Medical data analysis model construction method, analysis method, equipment and medium

Abstract

The invention discloses a medical data analysis model construction method, an analysis method, a device and a medium, and the method comprises the steps: obtaining a training data set which comprises medical training data, and the classification labeling information and classification labeling information of the medical training data; inputting the medical training data into the constructed initial analysis model, performing hierarchical classification processing on the medical training data, and outputting an initial prediction result and a loss value corresponding to the medical training data; performing iterative training on the initial analysis model according to a set loss threshold value and a loss value to generate a target analysis model; the target analysis model is used for grading and classifying the input to-be-analyzed medical data and outputting a prediction result corresponding to the to-be-analyzed medical data. According to the method and the device, automatic and accurate grading and classification processing of the medical data is realized, grading and classification only through a manual labeling mode are avoided, the analysis efficiency and accuracy are improved, and powerful support is provided for management and safety protection of the medical data.

Description

Technical Field

[0001] The present application relates to the field of big data technology, and in particular to a medical data analysis model construction method, analysis method, equipment and medium. Background Art

[0002] With the popularization and rapid development of medical information technology, current medical data has the characteristics of large data volume, multiple data types, high value, low value density, fast generation and fast processing. A large amount of medical data can be graded and classified to fully explore its medical value, provide a basis for improving medical standards, and is of great significance to promoting medical progress.

[0003] At present, the grading and classification of medical data is mainly achieved through manual labeling or by using intelligent data grading and classification models in other non-medical scenarios.

[0004] However, with the expansion of business and data models, the amount and types of medical data have become very large, making the classification and grading method based on manual labeling costly, inefficient and inaccurate. In addition, the actual medical data involves complex scenarios, which makes its classification and grading standards highly professional, and the intelligent classification and grading models used in other non-medical scenarios cannot meet the professional needs of medical scenarios. Summary of the invention

[0005] The purpose of this application is to provide a medical data analysis model construction method, analysis method, equipment and medium to achieve accurate and automatic grading and classification of medical data.

[0006] To achieve the above objectives, this application provides the following solutions: In a first aspect, the present application provides a method for constructing a medical data analysis model, the method comprising: Acquire a training data set, wherein the training data set includes medical training data and hierarchical labeling information and classification labeling information of each medical training data; Input the medical training data into the constructed initial analysis model, perform grading and classification processing on the medical training data, and output the initial prediction result and loss value corresponding to the medical training data, wherein the initial prediction result includes the initial grading label and the initial classification label of the medical training data, and the loss value is determined according to the initial grading label, the initial classification label, the grading annotation information, and the classification annotation information; According to the set loss threshold and the loss value, the initial analysis model is iteratively trained to generate a target analysis model, so that when the medical data is input into the target analysis model and the target prediction result and the target loss value corresponding to the medical training data are output, the target loss value satisfies the set loss threshold; the target analysis model is used to grade and classify the input medical data to be analyzed, and output the prediction results corresponding to the medical data to be analyzed, and the prediction results include the grading labels and classification labels corresponding to the medical data to be analyzed.

[0007] Optionally, the method for constructing a medical data analysis model provided in the present application, wherein the medical training data is input into the constructed initial analysis model, the medical training data is subjected to hierarchical classification processing, and the initial prediction result corresponding to the medical training data is outputted, comprises: Performing feature extraction on the medical training data to obtain a feature vector corresponding to the medical training data; The level and category of the feature vector are identified to obtain the initial prediction result.

[0008] Optionally, in the method for constructing a medical data analysis model provided in the present application, the initial analysis model is constructed based on a BERT module, and the feature extraction of the medical training data to obtain a feature vector corresponding to the medical training data includes: Performing word segmentation processing on the preprocessed medical training data to obtain word segmentation results corresponding to the medical training data; Generate an index sequence and filling information corresponding to the medical training data according to the word segmentation result; According to the index sequence and the filling information, deep semantic extraction is performed on the medical training data to obtain a feature vector corresponding to the medical training data.

