Training method and prediction device for multi-class gynecological disease prediction model

Abstract

The invention discloses a training method and a prediction device for a multi-class gynecological disease prediction model, and relates to the technical field of gynecological disease prediction, and the method comprises the steps: obtaining a gynecological electronic case data set, and carrying out the preprocessing of the data set; the preprocessed data set is divided into a training set and a verification set according to the proportion of 4: 1, and the training set and the verification set serve as input of training of at least two prediction models; and evaluating the trained prediction model, taking out the model with the best prediction effect as the prediction model, and the like. Compared with the training of the existing gynecological disease prediction model, the method has the advantages that the training time is shortened, and meanwhile, the execution efficiency is improved; 12 gynecological diseases can be predicted at the same time according to gynecological electronic case data, the generalization ability of the prediction model is enhanced, and the method can be practically popularized and used.

Description

Technical Field

[0001] The present invention relates to the technical field of gynecological disease prediction, and in particular to a training method and a prediction device for a multi-category gynecological disease prediction model. Background Art

[0002] In recent years, gynecological diseases have become a common physiological problem for women in various countries. The incidence rate continues to rise, and it shows a trend of covering multiple age groups. The global incidence rate of gynecological diseases has reached more than 65%. Against the background of the continued increase in the incidence rate of gynecological diseases, how to improve the diagnostic efficiency to reduce the average consultation time for each patient is an issue worthy of attention.

[0003] Among all types of gynecological diseases, the incidence of some types of gynecological diseases is generally higher than that of all gynecological diseases. The types of diseases included in this part often change with the region, time, age distribution, educational level and other attributes, but there is still a small number of diseases that have a higher incidence in most case data.

[0004] With the development of digital medicine, electronic medical records have gradually replaced paper medical records, making data statistics more convenient. It has become easier to use machine learning algorithms to predict diseases based on electronic medical records. The various algorithms included in machine learning algorithms help to achieve accurate disease prediction and provide prediction results to doctors to assist doctors in disease judgment, so as to achieve rapid diagnosis and provide corresponding diagnosis and treatment plans.

[0005] At present, there are a lot of literatures in the research community on using various machine learning methods to achieve disease prediction. For example, Md. Jalal Uddin Chowdhury collected data from 5110 patients, constructed artificial neural networks (ANN) and machine learning models to classify diseases, which included 11 input features. The performance of six different classification algorithms in the field of disease prediction was evaluated through experiments. The evaluation results showed that the artificial neural network (ANN) and k-nearest neighbor (KNN) algorithms had the highest prediction accuracy. However, if the number of data sets is reduced, it cannot be guaranteed that the constructed artificial neural network will still maintain a high prediction accuracy. For example, Asadi F used machine learning classification algorithms such as support vector machine (SVM), QUEST, C&R tree, and multilayer perceptron (MLP) to analyze 23 attributes in 145 case data to predict the probability of cervical cancer. The experimental results showed that the QUEST method had the highest accuracy, reaching 95%. However, the data set is small, it is easy to overfit during training, and the generalization ability of the constructed model needs to be considered.

[0006] In the above studies, researchers used various machine learning algorithms to build disease prediction models for different data sets, and determined the characteristic attributes related to the predicted disease type, with the number of characteristic attributes generally ranging from 10 to 24. Among the machine learning algorithms used, the best algorithm varies depending on the data set processing method and the number of characteristic attributes. However, the above studies are only for the prediction of a certain disease, and cannot simultaneously predict whether the patient is at risk of suffering from other diseases.

