state name dl, state name d2, state name d3, and state name d4.

[0085] The processing of all sub-selected groups can be completed by repeatedly performing step 123, so that the first number of final selected groups can be obtained. It should be understood that if there is a sub-selected group, all state names in the sub-selected group do not belong to the same frequent item set with any state name in any other sub-selected group, then the sub-selected group can be directly used as a final selected group.

[0086] It should be understood that when grouping the state names, the above-mentioned steps 121 to 123 can be used, or other grouping methods can be used, for example, the state names can be directly grouped according to the body part where the disease is located, or the state names can be directly grouped according to the diagnosis and treatment department corresponding to the disease, or the state names can be directly grouped according to the frequent item set to which the disease belongs. In addition to the above-mentioned attribute information, other attribute information can also be used for grouping, for example, the state names can be grouped according to the human system where the disease is located (such as the urinary system, the digestive system, the nervous system, etc.); the state names can also be grouped according to the disease type (such as trauma, poisoning, obstetrics, etc.); the state names can also be grouped according to the core disease diagnosis related group (ARDG group) corresponding to the disease. The grouping method used for the state names and the attribute information relied on should not be understood as a limitation of the present application.

[0087] The target state name can be the name of the disease as the primary diagnosis. In some embodiments, the deep learning-based primary diagnosis prediction method is limited by the fact that different hospitals and different doctors have different understandings of the meaning of the primary diagnosis, resulting in inconsistent selection of the primary diagnosis in similar medical record texts, and resulting in poor migration effect of the model trained on medical record text data in different hospitals. In the above implementation, the state grouping and sorting idea is used, which can effectively solve the problem of inconsistent selection of the primary diagnosis in similar medical records caused by different understandings of the primary diagnosis by different hospitals and different doctors, and can more effectively utilize the medical record text data.

[0088] 130、score each of the state groups according to state names included in each of the state groups and state description information, to obtain a score of each of the state groups, wherein the state description information is recorded in the current text information, and the score is positively correlated with the severity of the state.

[0089] The state description information is text information recorded in the current text information and used to describe all the state names included in the current text information. The severity of the state refers to the negative impact of the state corresponding to the state name on the natural person corresponding to the state. The higher the score, the more severe the severity; the lower the score, the less severe the severity.

[0090] Optionally, in a specific embodiment, the state description information includes a fourth number of state description types and a fifth number of state repair types. Correspondingly, step 130 can specifically include steps 131 to 137 as follows:

[0091] The state description type refers to different angles for describing the state name, and the fourth number of state description types means that there are a fourth number of angles for describing the state name. The state repair type refers to the angle for repairing or remedying the state corresponding to the state name, and the fifth number of state repair types means that there are a fifth number of angles for repairing or remedying the state corresponding to the state name. The specific values of the fourth number and the fifth number should not be understood as a limitation of the present application.

[0092] Continuing the above example, it is assumed that the fourth number is 4, and the angles for describing the state name can include the following 4 angles: medical history, course record, superior doctor ward round record, and diagnosis and treatment process; it is assumed that the fifth number is also 4, and the angles for repairing or remedying the state corresponding to the state name can include the following 4 angles: diagnosis and treatment department, surgery, drug, and treatment principle.

[0093] For the first state group in the first number of state groups, the score of the first state group can be obtained through steps 131 to 137 as follows, wherein the first state group is any state group in the first number of state groups, and therefore, steps 131 to 137 are performed on each state group to obtain the scores corresponding to each state group respectively.

[0094] 131、encode the sixth number of state names included in the first state group to obtain a first vector corresponding to the first state group.

[0095] The specific value of the sixth number should not be understood as a limitation of the present application.

[0096] The first state group can include a sixth number of state names. The first state group can correspond to a vector obtained by encoding the sixth number of state names.

[0097] Optionally, in an embodiment, the step 131 can include steps 1311-1314.

[0098] 1311. Perform first encoding on each of the state names to obtain a state name vector corresponding to each of the state names.

[0099] The first encoding can be performed on each of the sixth number of state names to obtain a state name vector corresponding to each of the state names, and there are a total of sixth number of state name vectors.

[0100] Optionally, in an embodiment, the step 1311 can include steps S1-S3.

[0101] The steps S1-S3 can be performed on a first state name of the sixth number of state names. The first state name can be any of the sixth number of state names.

[0102] S1. Encode m characters included in the first state name to obtain m character vectors, where m is a positive integer.

[0103] The first state name includes a total of m characters. The m character vectors can be obtained by encoding the m characters respectively.

[0104] For details, see Figure 3 For example, the first state name is "hypertension", the value of m is 3, and the first state name includes three characters: "high", "blood", and "pressure". The three characters can be input into the word embedding layer shown in the figure to obtain three d-dimensional vectors, where d is a positive integer. Figure 3

[0105] S2. Calculate the sum of the m character vectors to obtain a sum vector.

[0106] Continuing the example, there are m d-dimensional vectors. The sum of the m d-dimensional vectors can be calculated to obtain a sum vector. The sum vector can be denoted as e i c where e i c ∈R d .

[0107] ​Optionally, if the m d-dimensional vectors can be column vectors, then adding the m d-dimensional column vectors column-wise will yield a summed column vector; if the m d-dimensional vectors can be row vectors, then adding the m d-dimensional row vectors row-wise will yield a summed row vector. It should be understood that the specific vector types of the m d-dimensional vectors should not be construed as a limitation of this application.