[0009] Optionally, in the method for constructing a medical data analysis model provided in the present application, the initial analysis model is constructed based on a TextCNN module, and the feature extraction of the medical training data to obtain a feature vector corresponding to the medical training data includes: Converting text data of medical training data into a sequence of dictionary indices; Converting the dictionary index sequence into a word vector representation; Performing convolution processing on the word vector representation to obtain a convolution result, wherein the convolution result represents local association information between the word vectors; The convolution result is subjected to maximum pooling, and the pooling results of each convolution window are concatenated to obtain a feature vector corresponding to the medical training data.

[0010] Optionally, in the medical data analysis model construction method provided in the present application, the step of performing level and category identification on the feature vector to obtain the initial prediction result includes: Inputting the extracted feature vector into the fully connected layer, performing feature mapping on the feature vector, and obtaining the prediction scores of each category label and each level label corresponding to the medical training data; According to the prediction scores, the probability distribution corresponding to each category label and each level label is calculated; The category label and level label with the highest probability value are used as the initial prediction result.

[0011] Optionally, in the method for constructing a medical data analysis model provided by the present application, when the initial analysis model is constructed based on the BERT module, before the extracted feature vector is input into the fully connected layer, the method further includes: Some elements in the feature vector are randomly set to zero.

[0012] Optionally, the method for constructing a medical data analysis model provided in the present application, before inputting the medical training data into the constructed initial analysis model, the method further comprises: The method for constructing a medical data analysis model provided herein comprises: preprocessing the medical data: Performing data cleaning on the medical training data; According to the set conversion mode, the format of the cleaned medical training data is converted.

[0013] In a second aspect, the present application provides a medical data analysis method, the method comprising: Obtain medical data to be analyzed; The medical data is input into the target analysis model constructed in the medical data analysis model construction method as described in the first aspect, and the prediction result of the medical data to be analyzed is output, wherein the prediction result includes the grading label and classification label corresponding to the medical data to be analyzed.

[0014] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the medical data analysis model building method as described in the first aspect or the medical data analysis method as described in the second aspect.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the medical data analysis model construction method as described in the first aspect or the medical data analysis method as described in the second aspect is implemented.

[0016] According to the specific embodiments provided in this application, this application discloses the following technical effects: The medical data analysis model construction method, analysis method, device and medium provided in the present application, through the collected medical training data including medical scenarios, and the training data set of hierarchical annotation information and classification annotation information of the medical training data, the constructed initial analysis model is iteratively trained to construct a target analysis model that can accurately identify and classify the medical data in each scenario, so that the medical data to be analyzed can be hierarchically classified using the constructed target analysis model, and the corresponding classification results and classification results are output. That is, the present application uses the analysis model to learn and understand the characteristics of medical data containing a large number of professional terms and professional terms, and involving various scenarios, so as to construct the final target analysis model on the basis of multiple iterations and gradual optimization of parameters, and the target analysis model can also be adjusted and expanded according to new medical data and classification standards to adapt to the ever-changing medical data management needs, and finally realizes the automatic and accurate hierarchical classification of medical data, avoiding the hierarchical classification by manual annotation or by other non-medical scenario intelligent models, improving the analysis efficiency and accuracy, and providing strong support for the management and security protection of medical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flowchart of a method for constructing a medical data analysis model according to some embodiments of the present application; Figure 2 A flowchart of a method for constructing a medical data analysis model according to some other embodiments of the present application; Figure 3 A flowchart of a method for constructing a medical data analysis model according to some further embodiments of the present application; Figure 4 A schematic diagram of a process flow of a medical data analysis method according to some embodiments of the present application; Figure 5 A schematic diagram of the structure of a medical data analysis model building device in some embodiments of the present application; Figure 6 A schematic diagram of the structure of a medical data analysis device according to some embodiments of the present application; Figure 7A schematic diagram of the structure of a computer device provided for some embodiments of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0020] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0021] It can be understood that medical data refers to a collection of all data related to medical, health and life activities. It includes big data generated by health-related activities such as immunization, physical examinations, treatments, exercise, diet, etc. throughout the life cycle of an individual from birth to death, and also involves the aggregation of data from medical services, disease prevention and control, health protection, food safety, health care, etc.

[0022] The amount of data is gradually increasing, such as from GB to TB and then to PB; the types of data are also gradually increasing, including structured data, such as Oracle, MySql and other databases, and semi-structured data, such as XML documents, and unstructured data, such as Word and PDF documents, images, audio and video, etc.