[0007] At present, there are few studies in the research community on predicting whether a patient has multiple gynecological diseases at the same time, because it has strict requirements on the amount of data, the complexity of the model, the size of the feature dimension, etc. For example, ELHAM F.AHMAD predicts and classifies the 11 most common gynecological diseases in its data set (polyps, infections, myomas, prolapse, cancer, endometrial hyperplasia, displacement, amenorrhea, miscarriage, dysmenorrhea and infertility). It uses case data provided by 550 patients and adopts the multi-layer perceptron (MLP) feedforward neural network method. It contains 54 input variables to build a model, which is divided into 51 symptom information and 3 basic information, which is much higher than the number of attributes used for predicting a certain disease mentioned above, and the constructed model is also more complex. The experimental results show that the model finally achieved an accuracy of 94.5%. Although a high accuracy rate was achieved, the number of input symptom attributes was large, which resulted in a long training time and low efficiency for the model, and it cannot be applied in actual use scenarios. Summary of the invention

[0008] The embodiment of the present invention provides a training method and a prediction device for a multi-category gynecological disease prediction model, which are used to solve the above-mentioned problems.

[0009] A training method for a multi-class gynecological disease prediction model, comprising:

[0010] Obtain gynecological electronic medical record data set and preprocess the data set;

[0011] The preprocessed dataset is divided into a training set and a validation set in a ratio of 4:1, which is used as the input for training at least two prediction models;

[0012] The prediction models that have been trained are evaluated, and the model with the best prediction effect is selected as the prediction model.

[0013] Furthermore, the data set preprocessing steps include: data cleaning, word segmentation, feature extraction, vectorization, and oversampling.

[0014] Furthermore, the processing of the data set also includes the following steps:

[0015] Analyze the types and proportions of diseases;

[0016] Select the top 12 diseases with the largest proportion according to their proportion;

[0017] Obtain disease characteristics corresponding to 12 diseases;

[0018] The above disease characteristics are screened according to the set weight threshold, and the 12 screened diseases and corresponding disease characteristics are output;

[0019] The disease features corresponding to the 12 diseases were split and merged, and finally mapped to 35 attributes as input for prediction model training.

[0020] Furthermore, the 12 diseases include ectopic pregnancy, abnormal uterine bleeding, adhesions, condyloma acuminatum, abscesses, cysts, polyps, infections, myomas, prolapse, cancer and lesions, and migration.

[0021] Furthermore, the prediction effect of the prediction model is evaluated from two dimensions: accuracy and F1 score. The calculation formula is as follows:

[0022]

[0023] In the formula: F1-Score is the F1 score; Accuracy is the accuracy rate; Precision is the precision rate; Recall is the recall rate; TP_avg, FP_avg, TN_avg, and FN_avg represent the average TP value, FP value, TN value, and FN value of the 12 diseases, respectively.

[0024] Furthermore, the calculation formula of TP_avg is as follows:

[0025]

[0026] Where: n is the number of disease types; TP i represents the TP value of the ith disease.

[0027] Furthermore, the calculation formula of FP_avg is as follows:

[0028]

[0029] Where: n is the number of disease types; FP i represents the FP value of the i-th disease.

[0030] Furthermore, the calculation formula of TN_avg is as follows:

[0031]

[0032] Where: n is the number of disease types; TN i Represents the TN value of the ith disease.

[0033] Furthermore, the calculation formula of FN_avg is as follows:

[0034]

[0035] Where: n is the number of disease types; FN i represents the FN value of the ith disease.

[0036] The embodiment of the present invention provides a device for predicting multiple types of gynecological diseases, including:

[0037] The acquisition module is used to obtain gynecological electronic medical records data;

[0038] A preprocessing module is used to preprocess the acquired gynecological electronic medical record data to obtain features that can be input into the prediction model;

[0039] The prediction model is used to input features and make predictions to obtain prediction results.

[0040] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0041] 1. By preprocessing the acquired gynecological electronic medical record data set, the feature dimension can be greatly reduced, and the feature attributes after dimensionality reduction can be quantified, thereby reducing training time and improving execution efficiency.

[0042] 2. The prediction effects of multiple prediction models can be evaluated, so that the prediction model with the best prediction effect can be selected according to the evaluation results and used as the prediction model for multiple types of gynecological diseases.

