A data intelligent integrated management system

Abstract

The present invention relates to the technical field of medical data archiving, and specifically to a data intelligent integrated management system, including a data acquisition unit, a data processing unit, a text combination unit, a text classification unit, and a data archiving unit, specifically: the data acquisition unit is used to receive medical data uploaded by multiple preset data sources at every preset sampling period, and transmit the medical data to the data processing unit; the data processing unit is used to perform word segmentation operations on the medical data to obtain a medical word sequence corresponding to the medical data, and perform feature extraction on the medical word sequence. The present invention avoids errors and delays that may occur in manual processing through automated data acquisition, processing, and archiving processes, which is particularly important for data processing in the medical industry, because medical data is usually highly time-sensitive and complex, and automated processes can improve work efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data archiving, and in particular to an intelligent integrated data management system. Background Art

[0002] Medical data refers to information recorded in natural language that describes the patient's health status, diagnosis and treatment process, and treatment results. These text data are widely present in various medical documents, such as medical records, diagnosis reports, surgical records, doctors' and nurses' notes, examination reports, and drug prescriptions.

[0003] Traditional systems rely on manual data collection, processing, classification and archiving, which is not only susceptible to human errors, but also takes a long time to complete data processing. Especially when the amount of medical data is huge and changes frequently, manual operations can easily lead to delays in data processing, which in turn affects the timeliness and accuracy of medical services. Traditional systems usually rely on manual judgment when classifying and archiving data, which may result in inaccurate or inconsistent classification. Especially in the face of the complexity and diversity of medical data, manual classification is not only inefficient, but also prone to omission or incorrect classification of certain data. Traditional systems classify medical data only on the basis of raw medical data, but raw medical data often has no effective contextual associations. Simply classifying raw data often fails to uncover the deep information behind the data. Summary of the invention

[0004] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a data intelligent integrated management system.

[0005] The technical solution adopted to solve the above technical problems is: a data intelligent integrated management system, including a data acquisition unit, a data processing unit, a text combination unit, a text classification unit and a data archiving unit, specifically:

[0006] The data acquisition unit is used to receive medical data uploaded by multiple preset data sources at every preset sampling period, and transmit the medical data to the data processing unit;

[0007] The data processing unit is used to receive the medical data transmitted by the data acquisition unit, perform word segmentation on the medical data to obtain a medical word sequence corresponding to the medical data, perform feature extraction on the medical word sequence to obtain a word feature of each medical word in the medical word sequence, sort the medical word sequence based on the word feature to obtain a medical word sequence, and transmit the medical word sequence to the text combination unit;

[0008] The text combination unit is used to receive the medical word sequence transmitted by the data processing unit, construct a medical combination text based on the medical word sequence, and transmit the medical combination text to the text classification unit.

[0009] Preferably, the text classification unit is used to receive the medical combination text transmitted by the text combination unit, and perform text classification on the medical combination text based on a trained text classification model to obtain a text label corresponding to the medical combination text, that is, to obtain a data label corresponding to the medical data, and transmit the data label corresponding to the medical data to the data archiving unit.

[0010] Preferably, the data archiving unit is used to receive the data tags corresponding to the medical data transmitted by the text classification unit, and archive the medical data into preset medical archive databases of different types based on the data tags corresponding to the medical data.

[0011] Preferably, the formula for extracting the word features of the medical words is as follows:

[0012] ;

[0013] in, Indicates medical terms. Represents medical data, Medical terms In medical data The word features, Medical terms In medical data The number of occurrences in Medical terms The inverse medical data word frequency, represents the entropy calculation function, Indicates category The proportion of medical terms in all medical terms, Representing medical data Medical terms included The total number of Representing medical data The total number of Chinese medical terms.