[0023] In this application, in order to improve the accuracy and efficiency of medical data grading and classification, and avoid obtaining classification results that are not universal and uniform based on different classifications through manual labeling, a machine learning model is constructed by performing machine learning on a large amount of medical data, which can quickly and accurately grade and classify the input medical data, so as to obtain classification results that can meet the needs of any scenario.

[0024] In order to better understand the medical data analysis model construction method and analysis method provided in the embodiments of the present application, they are described in detail below with reference to the accompanying drawings.

[0025] Figure 1 A flowchart of a method for constructing a medical data analysis model provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes: S110, obtaining a training data set, where the training data set includes medical training data and hierarchical labeling information and classification labeling information of each medical training data.

[0026] S120, input the medical training data into the constructed initial analysis model, perform grading and classification processing on the medical training data, and output the initial prediction results and loss values ​​corresponding to the medical training data. The initial prediction results include the initial grading labels and initial classification labels of the medical training data. The loss value is determined based on the initial grading labels, the initial classification labels, the grading annotation information, and the classification annotation information.

[0027] S130, according to the set loss threshold and the loss value, the initial analysis model is iteratively trained to generate a target analysis model, so that when the medical data is input into the target analysis model and the target prediction result and the target loss value corresponding to the medical training data are output, the target loss value satisfies the set loss threshold; the target analysis model is used to grade and classify the input medical data to be analyzed, and output the prediction result corresponding to the medical data to be analyzed, and the prediction result includes the grade label and classification label corresponding to the medical data to be analyzed.

[0028] Specifically, in an embodiment of the present application, in order to achieve efficient and accurate hierarchical classification of medical data, a neural network model for hierarchical classification of medical data can be constructed through a machine learning model, such as a BERT model or a TextCNN model.

[0029] First, a training data set may be collected, which may include medical training data in each scenario and hierarchical labeling information and classification labeling information of each medical training data.

[0030] That is, the medical training data may be a pre-labeled medical data set, that is, the original medical training data may be labeled with information of the classification and grading information, thereby obtaining a labeled training data set.

[0031] For example, a medical training data set may include various medical data fields, such as patient name (patient_name), department name (department_name), diagnosis information, etc., and corresponding classification labels and grading labels may be annotated for each data.

[0032] Furthermore, after completing the collection of the medical training data set, the medical training data can be used to train the initial neural network model, and finally build a neural network that can accurately analyze and process the input medical data to be analyzed and output hierarchical classification results.

[0033] That is, the medical training data in the collected training data set is input into the constructed initial analysis model, and the medical training data is preliminarily analyzed and processed, namely, graded and classified, and then the initial prediction results and loss values ​​corresponding to the medical training data are output.

[0034] The initial prediction result may include an initial grading label and an initial classification label of the medical training data.

[0035] The loss value may be a cross entropy loss or other types of loss functions, which are not limited in this application.

[0036] In actual training, an output result is obtained, such as an initial prediction result, and the current loss value, such as the cross entropy loss, can be calculated based on the initial classification label, the initial classification label, the classification annotation information and the classification annotation information.

[0037] Finally, the initial analysis model can be iteratively trained according to the set loss threshold and the loss value calculated after this training. During the continuous training process, the model parameters of the initial analysis model are gradually updated, and finally a target analysis model is generated. When the medical training data is input into the target analysis model, the target prediction result and target loss value corresponding to the medical training data can be output. When the current target loss value meets the set loss threshold, it means that the training is completed and the iterative process is stopped. At this time, the analysis model can be used as the target analysis model.

[0038] It can be understood that the constructed target analysis model can be used to grade and classify the input medical data to be analyzed, and output the prediction results corresponding to the medical data to be analyzed, that is, the output includes the grading label and classification label corresponding to the medical data to be analyzed, to indicate the type of medical data to be analyzed, and the level.

[0039] That is, the classification label can represent the category information of the medical data to be analyzed, such as personal information and medical resources; the grading label can represent the sensitive information of the medical data to be analyzed, such as level one sensitive data, level two sensitive data, level three sensitive data and level four sensitive data, among which, the higher the level of medical data, the higher its sensitivity, and better anti-leakage measures can be configured.