[0108] S3. Concatenate the summation vector with the comparison vector corresponding to the first state name to obtain the state name vector corresponding to the first state name.

[0109] The comparison vector corresponding to the first state name is randomly generated, and its value can be continuously updated during the training process. The dimension of the comparison vector is the same as the dimension of the summation vector, which is d-dimensional. Let's denote the comparison vector as e. i p e i p ∈R d The summation vector e i c Comparison vector e i p After concatenation, we can obtain the state name vector e. i d , where e i d ∈R 2d .

[0110] 1312. Combine the sixth number of state name vectors into a state grouping matrix.

[0111] It should be understood that the specific value of the sixth quantity should not be construed as a limitation on this application. Let's denote the sixth quantity as n, then the n state name vectors e i d The combined state grouping matrix is ​​e d , where e d ∈R n×2d .

[0112] 1313. Using an attention mechanism, the sixth number of state name vectors are fused with the first trainable parameters to obtain a first weight value vector, wherein the first weight value vector includes the weight value of each state name vector.

[0113] The first trainable parameter is randomly set and will be continuously updated during training. Let's denote the first trainable parameter as H, where H∈R. 2d×1 .

[0114] Specifically, att can be used 0 =softmax(e d H) Calculate the first weight vector att0 wherein att 0 ∈ R n ×1 ; the first weight value vector att 0 includes a weight value of each of the n state name vectors.

[0115] 1314. Perform matrix operation on the first weight value vector and the state grouping matrix to obtain the first vector.

[0116] Specifically, the transpose of the first weight value vector can be matrix operated with the state grouping matrix to obtain the first vector e g = att 0T e d The first vector e g wherein e g ∈ R 2d .

[0117] The steps 1312 to 1314 described above can be performed in the group information fusion layer shown. Figure 3

[0118] 132. For each state description type, calculate a type sub-vector corresponding to each state description type according to the text data corresponding to each state description type and the first vector.

[0119] Continuing with the example mentioned above, it is mentioned that the number of state description types is four, namely: present illness history, course record, superior physician ward round record and diagnosis and treatment process; therefore, the type sub-vectors obtained in step 132 are also four, corresponding to the present illness history, course record, superior physician ward round record and diagnosis and treatment process respectively.

[0120] Optionally, in a specific embodiment, step 132 can include steps 1321 to 1324 as follows:

[0121] For a first state description type in the fourth number of state description types, wherein the first state description type can be any state description type in the fourth number of state description types, i.e. the first state description type can be any one of the present illness history, course record, superior physician ward round record and diagnosis and treatment process; for the sake of convenience, the case where the first state description type is the present illness history is taken as an example for illustration.

[0122] 1321. Encode t sentences included in the text data corresponding to the first state description type to obtain t sentence vectors, wherein t is a positive integer.

[0123] ​The current medical history corresponds to t pieces of text data in the current text information. Specifically, the manner of determining that the current medical history corresponds to t pieces of text data can be as follows:

[0124] According to “.”, the text data corresponding to the current medical history is divided into several sentences, and the number of sentences is counted. The number value is t, so it can be determined that the text data corresponding to the current medical history has t pieces of text data.

[0125] Optionally, the textCNN model can be used to encode each of the t pieces of text data, thereby obtaining a d-dimensional sentence vector e i 现病史s , wherein e i 现病史s ∈R d , e i 现病史s represents the sentence vector corresponding to the i-th sentence.

[0126] 1322, the t pieces of sentence vectors are combined into a first description matrix.

[0127] Each sentence corresponds to a sentence vector e i 现病史s , and there are t pieces of sentence vectors, so e 现病史s can be obtained, wherein e 现病史s ∈R t×d .

[0128] Using the hierarchical CNN shown in Figure 4 to encode e 现病史s , a first description matrix e 现病史s' can be obtained, wherein e 现病史s' ∈R t×d .

[0129] 1323, using the attention mechanism, the second trainable parameter, the first vector and the first description matrix are fused to obtain a second weight value vector.

[0130] The second trainable parameter is a randomly set parameter that will be updated with training. The second trainable parameter can be denoted as W, wherein W ∈R 2d×d .

[0131] Specifically, the second weight value vector att 1.现病史 can be calculated using att g = softmax(e 现病史s'T We 1.现病史 ), wherein att 1.现病史 ∈R 1×t . The second weight value vector att 1.现病史The meaning is: for each statement in the t statements of the medical history, obtain each statement and the first vector e g The weight values of the relationship of the corresponding state group, wherein each weight value is the second weight value vector att 1.现病史 The score of the corresponding position of this t-dimensional vector.

[0132] 1324, matrix operation is performed on the second weight value vector and the first description matrix to obtain a type sub-vector corresponding to the first state description type.

[0133] Specifically, the second weight value vector can be matrix operated with the first description matrix to obtain e 现病史 1. 现病史 e 现病史s' The type sub-vector e 现病史 , wherein e 现病史 ∈R d .

[0134] For the medical record, the attending physician's ward round record and the diagnosis and treatment process, the same calculation method as steps 1321 to 1324 can be used to obtain the corresponding type sub-vector e 病程记录 , e 上级医生查房记录 , and e 诊疗经过 , wherein e 病程记录 ∈R d , e 上级医生查房记录 ∈R d , and e 诊疗经过 ∈R d ; the calculation process is the same as steps 1321 to 1324, and will not be repeated here.