[0014] Preferably, constructing a medical combination text based on the medical word sequence includes:

[0015] Obtaining a combined weight of each medical word in the medical word sequence based on the sequence number of each medical word in the medical word sequence;

[0016] Calculating the number of combination words required for each medical word based on the combination weight of each medical word in the medical word sequence;

[0017] Matching each medical term in the medical term sequence with a preset medical term knowledge base to obtain a medical term interpretation corresponding to each medical term in the medical term sequence;

[0018] The medical term explanation is intercepted based on the number of combination characters required for each medical term in the medical term sequence to obtain a standard medical term explanation corresponding to each medical term in the medical term sequence;

[0019] A medical combination text is constructed based on a medical word sequence and a standard medical word interpretation corresponding to each medical word in the medical word sequence.

[0020] Preferably, the calculation formula for the number of combined words is as follows:

[0021] ;

[0022] in, Indicates the first The number of words required for a medical term. represents the sum of the combined weights of all medical terms, Indicates the total word count of the medical portfolio text;

[0023] The calculation formula of the combined weight is as follows:

[0024] ;

[0025] in, Indicates the first The combined weight of medical terms, Indicates the sequential number of medical terms.

[0026] Preferably, the text classification model includes a semantic feature representation module, an attention module and a feature classification module; the semantic feature representation module is used to capture the semantic features of the input medical combination text through a BiLSTM network, and divide the semantic features into multiple feature blocks; the attention module is used to obtain a label-specific weighted feature representation at a label level for each feature block; the feature classification module is used to inject fine-grained features into coarse-grained labels, and output the final text label through a classifier.

[0027] Preferably, the semantic feature representation module is used to capture the semantic features of the input medical combination text through a BiLSTM network, and divide the semantic features into a plurality of feature blocks, including:

[0028] Performing vector representation on each medical word in the medical combination text by word embedding technology to obtain an embedding vector matrix;

[0029] Inputting the embedding vector matrix into a BiLSTM network, and outputting the concatenated vector of the hidden state of each embedding vector in two directions through the BiLSTM network to obtain a semantic feature matrix;

[0030] The semantic feature matrix is ​​segmented from a feature dimension to be segmented into a plurality of identical sub-semantic feature matrices, and each of the sub-semantic feature matrices corresponds to a granularity level.

[0031] Preferably, the feature classification module is used to inject fine-grained features into coarse-grained labels and output the final text labels through a classifier, including:

[0032] For each label-specific weighted feature representation, mean pooling is used for dimensionality reduction to obtain a feature mapping vector, and the label-specific feature matrix is ​​concatenated with the feature mapping vector to obtain a feature concatenation vector, which is input into the classifier to obtain the probability of the label. The probability calculation formula of the label is as follows:

[0033] ;

[0034] in, Indicates The probability of labels at the granular level, Indicates Granular single-layer feedforward network with bias, Indicates The feature map vector at the granular level, Represents a vector concatenation operation.

[0035] Preferably, the attention module is used to obtain a label-specific weighted feature representation at a label level for each of the feature blocks, comprising:

[0036] Multiple sub-semantic feature matrices are classified by multiple classifiers, and corresponding attention weights are added to the classification of each granularity level to obtain a label-specific weighted feature representation of the feature block at a label level. The attention weight calculation formula is as follows:

[0037] ;

[0038] in, Indicates Granularity-level label-specific feature matrix, Indicates Granular level attention weight matrix, Indicates The sub-semantic feature matrix at the granular level, and , represents the soft maximum activation function, Indicates The transformation matrix of each granularity level, Indicates Through the granularity level is a matrix of function activation vectors, and , Indicates The weight matrix of each granularity level, represents the hyperbolic tangent activation function.

[0039] The beneficial effects of the present invention are as follows: (1) The present invention avoids errors and delays that may occur in manual processing by automating the data collection, processing and archiving process. This is particularly important for data processing in the medical industry because medical data is usually highly time-sensitive and complex. The automated process can improve work efficiency and ensure that data is processed and archived in a timely and accurate manner. It can also perform steps such as word segmentation, feature extraction, sorting and classification on medical data, and can convert unstructured or semi-structured medical text data into a structured standard format. This structured data facilitates subsequent query, analysis and utilization, reducing the complexity of data processing; (2) The present invention can accurately obtain the standard interpretation of medical terms by matching with the preset medical term knowledge base, and effectively intercept and integrate medical terms, which makes medical data not just raw text information, but can be analyzed and reasoned at a higher level, thereby supporting medical decision-making, research and application. Moreover, through the trained text classification model, the system can automatically label the medical data according to the content of the medical data, thereby realizing personalized classification and archiving, which not only reduces the need for manual intervention, but also improves the accuracy and precision of data archiving, facilitates subsequent retrieval and utilization, and because standard interpretations are added to medical data, classification and archiving are more accurate; (3) The present invention can archive medical data into different types of medical archive databases according to data labels, optimizes the storage structure of medical data, and improves the efficiency of data management, so that medical institutions can quickly locate and retrieve different categories of medical data as needed, improving the accuracy and convenience of archive management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of the system architecture of an overall system in an embodiment of the present invention.