[0040] It can be understood that in the embodiments of this application, by combining the scenario of medical data and using the BERT and TextCNN models, we have achieved automatic classification and grading of medical data, greatly improving the efficiency and accuracy of data processing. This not only meets the requirements of medical data security management, but also provides strong support for data analysis and application in the medical field.

[0041] Optionally, in some embodiments of the present application, in S120, using medical training data or pre-processed medical training data, inputting the medical training data into the constructed initial analysis model, and performing hierarchical classification processing on the medical training data may specifically include the following steps: S121, extracting features from the medical training data to obtain feature vectors corresponding to the medical training data.

[0042] S122, performing level and category identification on the feature vector to obtain the initial prediction result.

[0043] Specifically, in an embodiment of the present application, after the medical training data is input into the initial analysis model, the initial analysis model will extract the features of the medical training data, and then use the extracted features to perform hierarchical classification on the medical training data to output an initial prediction result.

[0044] In practice, in some embodiments of the present application, the analysis model can be built based on a BERT module or a TextCNN module.

[0045] Optionally, in some embodiments of the present application, when the initial analysis model is constructed based on the BERT model, in S121, if Figure 2 As shown, the following steps may be specifically included: S1211, performing word segmentation processing on the preprocessed medical training data to obtain word segmentation results corresponding to the medical training data.

[0046] S1212: Generate an index sequence and filling information corresponding to the medical training data according to the word segmentation result.

[0047] S1213, performing deep semantic extraction on the medical training data according to the index sequence and the filling information to obtain a feature vector corresponding to the medical training data.

[0048] Specifically, when using the initial analysis model built based on the BERT module to extract features of medical data, the preprocessed medical data can first be text encoded, that is, the fields in the medical data can be converted into an index sequence of the vocabulary and the corresponding attention mask to distinguish the actual text from the filler part.

[0049] For example, for the collected medical data "patient_name", the goal of the prediction result output by the model is to determine which medical data category "patient_name" belongs to.

[0050] In this step, we can first use BERT's tokenizer to tokenize and encode "patient_name": First, the tokenizer separates it into ["patient", "_", "name"], and then maps the above tokenization results into an index sequence of the vocabulary to obtain input_ids. Finally, an attention_mask is generated to complete the text encoding of "patient_name" to indicate the actual words and the filling part.

[0051] Furthermore, the model is input, that is, the encoded text can be input into the pre-trained BERT model, and the model performs deep semantic feature extraction on the input text to obtain the hidden representation of each word.

[0052] For example, you can input input_ids and attention_mask into the pre-trained BERT model, so that the BERT model can perform deep semantic understanding of the input text and generate hidden states for each word.

[0053] Finally, feature acquisition can be performed, that is, the vector of the special tag position ([CLS] position) output by the BERT model can be extracted as the semantic feature representation of the entire text, that is, the feature vector corresponding to the medical training data is obtained.

[0054] It can be understood that analyzing and processing medical data through an analysis model built based on the BERT module and building a target analysis model has the following advantages: Since the BERT model has been pre-trained on a large amount of general text, it has the ability to deeply understand natural language; and it has the characteristics of transfer learning, that is, through fine-tuning on medical data, the model can adapt to the professional terms and expressions in the medical field. In addition, for feature extraction, the BERT model can capture subtle semantic differences in medical data and improve classification accuracy.

[0055] Alternatively, in some other embodiments of the present application, when the initial analysis model is a TextCNN model, in S121, the preprocessed medical data is input into the initial analysis model, and an initial hierarchical classification result is output, such as Figure 3 As shown, the following steps may be specifically included: S1214, converting the text data of the medical training data into a dictionary index sequence.

[0056] S1215, convert the dictionary index sequence into a word vector representation.

[0057] S1216, performing convolution processing on the word vector representation to obtain a convolution result, which represents the local correlation information between the word vectors.

[0058] S1217, performing maximum pooling on the convolution result, and concatenating the pooling results of each convolution window to obtain a feature vector corresponding to the medical training data.