[0135] 133, the fourth number of type sub-vectors are spliced to obtain a second vector corresponding to the first state group.

[0136] After obtaining the fourth number of type sub-vectors, these type sub-vectors can be spliced to obtain a second vector. Continue to illustrate the above example, the second vector is e u = [e 现病史 ; e 病程记录 ; e 上级医生查房记录 ; e 诊疗经过 ], wherein e u ∈R 4d .

[0137] Steps 132 to 133 can be implemented by the text understanding system shown in the embodiment. Figure 3

[0138] ​​134、According to the historical text information, the sixth number of state names are obtained as target states, and each state repair type corresponds to a conditional probability of each state repair type.

[0139] The historical text information is text information that is close to the form of the current text information and is earlier in time. For example, if the current text information is a current medical record text, the corresponding historical text information is a historical medical record text. The target state is a state that has the most negative impact on a natural person. For example, if the state name is a disease name, the corresponding target disease name is the name of the disease that causes the most harm to the human body. If the state name is a crime name, the corresponding target crime name is the name of the most serious crime committed by the natural person.

[0140] Optionally, in a specific embodiment, step 134 can specifically include steps 1341-1344 as follows:

[0141] The following steps 1341-1344 can be performed for each of the first state repair types in the fifth number of state repair types, and the first state repair type is any of the fifth number of state repair types.

[0142] Continuing the example above, let's assume that the fifth number of state repair types is four, namely the diagnosis and treatment department, surgery, medication, and treatment principle. The first state repair type can be any of the four, and let's assume that the first state repair type is the diagnosis and treatment department.

[0143] 1341、From the current text information, r first state repair entities corresponding to the first state repair type are obtained, where r is a positive integer.

[0144] The first state repair entity is an entity that can repair the state reflected by the state name.

[0145] Continuing the example above, if the first state repair type corresponds to the diagnosis and treatment department, the first state repair entity is a specific department, i.e., all departments mentioned in the current text information. Let's assume that the number of all departments mentioned in the current text information is r, where r is a positive integer, and the r first state repair entities include department 1, department 2, …, and department r.

[0146] 1342、For each of the first state repair entities, the conditional probability of the first state repair entity being implemented when each of the sixth number of state names is a target state is calculated.

[0147] Let's assume that the sixth number is six, and the state name is a disease name. Continuing the example above:

[0148] When disease 1 is the target disease, the conditional probability of department i being registered is calculated according to the statistical data of historical medical record texts, where department i is one of the r departments mentioned above.

[0149] When disease 2 is the target disease, the conditional probability of department i being registered is calculated according to the statistical data of historical medical record texts.

[0150] When disease 3 is the target disease, the conditional probability of department i being registered is calculated according to the statistical data of historical medical record texts.

[0151] When disease 6 is the target disease, the conditional probability of department i being registered is calculated according to the statistical data of historical medical record texts.

[0152] In the above manner, the sixth number of conditional probabilities corresponding to department i can be calculated.

[0153] 1343、Calculate the sum of the conditional probability of each said first state repair entity being implemented, to obtain the conditional probability corresponding to each said first state repair entity respectively.

[0154] After obtaining the sixth number of conditional probabilities corresponding to department i, the sum of the sixth number of conditional probabilities can be calculated, and the obtained conditional probability sum is the conditional probability sum corresponding to department i.

[0155] 1344、Calculate the sum of the conditional probability sum corresponding to the r said first state repair entities respectively, to obtain the conditional probability of the first state repair type.

[0156] Since there are a total of r departments, after calculating the conditional probability sum corresponding to each department (i.e., r conditional probability sums) according to step 1343, the sum of the r conditional probability sums also needs to be further calculated, and at this time the obtained is the conditional probability p dep .

[0157] The above steps 1341 to 1344 can also be represented by the formula p dep =∑ i ∑ j p dep diag＝i,dep＝j , where p dep diag＝i,dep＝j represents the correlation probability between a certain disease i and the department j registered in the historical medical record.

[0158] For surgery, drugs, and treatment principles, the above steps 1341 to 1344 can be performed to obtain the conditional probabilities p operate , p drug , and ptreatment . Calculate p operate , p drug , p treatment The process of calculating p p , p dep , p operate , p drug corresponds to steps 1341-1344, which will not be repeated here.

[0159] 135. Concatenate the fifth number of conditional probabilities to obtain a third vector corresponding to the first state group.

[0160] After obtaining the fifth number of conditional probabilities, these conditional probabilities can be concatenated to obtain a third vector. Continue to illustrate the above example, the third vector is e p = [p dep ; p operate ; p drug ; p treatment ], wherein e p ∈ R 4 .

[0161] 136. Concatenate the first vector, the second vector and the third vector to obtain the feature vector of the first state group.

[0162] Concatenating the first vector e g , the second vector e u and the third vector e p , the feature vector of the first state group e f = [e g ; e u ; e p ] can be obtained, wherein e f ∈ R 6d+4 . Steps 135-136 can be implemented in the feature fusion layer shown in Figure 3 .

[0163] 137. Input the feature vector of the first state group into a fully connected layer to obtain the score of the first state group.