[0041] Figure numerals: 1. Data acquisition unit; 2. Data processing unit; 3. Text combination unit; 4. Text classification unit; 5. Data archiving unit. DETAILED DESCRIPTION

[0042] Embodiment 1, as Figure 1 As shown, the present invention proposes a data intelligent integrated management system, including a data acquisition unit 1, a data processing unit 2, a text combination unit 3, a text classification unit 4 and a data archiving unit 5, specifically:

[0043] The data acquisition unit 1 is used to receive medical data uploaded by multiple preset data sources at every preset sampling period, and transmit the medical data to the data processing unit 2;

[0044] The data processing unit 2 is used to receive the medical data transmitted by the data acquisition unit 1, and perform word segmentation operation on the medical data to obtain a medical word sequence corresponding to the medical data, perform feature extraction on the medical word sequence to obtain a word feature of each medical word in the medical word sequence, sort the medical word sequence based on the word feature to obtain a medical word sequence, and transmit the medical word sequence to the text combination unit 3;

[0045] The text combination unit 3 is used to receive the medical word sequence transmitted by the data processing unit 2, construct a medical combination text based on the medical word sequence, and transmit the medical combination text to the text classification unit 4.

[0046] In the present invention, the sampling period refers to the time interval at which the system automatically obtains data from multiple data sources. This period is set in advance, such as once every hour, every day or every week; the data source refers to the different sources from which the system obtains medical data, which can be various information systems of the hospital; the medical data obtained from the data source exists in the form of text, and the text data usually contains the doctor's diagnosis, the patient's symptom description, the treatment plan, the drug prescription, the examination report and other contents. Although these text data may contain some structured information (such as standardized drug names), in most cases, they are free-format texts containing natural language expressions; word segmentation refers to dividing continuous text into independent words or subwords. Word segmentation is a basic operation in natural language processing (NLP), especially in languages ​​without spaces such as Chinese. For example, "the patient's body temperature is 38 degrees and the blood pressure is low" can be segmented into "patient", "body temperature", "38 degrees", "blood pressure", "low".

[0047] In an optional embodiment, the text classification unit 4 is used to receive the medical combination text transmitted by the text combination unit 3, and perform text classification on the medical combination text based on a trained text classification model to obtain a text label corresponding to the medical combination text, that is, to obtain a data label corresponding to the medical data, and transmit the data label corresponding to the medical data to the data archiving unit 5.

[0048] It should be noted that the data label is the classification result output by the model, indicating which category the medical combination text belongs to. Depending on the task, the type of label is also different. Data labels include but are not limited to disease label text, treatment plan label text and symptom label text.

[0049] In an optional embodiment, the data archiving unit 5 is used to receive the data tags corresponding to the medical data transmitted by the text classification unit 4, and archive the medical data into preset medical archive databases of different types based on the data tags corresponding to the medical data.

[0050] It should be noted that different types of medical record databases refer to pre-set database types in medical institutions or medical information systems, which are used to store different types of medical data.

[0051] Embodiment 2, a data intelligent integrated management system proposed by the present invention, compared with embodiment 1, this embodiment further includes: the extraction formula of the word feature of the medical word is as follows:

[0052] ;

[0053] in, Indicates medical terms. Represents medical data, Medical terms In medical data The word features, Medical terms In medical data The number of occurrences in Medical terms The inverse medical data word frequency, represents the entropy calculation function, Indicates category The proportion of medical terms in all medical terms, Representing medical data Medical terms included The total number of Representing medical data The total number of Chinese medical terms.