[0059] Specifically, when using the initial analysis model built based on the TextCNN module to extract features from medical data, all words that appear in the medical training data can be collected first to build a dictionary, and then the medical data field name or content can be converted into a dictionary index sequence. That is, the preprocessed text data can be converted into a dictionary index sequence of fixed length.

[0060] For example, a dictionary might include {" <k> ":0, " <unk>":1, "patient":2, "name":3, "blood":4, "pressure":5} and other words.

[0061] The field name "blood_pressure" is segmented into ["blood", "pressure"], which is then mapped to the dictionary index sequence [4, 5].

[0062] For another example, "patient_name" may be split into ["patient", "name"]. Assuming that the index of "patient" in the dictionary is 100 and the index of "name" is 200, they can be mapped and converted to obtain the dictionary index sequence [100, 200].

[0063] Furthermore, word vector mapping can be performed based on the generated dictionary index sequence, that is, the dictionary index sequence can be converted into a word vector representation using the embedding layer (nn.Embedding) in the model.

[0064] Among them, each word index corresponds to a dense low-dimensional vector that captures the semantic information of the word.

[0065] Furthermore, after completing the word vector representation, convolutional feature extraction can be performed, that is, multiple one-dimensional convolutional layers can be used to perform convolution operations on the word vector sequence to extract different n-gram (such as 2-gram, 3-gram) features to capture the local correlation information between each word.

[0066] Among them, the convolution kernel widths are 2, 3, and 4 respectively.

[0067] Finally, the local association relationships can be pooled and spliced, that is, the convolution results are subjected to the maximum pooling operation to extract the key features, and then the pooling results of different convolution windows are spliced ​​into a feature vector, thereby forming a more comprehensive text feature representation, that is, combining the information of different n-gram features to serve as the feature vector of medical training data.

[0068] For example, for a medical data field name "patient_age", the processing process through the TextCNN module is as follows: Dictionary index: The word segmentation of "patient_age" is ["patient", "age"], and the mapped dictionary index sequence is [2, 6].

[0069] Furthermore, [2, 6] is represented by word vector.

[0070] Furthermore, convolution kernels of different widths are used to extract features.

[0071] For example, using a convolutional layer with a width of 2 to extract the features of ["patient", "age"] may capture the features of "patient age".

[0072] It is understandable that due to the short sequence length, convolutions with widths of 3 and 4 cannot be applied, but the model can still handle it effectively.

[0073] Furthermore, feature concatenation is performed, that is, the pooling results of each convolutional layer can be concatenated to form a comprehensive feature vector. Finally, feature classification is performed, that is, the feature vector can be input into the fully connected layer and the category probability can be output, which may be classified as A1-personal statistical information.

[0074] It can be understood that in the embodiment of the present application, by using TextCNN to extract features of medical training data and constructing a target analysis model, automatic classification and grading of medical data is achieved, greatly improving the efficiency and accuracy of data processing. This not only meets the requirements of medical data security management, but also provides strong support for data analysis and application in the medical field.

[0075] Optionally, in some embodiments of the present application, in S122, the level and category of the feature vector are identified to obtain the initial prediction result, such as Figure 4 As shown, the following steps may be specifically included: S1221, input the extracted feature vector into the fully connected layer, perform feature mapping on the feature vector, and obtain the prediction scores of each category label corresponding to the medical training data, as well as the prediction scores of each level of labels.

[0076] S1222: Calculate the probability distribution of each category label and each level label based on the prediction score.

[0077] S1223, taking the category label and level label with the highest probability value as the initial prediction result.

[0078] Specifically, first, a fully connected layer may be used for feature mapping, that is, the extracted feature vector may be input into the fully connected layer, i.e., the classifier, and mapped to a predefined category space.

[0079] Furthermore, the model outputs the predicted probability of each category, and the classification and grading results of the input data are determined according to the maximum probability value.

[0080] That is, first, the feature vector processed by dropout can be input into the fully connected layer (classifier). Then the classifier maps the high-dimensional features to the predefined category space and outputs the prediction scores of each category.

[0081] Furthermore, the softmax function can be used to calculate the probability distribution of each category based on the prediction scores of each category.

[0082] Finally, the category with the highest probability can be selected as the prediction result of the model, that is, the initial prediction result.