[0164] After calculating the feature vector e f of the first state group, inputting the feature vector into the multi-layer fully connected shown in Figure 3 , the score of the first state group can be obtained, which is the score 1 in Figure 3 .

[0165] Please refer to Figure 3 , Figure 3 , the first number of state groups is 4, that is, there are 4 groups of state groups, so steps 131-137 can be performed for the other three groups, and the scores corresponding to the other three groups can be obtained: score 2, score 3, score 4.

[0166] 140、from the first number of state groups, obtain a target state group with the highest score, and the target state group includes the name of at least one candidate state.

[0167] The first number of state groups each have their own corresponding score, a total of first number of scores can be obtained, then the first number of scores are compared to obtain the highest score, and the state group corresponding to the highest score is recorded as the target state group. The target state group includes the name of one or more candidate states.

[0168] 150、According to the pre-set selection condition, the severity of each candidate state and the resource consumption degree of each candidate state, a set number of target state names are selected from the name of at least one candidate state.

[0169] The pre-set selection condition can be a conditional adverbial selection condition, for example, if XX occurs, select the name of ZZ state in the name of the candidate state as the target state name. The severity refers to the degree of negative impact of the candidate state on the natural person, and the resource consumption degree refers to the consumption degree of manpower and material resources including the natural person. The target state name can be the name of the state that has the most negative impact on the natural person. The value of the set number is a positive integer, and the specific value should not be understood as a limitation of the present application.

[0170] The above example continues to illustrate that, in the case where the name of the candidate state is the name of the candidate disease, the pre-set selection condition is the target state name selection rule under a plurality of scenes agreed in advance; for example, the obstetric delivery medical record should select obstetric complications as the main diagnosis; the first admission of malignant tumor should select malignant tumor as the main diagnosis, and the second admission of malignant tumor should select chemotherapy as the main diagnosis, etc. In real-time above the pre-set selection condition, it can be realized by a trained semantic model, or it can be realized by text pattern matching or regular expression.

[0171] The target state name is the name of the main diagnosis, wherein the main diagnosis refers to the most serious, highest treatment cost and longest time disease.

[0172] The severity of the candidate state is the criticality of the candidate disease, which can be calculated according to the formula The criticality P of the candidate disease d is calculated according to the formula severe (d), wherein #(severe, d) represents the number of medical records recorded in the historical medical record text in which the patient is critically ill when the candidate disease d is the main diagnosis; the critically ill identifier includes "at risk of death", "death", "extremely critically ill", etc. #(d) represents the total number of times when the candidate disease d is the main diagnosis.

[0173] The resource consumption degree of the candidate state is the general resource consumption degree of the candidate disease, which can be calculated according to the formula The time consumption of the candidate disease d can be calculated according to the formula Wherein, # (days, d) represents the hospitalization time of the patient recorded in the historical medical record text when the candidate disease d is the main diagnosis.

[0174] The cost consumption of the candidate disease d can be calculated according to the formula Wherein, # (cost, d) represents the cost of the patient recorded in the historical medical record text when the candidate disease d is the main diagnosis.

[0175] Optionally, in a specific embodiment, step 150 can specifically include steps 151 to 154 as follows:

[0176] 151. According to the pre-set selection condition, the seventh number of target state names are selected from the target state group.

[0177] The seventh number is a natural number, and the specific value of the seventh number should not be understood as a limitation of the present application.

[0178] 152. If the seventh number is not less than the set number, the set number of target state names are selected from the seventh number of target state names according to the severity of each candidate state and the resource consumption degree of each candidate state.

[0179] The seventh number of target state names can be selected according to the pre-set selection condition first, and then the seventh number and the set number are compared. If the seventh number is greater than or equal to the set number, it means that the seventh number of target state names can be further selected to obtain the set number of target state names.

[0180] The further selection can be realized according to the severity of each candidate state and the resource consumption degree of each candidate state. For example, the score of the severity and the score of the resource consumption degree can be calculated, and then the score sum corresponding to each candidate state is obtained, and then the score sum is arranged in descending order, and the names of the first set number of candidate states are selected as the set number of target state names.

[0181] 153. If the seventh number is less than the set number, the difference between the set number and the seventh number is calculated.

[0182] 154. According to the severity of each candidate state and the resource consumption degree of each candidate state, the number of target state names corresponding to the difference value is selected from the remaining candidate states in the target state group.​

[0183] If the seventh quantity is less than the set quantity, it indicates that the quantity selected by the pre-set selection condition cannot meet the quantity requirement of the target state name, and therefore, a difference between the set quantity and the seventh quantity can be calculated, and for the candidate states remaining in the target state group, a quantity value corresponding to the difference is selected from the target state names according to the severity of each candidate state and the resource consumption degree of each candidate state.

[0184] In the above embodiment, the pre-set selection condition can be used as a higher priority, so that the consumption of physical resources by the screening work can be reduced and the efficiency can be improved in the case of meeting the screening requirements. The in-group sorting according to the pre-set selection condition, the severity of the disease, and the resource consumption degree is more in line with the main diagnosis selection rule and is less likely to have a large error.

