CL-FCM-based Data Preprocessing Method and System for Rheumatic Immune Diseases

Abstract

The present invention provides a method and system for preprocessing rheumatic immune disease data based on CL-FCM. The method includes: determining attributes related to disease diagnosis based on rheumatic immune disease clinical data and an expert database; screening available information, removing redundant information, and performing encoding transformation and missing value processing on each attribute; determining the optimal number of clusters using cross-entropy and the elbow method; performing data augmentation on single or multiple attributes to form multiple view clustering results, calculating the overall loss, and updating the original feature representation; using the FCM algorithm to cluster the updated data set, and obtaining clustering results that can be directly used in downstream prediction models for disease prediction. The present invention can accurately enhance key attributes and improve disease diagnosis efficiency when facing a data set with a large number of relevant attributes; can reduce redundant features and calculation amount for subsequent inference models, and can help doctors more comprehensively understand the key features of rheumatic diseases.

Description

Technical Field

[0001] The present invention integrates knowledge in the fields of medicine, information science, computer applications, etc., and invents a method and system for preprocessing rheumatism and immunology disease data based on CL-FCM. Background Art

[0002] Rheumatism and immunology diseases are a group of diseases involving joints, bones, muscles and the immune system. This group of diseases includes more than 100 diseases, and the most common ones are rheumatoid arthritis, ankylosing spondylitis, systemic lupus erythematosus, osteoarthritis, gout, etc. The symptoms and manifestations of rheumatism and immunology diseases are diverse, and different diseases may have similar symptoms and signs, making diagnosis difficult. The diagnosis of rheumatism and immunology diseases usually requires integrating multiple data sources, including clinical data, auxiliary examinations, medical history, etc. These data have different dimensions and characteristics. In order to better determine the type of rheumatism and immunology disease suffered by a patient, currently, a big data analysis prediction model is often used to analyze the manifestations of each symptom and symptom of the patient, combined with the patient's clinical data, auxiliary examination, medical history and other data, to assist doctors in diagnosis. Therefore, it is necessary to preprocess the huge clinical data to provide basic data for the big data analysis prediction model. Summary of the Invention

[0003] The present invention proposes a method for preprocessing rheumatism and immunology disease data based on CL-FCM, extracts the past medical record information of rheumatism patients, and makes full use of patient information by means of technologies such as coding transformation, missing value processing, fuzzy C-means algorithm and multi-view contrast learning mechanism to assist doctors in treatment and help improve the efficiency of medical diagnosis and treatment.

[0004] To achieve the above object, the technical solution adopted by the present invention is: A method for preprocessing rheumatism and immunology disease data based on CL-FCM, comprising the following steps:

[0005] Determine the attributes related to disease diagnosis based on the clinical data of rheumatism and immunology diseases and the expert database;

[0006] Based on the attributes related to the disease diagnosis, screen and collect the available information of the patient in the information, including text data and numerical data, remove redundant information, and form a data set;

[0007] Based on the data set, perform coding transformation and missing value processing on each attribute to obtain a complete data set;

[0008] Use the elbow method, Xie-Beni index and cross-entropy index to determine the optimal number of clusters of the complete data set;

[0009] Based on the optimal number of clusters, perform data augmentation on single or multiple attributes to form multiple view clustering results. According to the multiple view clustering results, calculate the contrast loss and clustering loss, and update the original feature representation through backpropagation to obtain the dataset after updating the original feature representation;

[0010] Use the FCM algorithm to perform final clustering on the dataset after updating the original feature representation to obtain the final number of clusters and the cluster centers.

[0011] Furthermore, determine the attributes related to disease diagnosis based on the clinical data of rheumatic immune diseases and the expert database. By using literature analysis, communication with professional rheumatologists, and analysis of existing medical data, select basic information, family history, specialist conditions, laboratory test results, and diagnosis results as the attributes related to disease diagnosis. The attributes related to disease diagnosis are as follows:

[0012] Basic information: gender and age;

[0013] Family history: whether suffering from a disease similar to the patient;

[0014] Specialist conditions: morning stiffness >= 30 min, involved joints, number of involved joints, degree of joint pain, whether joint movement is restricted, whether the joint is swollen, and bone friction sound;

[0015] Laboratory test results: erythrocyte sedimentation rate, C-reactive protein, antistreptolysin O, rheumatoid factor, blood uric acid, anti-cyclic citrullinated peptide antibody, and antinuclear antibody;

[0016] Diagnosis results: rheumatoid arthritis, osteoarthritis, gout, ankylosing spondylitis, systemic lupus erythematosus.