[0083] For example, for the above "patient_name" medical data, after the above processing, the model may output its classification result as A2-personal identity information, because it contains the patient's name and belongs to the personal identity category.

[0084] For another example, for "department_name" (department name), the model will be encoded by the tokenizer and may get ["department", "_", "name"].

[0085] Finally, the classifier outputs a prediction result, which may be classified as E-health resource data because the department name belongs to the resource information of the hospital.

[0086] Optionally, in some embodiments of the present application, for the feature vector extracted by the analysis model built based on the BERT module, in order to prevent the model from overfitting, before the feature vector is input into the fully connected layer, that is, before the classifier is executed, the method may further include: S1220, randomly set some elements in the feature vector to zero.

[0087] Specifically, a dropout layer is applied to the extracted [CLS] vector to randomly set some elements to zero to prevent the model from overfitting the training data.

[0088] Optionally, in some embodiments of the present application, preprocessing of the medical training data may specifically include the following measures: S01, data cleaning.

[0089] S02, format conversion.

[0090] Specifically, we can first remove duplicate data records, process missing values ​​and outliers, and ensure the integrity and accuracy of the data.

[0091] Furthermore, text normalization can be performed: replace underscores in field names with spaces, convert all letters to lowercase, and unify the format to facilitate effective model training.

[0092] It can be understood that the above steps of preprocessing the medical training data can be performed in sequence of S01 and S02 according to actual conditions, or S01 or S02 can be performed selectively, and the present application does not impose any limitation on this.

[0093] It can be understood that through the above steps, the BERT and TextCNN models can be used to extract features and classify medical data, realize automatic classification and grading of medical data fields, and provide strong support for the management and security protection of medical data.

[0094] On the other hand, in some embodiments of the present application, after the target analysis model is constructed through the above-mentioned embodiments, the constructed target analysis model can be used to perform actual analysis and processing of the medical data, that is, to perform hierarchical classification processing.

[0095] like Figure 5 As shown, the analysis process specifically includes the following steps: S210, obtaining medical data to be analyzed.

[0096] S220, input the medical data into the target analysis model constructed in the pre-constructed medical analysis model construction method, and output the prediction result of the medical data to be analyzed, which includes the grading label and classification label corresponding to the medical data to be analyzed.

[0097] Specifically, in an embodiment of the present application, the acquired medical data to be analyzed can be directly input into the target analysis model constructed in the above embodiment to grade and classify the input medical data to be analyzed, and finally output the graded classification result, that is, the prediction result, to indicate the category and level of the medical data to be analyzed.

[0098] It can be understood that the process of feature extraction and analysis of the medical data to be analyzed using the target analysis model is similar to the feature extraction process of the medical training data when the analysis model is constructed through training as mentioned above, and will not be elaborated here.

[0099] It can also be understood that in the embodiments of the present application, by using the BERT and TextCNN models in the application scenarios of medical data, automatic classification and grading of large amounts of medical data can be achieved.

[0100] That is, the constructed analysis model combines the characteristics of medical data for model design: Rich in professional terms: Medical data contains a large amount of professional terms and proper nouns, and the model needs to be able to understand these words.

[0101] Diverse data types: Medical data includes personal information, diagnostic information, test results, etc. There are many categories, and the model needs to distinguish them accurately.

[0102] Model training and optimization: Annotated dataset: Use manually annotated medical data fields and their corresponding categories and levels as training data for the model.

[0103] Multi-round iterative training: The model iterates multiple times on the training data to gradually optimize the parameters and improve the classification performance.

[0104] Performance evaluation: Evaluate the model's accuracy, precision, recall, F1 value and other indicators on the validation set and test set to ensure the reliability of the model.

[0105] Finally, classification prediction in practical applications: Enter new data: In actual use, enter new medical data field names, such as "patient_id" or "diagnosis_result".

[0106] Model prediction: The model processes the input data and outputs the predicted category and rating, for example: Category: A2-Personally Identifiable Information, which means that the field name belongs to the category of Personally Identifiable Information.

[0107] Level: Level 4 sensitive data, indicating the sensitivity of the data and the need for strict protection.

[0108] Application of results: According to the prediction results of the model, formulate corresponding data processing strategies, such as data desensitization and access control, to ensure the security of medical data.