[0054] In an optional embodiment, a medical combination text is constructed based on a sequential sequence of medical words, including:

[0055] Obtaining a combined weight of each medical word in the medical word sequence based on the sequence number of each medical word in the medical word sequence;

[0056] Calculating the number of combination words required for each medical word based on the combination weight of each medical word in the medical word sequence;

[0057] Matching each medical term in the medical term sequence with a preset medical term knowledge base to obtain a medical term explanation corresponding to each medical term in the medical term sequence;

[0058] The medical word explanation is intercepted based on the number of combined characters required for each medical word in the medical word sequence to obtain a standard medical word explanation corresponding to each medical word in the medical word sequence;

[0059] A medical composite text is constructed based on a medical word sequence and a standard medical word interpretation corresponding to each medical word in the medical word sequence.

[0060] It should be noted that the sequential sequence of medical terms refers to a list of medical terms arranged in order, and each medical term has a specific position (sequence number) in the sequence; the medical term explanation is a detailed description or definition of medical terms or terms, which usually include pathology, treatment methods, symptoms and other aspects; truncation refers to adjusting the length of the medical term explanation to ensure that the number of words in each term explanation meets the requirements. Usually, truncation is carried out on the basis of ensuring the completeness and accuracy of the explanation to avoid the explanation being too long or too short and to ensure that its information content is moderate; standard medical term explanation refers to standardized explanations extracted from the medical term knowledge base and in line with industry norms. These explanations are defined by medical experts or standard-setting organizations in related fields and stored and managed through a systematic knowledge base; medical composite text is the final text constructed based on the sequential sequence of medical terms and the standard explanation of each medical term. It organizes multiple medical terms and their corresponding explanations in a certain logical order to form a complete medical description.

[0061] In an optional embodiment, the calculation formula for the number of combined words is as follows:

[0062] ;

[0063] in, Indicates the first The number of words required for a medical term. represents the sum of the combined weights of all medical terms, Indicates the total word count of the medical portfolio text;

[0064] The formula for calculating the combined weight is as follows:

[0065] ;

[0066] in, Indicates the first The combined weight of medical terms, Indicates the sequential number of medical terms.

[0067] In an optional embodiment, the text classification model includes a semantic feature representation module, an attention module and a feature classification module; the semantic feature representation module is used to capture the semantic features of the input medical combination text through a BiLSTM network, and divide the semantic features into multiple feature blocks; the attention module is used to obtain a label-specific weighted feature representation at a label level for each feature block; the feature classification module is used to inject fine-grained features into coarse-grained labels, and output the final text label through a classifier.

[0068] It should be noted that the role of the semantic feature representation module is to convert the input text (here is the medical combination text) into a vector representation with semantic information. Specifically, the semantic feature representation module extracts useful semantic information from the text through certain models (such as BiLSTM). These semantic information are further converted into feature blocks for subsequent processing; Bidirectional long short-term memory network (BiLSTM) is an improved LSTM (long short-term memory) model, which can better capture the semantic features in the text by reading the context information of the text in both directions (from left to right and from right to left). LSTM itself is a neural network for sequence data (such as text, speech), which can maintain information in longer sequences and solve long-term dependency problems; The attention module is to enable the model to give different weighted feature representations according to different label requirements when processing each feature block; Coarse-grained labels refer to relatively macro labels, usually representing large categories or categories with strong generalization; Fine-grained features: refers to more specific and detailed information.

[0069] In an optional embodiment, the semantic feature representation module is used to capture the semantic features of the input medical combination text through a BiLSTM network, and divide the semantic features into multiple feature blocks, including:

[0070] Each medical word in the medical combination text is represented by a vector using word embedding technology to obtain an embedding vector matrix;

[0071] The embedding vector matrix is ​​input into the BiLSTM network, and the BiLSTM network outputs the concatenated vector of the hidden state of each embedding vector in two directions to obtain the semantic feature matrix;

[0072] The semantic feature matrix is ​​segmented from the feature dimension to be segmented into a plurality of identical sub-semantic feature matrices, and each sub-semantic feature matrix corresponds to a granularity level.