[0185] In the medical data processing method provided by the embodiment of the present application, a plurality of state names included in the current text information can be obtained; and the plurality of state names are grouped according to attribute information corresponding to each state name in the plurality of state names to obtain a first quantity of state groups, wherein each state group has one or more state names corresponding thereto. Then, each state group is scored according to the state names and state description information included in each state group, and the state group with the highest score is taken as a target state group. The one or more state names included in the target state group are candidate state names. Finally, the target state names are selected from the one or more candidate state names according to the pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state, and the quantity of the target state names can be a set quantity.

[0186] In the present application, the plurality of state names can be grouped according to the attribute information first: the same attribute information is classified into the same group. Then, the groups are scored to obtain the group with the highest score. Then, the plurality of state names included in the group with the highest score are screened to obtain the target state names. Compared with the prior art, the present application can quickly determine the target state names from the plurality of state names included in the current text information, reduce the consumption of manpower, and improve the screening efficiency. Moreover, the sorting method based on text information understanding proposed in the present application makes the model pay attention to the potential information in the historical text information and the current text information, which is conducive to exerting the ability of the deep model.

[0187] To better implement the above method, an embodiment of the present application further provides a medical data processing apparatus, which can be integrated in an electronic device, such as a terminal, a server, and the like. The terminal can be a mobile phone, a tablet computer, a smart Bluetooth device, a notebook computer, or a personal computer (PC), and the like. The server can be a single server or a server cluster composed of multiple servers. For example, as shown in FIG. 8, the medical data processing apparatus can include: Figure 5

[0188] a state name acquisition unit 501 configured to acquire a plurality of determined state names included in current text information;

[0189] a state grouping acquisition unit 502 configured to group the plurality of state names according to attribute information corresponding to each of the state names, to obtain a first number of state groupings, each of the state groupings including at least one state name;

[0190] a state grouping scoring unit 503 configured to score each of the state groupings according to state names included in each of the state groupings and state description information, to obtain a score of each of the state groupings, wherein the state description information is recorded in the current text information, and the score is positively correlated with the severity of the state;

[0191] a high score acquisition unit 504 configured to acquire a target state grouping with the highest score from the first number of state groupings, the target state grouping including names of at least one candidate state;

[0192] a target state screening unit 505 configured to screen a set number of target state names from the names of the at least one candidate state according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state.

[0193] In some embodiments, the attribute information includes a human body part where the state is located, a repair site corresponding to the state, and a frequent item set where the state is located. The state grouping acquisition unit 502 includes:

[0194] a preliminary grouping acquisition subunit configured to acquire a human body part where each of the state names is located, and group state names with the same human body part into the same group to obtain a second number of preliminary groupings;

[0195] ​a second selected group obtaining subunit, configured to, for any two of the second quantity of first selected groups, if there is a state represented by a state name in one of the first selected groups corresponding to a state represented by a state name in another of the first selected groups being the same repair site, merge the above two first selected groups into a second selected group, to obtain a third quantity of second selected groups;

[0196] a final selected group obtaining subunit, configured to, for any two of the third quantity of second selected groups, if there is a state represented by a state name in one of the second selected groups being the same frequent item set as a state represented by a state name in another of the second selected groups, merge the above two second selected groups into a final selected group, to obtain a first quantity of final selected groups, the final selected group being the state group.

[0197] In some embodiments, the state description information includes a fourth quantity of state description types and a fifth quantity of state repair types; the state group scoring unit 503 includes:

[0198] a first vector subunit, configured to perform encoding processing on a sixth quantity of state names included in the first state group, to obtain a first vector corresponding to the first state group, wherein the first state group is any state group of the first quantity of state groups;

[0199] a type subvector subunit, configured to, for each of the state description types, calculate a type subvector corresponding to each of the state description types according to the text data corresponding to each of the state description types and the first vector;

[0200] a second vector subunit, configured to splice the fourth quantity of type subvectors to obtain a second vector corresponding to the first state group;

[0201] a conditional probability subunit, configured to, according to historical text information, obtain respective conditional probabilities of each of the state repair types corresponding to each of the sixth quantity of state names as a target state;

[0202] a third vector subunit, configured to splice the fifth quantity of conditional probabilities to obtain a third vector corresponding to the first state group;

[0203] a feature vector subunit, configured to splice the first vector, the second vector, and the third vector to obtain a feature vector of the first state group;

[0204] a score obtaining subunit, configured to input the feature vector of the first state group into a fully connected layer to obtain a score of the first state group.

[0205] In some embodiments, the first vector subunit includes:

[0206] a first encoding subunit, configured to perform first encoding processing on each of the state names to obtain a state name vector corresponding to each of the state names;

[0207] a state grouping matrix subunit, configured to combine the sixth number of state name vectors into a state grouping matrix;

[0208] a first fusion subunit, configured to fuse the sixth number of state name vectors and first trainable parameters by using an attention mechanism to obtain a first weight value vector, the first weight value vector including a weight value of each of the state name vectors;

[0209] a first matrix operation subunit, configured to perform matrix operation on the first weight value vector and the state grouping matrix to obtain the first vector.

[0210] In some embodiments, the first encoding subunit is specifically configured to perform encoding on m characters included in a first state name in the sixth number of state names to obtain m character vectors, where the first state name is any of the sixth number of state names, and m is a positive integer; calculate a sum of the m character vectors to obtain a sum vector; and splice the sum vector and a comparison vector corresponding to the first state name to obtain a state name vector corresponding to the first state name.