[0043] 3. It can predict 12 gynecological diseases at the same time based on gynecological electronic medical record data. Compared with the existing technology, it enhances the generalization ability of the prediction model while reducing the running training time and improving the execution efficiency.

[0044] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0045] The technical solution of the present invention is further described in detail below through the drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0047] Figure 1 A flow chart of a training method for a multi-category gynecological disease prediction model disclosed in an embodiment of the present invention;

[0048] Figure 2 A flowchart of word segmentation and feature extraction processing disclosed in an embodiment of the present invention;

[0049] Figure 3 A distribution diagram of the types of diseases and their proportions disclosed in the embodiments of the present invention;

[0050] Figure 4 It is the ROC curve of the prediction model disclosed in the embodiment of the present invention. DETAILED DESCRIPTION

[0051] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0052] The training method of multiple gynecological disease prediction models provided in the embodiment of the present application can be applied to terminal devices such as laptops, personal computers, servers, etc. The embodiment of the present application does not impose any restrictions on the specific type of terminal devices.

[0053] Table 1 (a preferred application environment)

[0054]

[0055] The dataset used in this application is a gynecological electronic medical record dataset collected by a domestic tertiary hospital from 2023 to 2024. By dividing the disease types in the dataset and counting the proportion of the number, we get the following: Figure 3 Distribution chart of the types and number of diseases shown.

[0056] Through Figure 1 By sorting the proportion of each disease type, we can get the top 12 disease types with the largest proportion: ectopic pregnancy, abnormal uterine bleeding, adhesions, condyloma acuminatum, abscesses, cysts, polyps, infections, myomas, prolapse, cancer and lesions, and migration. Since the proportion of diseases in the dataset is affected by factors such as the hospital's location, time, age distribution, and educational level, the types of common gynecological diseases will vary in different hospitals, but the common types of diseases are still mostly similar. For example, in some hospitals, they are: polyps, infections, myomas, prolapse, cancer, endometrial hyperplasia, displacement, amenorrhea, miscarriage, dysmenorrhea, and infertility. This application analyzes and predicts the top 12 gynecological disease cases with the largest number, with a total of 1,136 cases.

[0057] The characteristics of each of the twelve diseases were determined, and the multiple characteristics of each disease were sorted according to the weights. The characteristics with smaller weights were deleted to achieve the least sacrifice of prediction accuracy, reduce the number of input attributes required for model training as much as possible, and improve the efficiency of model training. Finally, the disease type and characteristic correspondence table shown in Table 1 was obtained. The characteristics were appropriately split and merged, and finally mapped to 35 attributes as the input of model training.

[0058] Table 2 (12 diseases and some characteristics)

[0059]

[0060]

[0061] Figure 1 A flow chart of a training method for a multi-class gynecological disease prediction model is shown, which includes:

[0062] S1, obtain the gynecological electronic medical record dataset and preprocess the dataset.

[0063] The purpose of data preprocessing is to reduce the training time of the prediction model and improve the execution efficiency. To this end, the following methods are adopted:

[0064] 1. Data cleaning.

[0065] Data cleaning mainly removes erroneous data and duplicate data in the gynecological electronic medical record data set, processes missing values, and corrects inconsistencies in the data. This is to ensure data quality and make the gynecological electronic medical record data set more suitable for data analysis or training models. The present invention removes data with incomplete medical record information, typos, and data that does not belong to the twelve common diseases.

[0066] 2. Participle.

[0067] In text processing, word segmentation is the process of breaking down a continuous text sequence into meaningful elements, usually words, phrases or other symbols, to simplify the model processing process and improve processing efficiency. The present invention processes a Chinese medical record dataset, so in addition to normal word segmentation of the dataset, in order to retain the original context and time elements in a sentence, each case data is also sentence-processed into short sentences for subsequent feature extraction operations, such as Figure 2 shown.

[0068] 3. Feature extraction.