[0073] It should be noted that word embedding is a technology that maps words into dense vectors of fixed dimensions. Usually, such vectors can capture the semantic and contextual information of words. Word embedding enables computers to process semantic relationships in language by representing words as vectors. The embedding vector matrix is ​​a matrix composed of the embedding vectors of multiple words. Each row represents the vector representation of a word, so the entire matrix represents the embedding representation of all words in the input text. The semantic feature matrix is ​​a matrix composed of the concatenated vectors (the hidden state of each word) output by BiLSTM. Splitting refers to dividing the semantic feature matrix into multiple sub-matrices according to certain dimensions (usually feature dimensions). The semantic feature matrix is ​​divided into multiple sub-matrices according to the granularity level, and each sub-matrix contains semantic information of different granularities.

[0074] In an optional embodiment, the feature classification module is used to inject fine-grained features into coarse-grained labels and output the final text labels through a classifier, including:

[0075] For each label-specific weighted feature representation, mean pooling is used to reduce the dimension to obtain a feature mapping vector. The label-specific feature matrix is ​​concatenated with the feature mapping vector to obtain a feature concatenation vector. The feature concatenation vector is input into the classifier to obtain the probability of the label. The probability calculation formula of the label is as follows:

[0076] ;

[0077] in, Indicates The probability of labels at the granular level, Indicates Granular single-layer feedforward network with bias, Indicates The feature map vector at the granular level, Represents a vector concatenation operation.

[0078] It should be noted that the pooling operation is a downsampling technique used to reduce the dimension of the feature map. Here, mean pooling refers to averaging the values ​​of each area of ​​the input feature map to generate a lower dimensional feature vector. This process is usually used to reduce the spatial size of the data, thereby reducing computational complexity; the feature map vector refers to a low-dimensional representation of features extracted through certain operations (such as convolution, pooling, etc.) in the network; a single-layer feedforward network with bias is a simple neural network layer, usually including weights, biases, and activation functions. The bias term is a parameter in the neural network used to adjust the output. A feedforward network refers to a unidirectional flow of information in the network without involving loops.

[0079] In an optional embodiment, the attention module is used to obtain a label-specific weighted feature representation at a label level for each feature block, including:

[0080] Multiple sub-semantic feature matrices are classified by multiple classifiers, and corresponding attention weights are added to the classification of each granularity level to obtain a feature block to obtain a label-specific weighted feature representation at a label level. The attention weight calculation formula is as follows:

[0081] ;

[0082] in, Indicates Granularity-level label-specific feature matrix, Indicates Granular level attention weight matrix, Indicates The sub-semantic feature matrix at the granular level, and , represents the soft maximum activation function, Indicates The transformation matrix of each granularity level, Indicates Through the granularity level is a matrix of function activation vectors, and , Indicates The weight matrix of each granularity level, represents the hyperbolic tangent activation function.