[0211] In some embodiments, the type sub-vector unit includes:

[0212] a sentence vector subunit, configured to perform encoding on t sentences included in the text data corresponding to the first state description type to obtain t sentence vectors, where the first state description type is any of the fourth number of state description types, and t is a positive integer;

[0213] a description matrix subunit, configured to combine the t sentence vectors into a first description matrix;

[0214] a second fusion subunit, configured to fuse the second trainable parameters, the first vector, and the first description matrix by using an attention mechanism to obtain a second weight value vector;

[0215] a second matrix operation subunit, configured to perform matrix operation on the second weight value vector and the first description matrix to obtain a type sub-vector corresponding to the first state description type.

[0216] In some embodiments, the conditional probability subunit includes:

[0217] The entity obtaining subunit is configured to obtain r first status repair entities corresponding to the first status repair type from the current text information, where the first status repair type is any one of the fifth number of status repair types, and r is a positive integer.

[0218] The conditional probability subunit is configured to calculate, for each of the first status repair entities, a conditional probability that the first status repair entity is implemented when each of the sixth number of status names is taken as a target status.

[0219] The probability summing subunit is configured to sum the conditional probabilities that each of the first status repair entities is implemented to obtain a conditional probability sum corresponding to each of the first status repair entities.

[0220] The summing calculation subunit is configured to sum the conditional probability sums corresponding to the r first status repair entities to obtain a conditional probability of the first status repair type.

[0221] In some embodiments, the target status screening unit 505 includes:

[0222] The target status name screening subunit is configured to screen, according to the pre-set selection condition, a seventh number of target status names from the target status group.

[0223] The first comparison subunit is configured to screen, when the seventh number is not less than the set number, the set number of target status names from the seventh number of target status names according to the severity of each candidate status and the resource consumption degree of each candidate status.

[0224] In some embodiments, the target status screening unit further includes:

[0225] The second comparison subunit is configured to calculate, when the seventh number is less than the set number, a difference between the set number and the seventh number.

[0226] The remaining screening subunit is configured to screen, according to the severity of each candidate status and the resource consumption degree of each candidate status, a number of target status names corresponding to the difference from the remaining candidate status in the target status group.

[0227] In implementation, each of the above units can be implemented as an independent entity, or can be combined as the same or several entities. The implementation of each of the above units can be referred to the method embodiments above, and will not be described here.

[0228] The medical data processing method provided in the embodiments of the present application can first group a plurality of state names according to attribute information: the same attribute information is classified into the same group. Then, scoring is performed in units of groups to obtain a group with the highest score. Then, the plurality of state names included in the group with the highest score are screened, so that the target state name is obtained.

[0229] In the present application, the screening efficiency of the target state name can be improved.

[0230] In one specific embodiment of the present application, the state name is a disease name, the target state name is the name of a disease that is identified as the main diagnosis, and the current text information is the complete medical record of a patient and all discharge diagnoses corresponding to the medical record. Using the method provided in the present application, after providing the complete medical record of the patient and all discharge diagnoses corresponding to the medical record, the method can be used to predict and select a set number of most likely main diagnoses from the provided complete medical record and all discharge diagnoses corresponding to the medical record. The method provided in the present application is based on an existing deep learning method framework for main diagnosis selection, but overcomes three main problems existing in the framework: 1) Different doctors have different understandings of the principles of main diagnosis selection in medical record data. Therefore, for similar medical records, different doctors may select different main diagnoses, which causes difficulty in learning, and because different doctors in different hospitals have different understandings of the main diagnosis, the generalization performance between different hospitals is poor. 2) Directly comparing and sorting each disease separately makes it difficult to compare some diseases of the same type, for example, the description of the symptoms and treatment of cerebral hemorrhage and cerebral hernia in the medical record may be similar, which makes it difficult for the model to distinguish between the two, which increases the difficulty of the model in learning the comparison relationship between diseases. 3) Only using a probability method to measure the relationship between diseases and treatment processes has a single dimension, which achieves better results.

[0231] The embodiments of the present application propose using the state grouping and sorting idea, which can effectively solve the problem of inconsistent selection of main diagnoses for similar medical records caused by different understandings of main diagnoses by different doctors in different hospitals; can also improve the problem of comparing and sorting each disease separately, which makes it difficult to compare some diseases of the same type; grouping first, then selecting the group with the highest score, and then screening the plurality of disease names in the group with the highest score can improve the problem of measuring the relationship between diseases and treatment processes only by a probability method, which has a single dimension.

[0232] The embodiments of the present application can more effectively utilize medical record text data.

[0233] This application also provides an electronic device, which can be a terminal, a server, or other similar device. The terminal can be a mobile phone, tablet computer, smart Bluetooth device, laptop computer, personal computer, etc.; the server can be a single server or a server cluster composed of multiple servers, etc.

[0234] In some embodiments, the medical data processing device may also be integrated into multiple electronic devices, such as multiple servers, with the medical data processing method of this application being implemented by the multiple servers.

[0235] In this embodiment, the electronic device will be used as an example for detailed description, such as... Figure 6 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:

[0236] The electronic device may include components such as a processor 601 with one or more processing cores, a memory 602 with one or more computer-readable storage media, a power supply 603, an input module 604, and a communication module 605. Those skilled in the art will understand that... Figure 6 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0237] The processor 601 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 602, and by calling data stored in the memory 602, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. In some embodiments, the processor 601 may include one or more processing cores; in some embodiments, the processor 601 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor 601.