[0069] Feature extraction is the process of extracting information from raw data to construct the feature set of the algorithm. This includes extracting important information and attributes that are helpful for model learning from various types of data. The feature extraction operation of the present invention is to extract from the short words and short sentences obtained after word segmentation, where short words are mainly used to extract disease feature information, and short sentences are mainly used to extract contextual elements such as time and place related to the feature, so that the extracted feature information is more accurate and complete, such as Figure 2 shown.

[0070] 4. Vectorization.

[0071] Vectorization is the process of converting non-numeric data into numerical form, especially in natural language processing, converting text into digital or vector format so that it can be processed by computer and facilitates subsequent vectorization processing. The present invention maps the extracted features to obtain 35 attributes and vectorizes them. Each attribute can take different values. For example, some attributes have a value range of [0, 1, 2, 3, 4, 5, 6], which means that there are 7 cases of this feature related to twelve diseases, while for the range of [0, 1], there are only 2 cases.

[0072] 5. Oversampling.

[0073] The vectorized dataset is subjected to intra-class oversampling to balance the number of positive and negative samples for each disease.

[0074] The processing of the data set also includes the following steps:

[0075] S11, analysis of disease types and proportions;

[0076] S12, select the top 12 diseases with the largest proportion according to their proportion;

[0077] S13, obtaining disease characteristics corresponding to 12 diseases;

[0078] S14, screening the above disease characteristics according to the set weight threshold, and outputting the screened 12 diseases and corresponding disease characteristics;

[0079] S15, the disease features corresponding to the 12 diseases were split and merged, and finally mapped to 35 attributes as the input for prediction model training.

[0080] S2, the preprocessed dataset is divided into a training set and a validation set in a ratio of 4:1, which serves as the input for training at least two prediction models.

[0081] In this embodiment, the prediction models adopt four types of models: logistic regression, support vector machine, random forest, and artificial neural network.

[0082] The above four prediction models are applied to the environment in Table 1.

[0083] The effect of the prediction model can be evaluated from two dimensions: accuracy and F1 score. The calculation formula is as follows:

[0084]

[0085] Where: F1-Score is the F1 score; Accuracy is the accuracy rate; Precision is the precision rate; Recall is the recall rate; TP_avg, FP_avg, TN_avg, FN_avg represent the average TP value, FP value, TN value, FN value of 12 diseases respectively; n is the number of disease types; TP i represents the TP value of the i-th disease; FP i represents the FP value of the i-th disease; TN i Indicates the TN value of the i-th disease; FN i represents the FN value of the ith disease.

[0086] Explanation of the above terms:

[0087] TP: True Positive (TP), the patient appears to be ill, confirmed by a positive test.

[0088] FP: False Positive (FP), although the test is positive, the patient is negative for the disease.

[0089] TN: True Negative (TN), the patient is sick but the test result is negative.

[0090] FN: False Negative (FN), the test result for the disease is negative, so the patient does not have the disease.

[0091] The above formulas (1) to (8) can be used to simultaneously evaluate the prediction of 12 diseases, greatly improving the evaluation efficiency of prediction model training.

[0092] S3, evaluating the prediction models that have completed training, and taking out the model with the best prediction effect as the prediction model.

[0093] Table 3 (Performance analysis of four prediction models)

[0094]

[0095] From the performance analysis of the above four prediction models and Figure 4It can be seen that the accuracy and F1 score of the artificial neural network are both optimal, and its prediction effect is the best. The artificial neural network model constructed in this embodiment takes 312 seconds to train. Although the training is not completed within tens of seconds, it has achieved an accuracy of 96.51%. And while ensuring high accuracy, the time level does not reach thousands, and various indicators are also much higher than other machine learning methods.

[0096] A multi-category gynecological disease prediction device can be deployed on terminal devices such as laptops, personal computers, servers, etc., and includes: a prediction model trained by the above-mentioned multi-category gynecological disease prediction model training method, an acquisition module, and a preprocessing module.