[0083] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A data intelligent integrated management system, characterized in that: The system comprises a data collection unit (1), a data processing unit (2), a text combination unit (3), a text classification unit (4) and a data archiving unit (5), specifically: The data acquisition unit (1) is used to receive medical data uploaded by a plurality of preset data sources at every preset sampling period, and transmit the medical data to the data processing unit (2); The data processing unit (2) is used to receive the medical data transmitted by the data acquisition unit (1), perform word segmentation on the medical data to obtain a medical word sequence corresponding to the medical data, perform feature extraction on the medical word sequence to obtain a word feature of each medical word in the medical word sequence, sort the medical word sequence based on the word feature to obtain a medical word sequence, and transmit the medical word sequence to the text combination unit (3); The text combination unit (3) is used to receive the medical word sequence transmitted by the data processing unit (2), construct a medical combination text based on the medical word sequence, and transmit the medical combination text to the text classification unit (4); The text classification unit (4) is used to receive the medical combination text transmitted by the text combination unit (3), and to perform text classification on the medical combination text based on a trained text classification model to obtain a text label corresponding to the medical combination text, that is, to obtain a data label corresponding to the medical data, and to transmit the data label corresponding to the medical data to the data archiving unit (5); The data archiving unit (5) is used to receive the data labels corresponding to the medical data transmitted by the text classification unit (4), and archive the medical data into preset medical archive databases of different types based on the data labels corresponding to the medical data; The extraction formula of the word features of the medical words is as follows: ； in, Indicates medical terms. Represents medical data, Medical terms In medical data The word features, Medical terms In medical data The number of occurrences in Medical terms The inverse medical data word frequency, represents the entropy calculation function, Indicates category The proportion of medical terms in all medical terms, Representing medical data Medical terms included The total number of Representing medical data The total number of TCM terms; Constructing a medical combination text based on the medical word sequence, including: Obtaining a combined weight of each medical word in the medical word sequence based on the sequence number of each medical word in the medical word sequence; Calculating the number of combination words required for each medical word based on the combination weight of each medical word in the medical word sequence; Matching each medical term in the medical term sequence with a preset medical term knowledge base to obtain a medical term interpretation corresponding to each medical term in the medical term sequence; The medical term explanation is intercepted based on the number of combination characters required for each medical term in the medical term sequence to obtain a standard medical term explanation corresponding to each medical term in the medical term sequence; Constructing a medical combination text based on a medical word sequence and a standard medical word interpretation corresponding to each medical word in the medical word sequence; The text classification model includes a semantic feature representation module, an attention module and a feature classification module; the semantic feature representation module is used to capture the semantic features of the input medical combination text through a BiLSTM network, and divide the semantic features into multiple feature blocks; the attention module is used to obtain a label-specific weighted feature representation at a label level for each feature block; the feature classification module is used to inject fine-grained features into coarse-grained labels, and output the final text label through a classifier; The semantic feature representation module is used to capture the semantic features of the input medical combination text through the BiLSTM network, and divide the semantic features into multiple feature blocks, including: Performing vector representation on each medical word in the medical combination text by word embedding technology to obtain an embedding vector matrix; Inputting the embedding vector matrix into a BiLSTM network, and outputting the concatenated vector of the hidden state of each embedding vector in two directions through the BiLSTM network to obtain a semantic feature matrix; The semantic feature matrix is ​​divided from a feature dimension to divide the semantic feature matrix into a plurality of identical sub-semantic feature matrices, and each of the sub-semantic feature matrices corresponds to a granularity level; The feature classification module is used to inject fine-grained features into coarse-grained labels and output the final text labels through a classifier, including: For each label-specific weighted feature representation, mean pooling is used for dimensionality reduction to obtain a feature mapping vector, and the label-specific feature matrix is ​​concatenated with the feature mapping vector to obtain a feature concatenation vector, which is input into the classifier to obtain the probability of the label. The probability calculation formula of the label is as follows: ； in, Indicates The probability of labels at the granular level, Indicates Granular single-layer feedforward network with bias, Indicates The feature map vector at the granular level, Represents the concatenation operation of vectors; The attention module is used to obtain a label-specific weighted feature representation at a label level for each of the feature blocks, including: Multiple sub-semantic feature matrices are classified by multiple classifiers, and corresponding attention weights are added to the classification of each granularity level to obtain a label-specific weighted feature representation of the feature block at a label level. The attention weight calculation formula is as follows: ； in, Indicates Granularity-level label-specific feature matrix, Indicates Granular level attention weight matrix, Indicates The sub-semantic feature matrix at the granular level, and , represents the soft maximum activation function, Indicates The transformation matrix of each granularity level, Indicates Through the granularity level is a matrix of function activation vectors, and , Indicates The weight matrix of each granularity level, represents the hyperbolic tangent activation function.

2. According to claim 1, a data intelligent integrated management system is characterized in that: The calculation formula of the number of combined words is as follows: ； in, Indicates the first The number of words required for a medical term. represents the sum of the combined weights of all medical terms, Indicates the total word count of the medical portfolio text; The calculation formula of the combined weight is as follows: ； in, Indicates the first The combined weight of medical terms, Indicates the sequential number of medical terms.

Images