[0109] Finally, the interpretability and scalability of the model: Explainability: By analyzing the model’s feature extraction process, we can understand how the model uses the features of medical data for classification, which helps improve the credibility of the model.

[0110] Scalability: The model can be adjusted and expanded according to new medical data and classification standards to adapt to the changing needs of medical data management.

[0111] That is, by combining the scenario of medical data, using the BERT and TextCNN models, we have achieved automatic classification and grading of medical data, greatly improving the efficiency and accuracy of data processing. This not only meets the requirements of medical data security management, but also provides strong support for data analysis and application in the medical field.

[0112] On the other hand, the present application also provides a medical data analysis model construction device, such as Figure 6 Said device comprises: An acquisition module 310 is used to acquire a training data set, wherein the training data set includes medical training data and hierarchical labeling information and classification labeling information of each medical training data; A first training module 320 is used to input the medical training data into the constructed initial analysis model, perform classification and processing on the medical training data, and output an initial prediction result and a loss value corresponding to the medical training data, wherein the initial prediction result includes an initial classification label and an initial classification label of the medical training data, and the loss value is determined according to the initial classification label, the initial classification label, the classification annotation information, and the classification annotation information; The second training module 330 is used to iteratively train the initial analysis model according to the set loss threshold and the loss value to generate a target analysis model, so that when the medical data is input into the target analysis model and the target prediction result and target loss value corresponding to the medical training data are output, the target loss value satisfies the set loss threshold; the target analysis model is used to classify and classify the input medical data to be analyzed, and output the prediction result corresponding to the medical data to be analyzed, and the prediction result includes the grading label and classification label corresponding to the medical data to be analyzed.

[0113] Optionally, in the medical data analysis model construction device of the embodiment of the present application, the first training module is specifically used for: Performing feature extraction on the medical training data to obtain a feature vector corresponding to the medical training data; The level and category of the feature vector are identified to obtain the initial prediction result.

[0114] Optionally, in the medical data analysis model construction device of the embodiment of the present application, the initial analysis model is constructed based on the BERT module, and the first training module is specifically used for: Performing word segmentation processing on the preprocessed medical training data to obtain word segmentation results corresponding to the medical training data; Generate an index sequence and filling information corresponding to the medical training data according to the word segmentation result; According to the index sequence and the filling information, deep semantic extraction is performed on the medical training data to obtain a feature vector corresponding to the medical training data.

[0115] Optionally, in the medical data analysis model construction device of the embodiment of the present application, the initial analysis model is constructed based on the TextCNN module, and the first training module is specifically used for: Converting text data of medical training data into a sequence of dictionary indices; Converting the dictionary index sequence into a word vector representation; Performing convolution processing on the word vector representation to obtain a convolution result, wherein the convolution result represents local association information between the word vectors; The convolution result is subjected to maximum pooling, and the pooling results of each convolution window are concatenated to obtain a feature vector corresponding to the medical training data.

[0116] Optionally, in the medical data analysis model construction device of the embodiment of the present application, the first training module is specifically used for: Inputting the extracted feature vector into the fully connected layer, performing feature mapping on the feature vector, and obtaining the prediction scores of each category label and each level label corresponding to the medical training data; According to the prediction scores, the probability distribution corresponding to each category label and each level label is calculated; The category label and level label with the highest probability value are used as the initial prediction result.

[0117] Optionally, in the medical data analysis model construction device of the embodiment of the present application, the first training module is further used for: Some elements in the feature vector are randomly set to zero.

[0118] Optionally, the medical data analysis model construction device of the embodiment of the present application further includes: A preprocessing module 340 is used to clean the medical training data; And it is used to convert the format of the cleaned medical training data according to the set conversion mode.

[0119] On the other hand, an embodiment of the present application further provides a medical data analysis device, the medical data analysis device comprising: An acquisition unit, used for acquiring medical data to be analyzed; The analysis unit is used to input the medical data into the target analysis model constructed in the pre-constructed medical analysis model construction method, and output the prediction result of the medical data to be analyzed, wherein the prediction result includes the graded label and the classification label corresponding to the medical data to be analyzed.

[0120] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store video tag processing data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a meteorological data transmission method is implemented.