[0238] The memory 602 can be used to store software programs and modules, and the processor 601 can execute various function applications and data processing by running the software programs and modules stored in the memory 602. The memory 602 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the electronic device, etc. In addition, the memory 602 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device. Accordingly, the memory 602 can also include a memory controller to provide the processor 601 with access to the memory 602.

[0239] The electronic device also includes a power supply 603 for powering the various components, and in some embodiments, the power supply 603 can be logically connected to the processor 601 through a power management system, so that the power management system can be used to manage charging, discharging, and power consumption management, etc. The power supply 603 can also include one or more direct current or alternating current power supplies, recharging systems, power failure detection circuits, power converters or inverters, power status indicators, and the like.

[0240] The electronic device can also include an input module 604, which can be used to receive input digital or character information, and to generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0241] The electronic device can also include a communication module 605, which in some embodiments can include a wireless module, and the electronic device can use the wireless module of the communication module 605 for short-range wireless transmission, thereby providing the user with wireless broadband Internet access. For example, the communication module 605 can be used to help the user send and receive emails, browse web pages, and access streaming media, etc.

[0242] Although not shown, the electronic device can also include a display unit, etc., which will not be described here. In particular, in the present embodiment, the processor 601 in the electronic device will load one or more application program processes corresponding to the executable files into the memory 602 according to the following instructions, and the processor 601 will run the application programs stored in the memory 602, thereby implementing various functions, as follows:

[0243] obtaining a plurality of determined state names included in current text information; grouping the plurality of state names according to attribute information corresponding to each of the state names, to obtain a first number of state groups, each of the state groups including at least one state name; scoring each of the state groups according to state names included in the state group and state description information, to obtain a score of each of the state groups, wherein the state description information is recorded in the current text information, and the score is positively correlated with the severity of the state; obtaining a target state group with the highest score from the first number of state groups, the target state group including names of at least one candidate state; and filtering a set number of target state names from the names of the at least one candidate state according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state.

[0244] The specific implementation of each operation can be referred to the foregoing embodiments, which will not be described herein.

[0245] Those skilled in the art can understand that all or part of the steps in the various methods of the foregoing embodiments can be completed by instructions, or by related hardware controlled by the instructions, which can be stored in a computer readable storage medium and loaded and executed by a processor.

[0246] To this end, the embodiment of the present application provides a computer readable storage medium, which stores a plurality of instructions. The instructions can be loaded by a processor to execute steps in any medical data processing method provided by the embodiment of the present application. For example, the instructions can execute the following steps:

[0247] obtaining a plurality of determined state names included in current text information; grouping the plurality of state names according to attribute information corresponding to each of the state names, to obtain a first number of state groups, each of the state groups including at least one state name; scoring each of the state groups according to state names included in the state group and state description information, to obtain a score of each of the state groups, wherein the state description information is recorded in the current text information, and the score is positively correlated with the severity of the state; obtaining a target state group with the highest score from the first number of state groups, the target state group including names of at least one candidate state; and filtering a set number of target state names from the names of the at least one candidate state according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state.

[0248] The storage medium can include a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or the like.

[0249] According to an aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the method provided in any of the optional implementation manners provided in the above embodiments.

[0250] Due to the instructions stored in the storage medium, the steps of any of the medical data processing methods provided in the embodiments of the present application can be executed, thus the beneficial effects of any of the medical data processing methods provided in the embodiments of the present application can be achieved, which are described in detail in the above embodiments and will not be repeated here.

[0251] The above describes in detail a medical data processing method, device, electronic device and computer readable storage medium provided in the embodiments of the present application. The principles and implementation manners of the present application are described by applying specific examples. The above embodiment descriptions are only used to help understand the method of the present application and its core idea. Meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges can be changed. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A medical data processing method, characterized by, A method for screening a target state name from a plurality of state names included in current text information, the method comprising: obtaining a plurality of determined state names included in the current text information; grouping the plurality of state names according to attribute information corresponding to each of the state names to obtain a first number of state groups, each of the state groups including at least one state name; scoring each of the state groups according to state names included in each of the state groups and state description information recorded in the current text information, to obtain a score of each of the state groups, wherein the score is positively correlated with the severity of the state, and the severity of the state refers to the negative impact of the state corresponding to the state name on the natural person corresponding to the state; obtaining a target state group with the highest score from the first number of state groups, the target state group including at least one candidate state name; screening a set number of target state names from the at least one candidate state name according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state.