[0097] in,

[0098] The acquisition module is used to obtain gynecological electronic medical records data;

[0099] A preprocessing module is used to preprocess the acquired gynecological electronic medical record data to obtain features that can be input into the prediction model;

[0100] The above preprocessing includes data cleaning, word segmentation, feature extraction and vectorization. The preprocessed data is convenient for the prediction model to make predictions.

[0101] The prediction model is used to input features and make predictions to obtain prediction results.

[0102] After the multi-category gynecological disease prediction device is deployed on the equipment, the prediction results can be output after the gynecological electronic case data is imported, which can effectively play an auxiliary role, reduce the diagnosis pressure of doctors, and improve the work efficiency of doctors.

[0103] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0104] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0105] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments of the present invention may all be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps described above are generally described around their functions. Whether such functions are implemented as hardware or software depends on the specific application and the design constraints imposed on the entire system. A skilled person may implement the described functions in an alternative manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of the present disclosure.

[0106] The steps of the method or algorithm described in conjunction with the embodiments of the present invention may be directly embodied as hardware, a software module executed by a processor, or a combination thereof. The software module may be located in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a mobile disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also be present in a user terminal as discrete components.

[0107] For software implementation, the techniques described in this application can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is coupled to the processor in a communication manner via various means, which are well known in the art.

[0108] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described in the present invention are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

Claims

1. A training method for a multi-class gynecological disease prediction model, characterized in that: include: Obtain gynecological electronic medical record data set and preprocess the data set; The preprocessed dataset is divided into a training set and a validation set in a ratio of 4:1, which is used as the input for training at least two prediction models; The prediction models that have been trained are evaluated, and the model with the best prediction effect is selected as the prediction model.

2. The method according to claim 1, characterized in that The data set preprocessing steps include: data cleaning, word segmentation, feature extraction, vectorization, and oversampling.

3. The method according to claim 1, characterized in that The processing of the data set also includes the following steps: Analyze the types and proportions of diseases; Select the top 12 diseases with the largest proportion according to their proportion; Obtain disease characteristics corresponding to 12 diseases; The above disease characteristics are screened according to the set weight threshold, and the 12 screened diseases and corresponding disease characteristics are output; The disease features corresponding to the 12 diseases were split and merged, and finally mapped to 35 attributes as input for prediction model training.

4. The method according to claim 3, characterized in that The 12 diseases include ectopic pregnancy, abnormal uterine bleeding, adhesions, genital warts, abscesses, cysts, polyps, infections, myomas, prolapse, cancer and lesions, and metastasis.

5. The method according to claim 1, characterized in that The prediction effect of the prediction model is evaluated from two dimensions: accuracy and F1 score. The calculation formula is as follows: In the formula: F1-Score is the F1 score; Accuracy is the accuracy rate; Precision is the precision rate; Recall is the recall rate; TP_avg, FP_avg, TN_avg, and FN_avg represent the average TP value, FP value, TN value, and FN value of the 12 diseases, respectively.

6. The method according to claim 5, characterized in that The calculation formula of TP_avg is as follows: Where: n is the number of disease types; TP i represents the TP value of the ith disease.

7. The method according to claim 5, characterized in that The calculation formula of FP_avg is as follows: Where: n is the number of disease types; FP i represents the FP value of the i-th disease.

8. The method according to claim 5, characterized in that The calculation formula of TN_avg is as follows: Where: n is the number of disease types; TN i Represents the TN value of the ith disease.

9. The method according to claim 5, characterized in that The calculation formula of FN_avg is as follows: Where: n is the number of disease types; FN i represents the FN value of the ith disease.

10. A device for predicting multiple gynecological diseases, characterized in that: include: The acquisition module is used to obtain gynecological electronic medical records data; A preprocessing module is used to preprocess the acquired gynecological electronic medical record data to obtain features that can be input into the prediction model; The prediction model is used to input features and make predictions to obtain prediction results.

Images