[0121] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0122] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0123] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0124] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0125] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0126] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0127] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a computer device, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0128] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0129] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.< / unk> < / k>

Claims

1. A method for constructing a medical data analysis model, characterized in that: The method comprises: Acquire a training data set, wherein the training data set includes medical training data and hierarchical labeling information and classification labeling information of each medical training data; Input the medical training data into the constructed initial analysis model, perform grading and classification processing on the medical training data, and output the initial prediction result and loss value corresponding to the medical training data, wherein the initial prediction result includes the initial grading label and the initial classification label of the medical training data, and the loss value is determined according to the initial grading label, the initial classification label, the grading annotation information, and the classification annotation information; According to the set loss threshold and the loss value, the initial analysis model is iteratively trained to generate a target analysis model, so that when the medical training data is input into the target analysis model and the target prediction result and target loss value corresponding to the medical training data are output, the target loss value satisfies the set loss threshold; the target analysis model is used to grade and classify the input medical data to be analyzed, and output the prediction result corresponding to the medical data to be analyzed, and the prediction result includes the grade label and classification label corresponding to the medical data to be analyzed.

2. The method for constructing a medical data analysis model according to claim 1, characterized in that: The step of inputting the medical training data into the constructed initial analysis model, performing hierarchical classification processing on the medical training data, and outputting the initial prediction results corresponding to the medical training data comprises: Performing feature extraction on the medical training data to obtain a feature vector corresponding to the medical training data; The level and category of the feature vector are identified to obtain the initial prediction result.

3. The method for constructing a medical data analysis model according to claim 2, characterized in that: The initial analysis model is constructed based on the BERT module, and the feature extraction of the medical training data to obtain the feature vector corresponding to the medical training data includes: Performing word segmentation processing on the preprocessed medical training data to obtain word segmentation results corresponding to the medical training data; Generate an index sequence and filling information corresponding to the medical training data according to the word segmentation result; According to the index sequence and the filling information, deep semantic extraction is performed on the medical training data to obtain a feature vector corresponding to the medical training data.

4. The method for constructing a medical data analysis model according to claim 2, characterized in that: The initial analysis model is constructed based on the TextCNN module, and the feature extraction of the medical training data to obtain the feature vector corresponding to the medical training data includes: Converting text data of medical training data into a sequence of dictionary indices; Converting the dictionary index sequence into a word vector representation; Performing convolution processing on the word vector representation to obtain a convolution result, wherein the convolution result represents local association information between the word vectors; The convolution result is subjected to maximum pooling, and the pooling results of each convolution window are concatenated to obtain a feature vector corresponding to the medical training data.

5. The method for constructing a medical data analysis model according to any one of claims 2 to 4, characterized in that: The performing level and category identification on the feature vector to obtain the initial prediction result comprises: Inputting the extracted feature vector into the fully connected layer, performing feature mapping on the feature vector, and obtaining the prediction scores of each category label and each level label corresponding to the medical training data; According to the prediction scores, the probability distribution corresponding to each category label and each level label is calculated; The category label and level label with the highest probability value are used as the initial prediction result.

6. The method for constructing a medical data analysis model according to claim 5, characterized in that: When the initial analysis model is constructed based on the BERT module, before the extracted feature vector is input into the fully connected layer, the method further includes: Some elements in the feature vector are randomly set to zero.

7. The method for constructing a medical data analysis model according to claim 1, characterized in that: Before inputting the medical training data into the constructed initial analysis model, the method further includes: Performing data cleaning on the medical training data; According to the set conversion mode, the format of the cleaned medical training data is converted.

8. A medical data analysis method, characterized in that: The method comprises: Obtain medical data to be analyzed; The medical data is input into the target analysis model constructed in the medical data analysis model construction method as described in any one of claims 1 to 7, and the prediction result of the medical data to be analyzed is output, wherein the prediction result includes the grading label and classification label corresponding to the medical data to be analyzed.

9. A computer device, characterized in that: The computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the medical data analysis model building method as described in any one of claims 1 to 7 or the medical data analysis method as described in claim 8.

10. A computer-readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed by a processor, implements the medical data analysis model construction method according to any one of claims 1 to 7 or the medical data analysis method according to claim 8.

Images