2. The method of claim 1, wherein, The attribute information includes a body part where the state is located, a repair site corresponding to the state, and a frequent item set where the state is located. The attribute information includes a body part where the state is located, a repair site corresponding to the state, and a frequent item set where the state is located. The grouping of the plurality of state names according to attribute information corresponding to each of the state names to obtain a first number of state groups comprises: obtaining a body part where each of the state names is located, and grouping state names with the same body part into the same group to obtain a second number of preliminary groups; for any two of the second number of preliminary groups, if a state represented by a state name in one preliminary group and a state represented by a state name in another preliminary group correspond to the same repair site, then the two preliminary groups are combined into a secondary group to obtain a third number of secondary groups; 3. The method of claim 1, wherein, for any two of the third number of secondary groups, if a state represented by a state name in one secondary group and a state represented by a state name in another secondary group are in the same frequent item set, then the two secondary groups are combined into a final group to obtain a first number of final groups, and the final group is the state group. The state description information includes a fourth number of state description types and a fifth number of state repair types, and the state repair type refers to the angle of repairing or remedying the state corresponding to the state name. The scoring of each of the state groups according to state names included in each of the state groups and state description information recorded in the current text information to obtain a score of each of the state groups comprises: for a first state group in the first number of state groups, encoding a sixth number of state names included in the first state group to obtain a first vector corresponding to the first state group, wherein the first state group is any state group in the first number of state groups; According to the text data corresponding to each state description type and the first vector, a type sub-vector corresponding to each state description type is calculated; The fourth number of type sub-vectors are spliced to obtain a second vector corresponding to the first state group; According to the historical text information, the sixth number of state names are obtained as the target state, and the respective conditions of each state repair type are obtained. The fifth number of conditional probabilities are spliced to obtain a third vector corresponding to the first state group. The first vector, the second vector, and the third vector are spliced to obtain a feature vector of the first state group. The first state group is input into a fully connected layer to obtain a score of the first state group.

4. The method of claim 3, wherein, The first state group includes a sixth number of state names, and the first state group corresponds to a first vector, which includes: Each state name is subjected to first encoding processing to obtain a state name vector corresponding to each state name. Sixth number of state name vectors are combined into a state group matrix. The sixth number of state name vectors and the first trainable parameter are fused using an attention mechanism to obtain a first weight value vector, which includes a weight value of each state name vector. The first weight value vector and the state group matrix are subjected to matrix operation to obtain the first vector.

5. The method of claim 4, wherein, The first state name in the sixth number of state names, The m characters included in the first state name are respectively encoded to obtain m character vectors, where the first state name is any state name in the sixth number of state names, and m is a positive integer. The sum of the m character vectors is calculated to obtain a sum vector. The sum vector and the comparison vector corresponding to the first state name are spliced to obtain a state name vector corresponding to the first state name. The first state description type in the fourth number of state description types, 6. The method of claim 3, wherein, The t sentences included in the text data corresponding to the first state description type are respectively encoded to obtain t sentence vectors, where the first state description type is any state description type in the fourth number of state description types, and t is a positive integer. The t sentence vectors are combined into a first description matrix. The second trainable parameter, the first vector, and the first description matrix are fused using an attention mechanism to obtain a second weight value vector. The second weight value vector and the first description matrix are subjected to matrix operation to obtain a type sub-vector corresponding to the first state description type. ​ ​ 7. The method of claim 3, wherein, The condition probability of each state repair type corresponding to each state name as a target state is obtained according to historical text information. A first state repair type in the fifth number of state repair types is obtained from the current text information, wherein the first state repair type is any state repair type in the fifth number of state repair types, and r is a positive integer. For each first state repair entity, the condition probability of the first state repair entity being implemented when each state name in the sixth number of state names is a target state is calculated. The sum of the condition probability of each first state repair entity being implemented is calculated to obtain the sum of the condition probability corresponding to each first state repair entity. The sum of the condition probability corresponding to each first state repair entity is calculated to obtain the condition probability of the first state repair type. The set number of target state names is selected from the names of the at least one candidate state according to a pre-set selection condition, the severity of each candidate state, and the resource consumption degree of each candidate state.

8. The method of claim 1, wherein, The seventh number of target state names is selected from the target state groups according to the pre-set selection condition. If the seventh number is not less than the set number, the set number of target state names is selected from the seventh number of target state names according to the severity of each candidate state and the resource consumption degree of each candidate state. After the seventh number of target state names is selected from the target state groups according to the pre-set selection condition, the method further comprises:

9. The method of claim 8, wherein, If the seventh number is less than the set number, the difference between the set number and the seventh number is calculated. The difference value corresponding to the number of target state names is selected from the remaining candidate states in the target state groups according to the severity of each candidate state and the resource consumption degree of each candidate state. The device for selecting target state names from a plurality of state names included in current text information comprises:

10. A medical data processing apparatus characterized by comprising: A state name acquisition unit is configured to acquire a plurality of determined state names included in current text information. A state group acquisition unit is configured to group the plurality of state names according to attribute information corresponding to each state name to obtain a first number of state groups, each of which includes at least one state name. A state group scoring unit is configured to score each state group according to the state names and state description information included in each state group, wherein the state description information is recorded in the current text information, and the score is positively correlated with the severity of the state, and the severity of the state refers to the negative impact of the state corresponding to the state name on the natural person corresponding to the state. ​ A high-score obtaining unit is configured to obtain a target state group with the highest score from the first number of state groups, the target state group including names of at least one candidate state; A target state screening unit is configured to screen a set number of target state names from the names of the at least one candidate state according to a pre-set selection condition, a severity of each candidate state, and a resource consumption degree of each candidate state.

11. An electronic device, comprising: A processor and a memory are included, and the memory stores a plurality of instructions; the processor loads the instructions from the memory to perform the steps in the medical data processing method according to any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a plurality of instructions, and the instructions are adapted to be loaded by a processor to perform the steps in the medical data processing method according to any one of claims 1-9.

Images