[0017] Furthermore, based on the attributes related to disease diagnosis, screen the available information of patients in the collected information, remove redundant information, and form a dataset including:

[0018] There is unstructured information in the original dataset. Screen the available information according to the attributes related to disease diagnosis, including text data and numerical data.

[0019] Furthermore, based on the dataset, perform encoding transformation and missing value processing on each attribute, including:

[0020] When performing encoding transformation on each attribute, the screened data needs to be encoded and transformed to form the final dataset. Specifically, for the screened numerical attributes, directly store them in the dataset; for the multiple-choice text attributes, numerically encode the optional values and then store them in the dataset;

[0021] Fill in the missing values in the dataset after encoding transformation. For the attribute features without data values, define a missing null value according to the meaning of the attribute features, and fill it with the mean, median or mode, interpolation method or random sampling method.

[0022] Furthermore, use the elbow method, Xie-Beni index and cross-entropy index to determine the optimal number of clusters of the complete dataset: by comparing the optimal number of clusters obtained by the elbow method, cross-entropy and Xie-Beni index, if the same result is obtained, then determine the result as the optimal number of clusters; otherwise, calculate the optimal number of clusters by using a weighted method.

[0023] Furthermore, use random perturbation, Gaussian noise and feature combination methods when performing data augmentation on single or multiple attributes. The feature combination is based on the combination of existing features to generate new features.

[0024] Furthermore, use the FCM algorithm to perform final clustering on the dataset after updating the original feature representation, including:

[0025] Cluster the updated dataset, determine the center of each cluster according to the input data and the expected number of clusters, and assign each data point to a specific cluster.

[0026] Calculate the minimum distance from each sample point to the centers of each cluster.

[0027] The training steps of the FCM clustering method are as follows:

[0028] Design the number of clusters and hyperparameters, and initialize the relationship matrix.

[0029] Update the cluster centers and update the relationship matrix.

[0030] Repeat updating the cluster centers and updating the relationship matrix until the difference between the relationship matrices obtained from two calculations is less than the set value or remains unchanged.

[0031] Based on the technical concept of the method, the present invention provides a rheumatic immune disease data preprocessing system based on CL-FCM, including a data extraction module, a comparison module and a clustering module;

[0032] The data extraction module is used to screen the information of rheumatic immune disease patients, extract the attributes related to the disease diagnosis, screen and collect the available information of the patients in the information, including text data and numerical data, remove redundant information, and form a dataset.

[0033] The comparison module is used to perform encoding transformation and missing value processing on each attribute based on the dataset to obtain a complete dataset; use multiple indexes such as the elbow method, Xie-Beni index and cross-entropy to determine the optimal number of clusters.

[0034] The clustering module performs data augmentation on single or multiple attributes based on the optimal number of clusters, forms a multi-view clustering result, calculates the contrast loss and the clustering loss, updates the original feature representation through backpropagation, and obtains a dataset after updating the original feature representation; uses the FCM algorithm to perform final clustering on the dataset after updating the original feature representation.

[0035] The present invention can also provide a computer device, including a processor and a memory. The memory is used to store computer-executable programs. The processor reads the computer-executable programs from the memory and executes them. When the processor executes the computer-executable programs, it can implement the method for preprocessing rheumatic immune disease data based on CL-FCM of the present invention.

[0036] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it can implement the method for preprocessing rheumatic immune disease data based on CL-FCM of the present invention.

[0037] Compared with the prior art, the present invention has at least the following beneficial effects:

[0038] The present invention uses technologies such as encoding transformation, missing value processing, fuzzy C-means clustering algorithm, and multi-view contrast mechanism to preprocess rheumatic immune disease data. Compared with the existing methods for preprocessing rheumatic immune disease data, it encodes the multi-selective text attributes using the encoding transformation technology to obtain a normalized dataset, and fuses contrast learning with the FCM algorithm to obtain the final number of clusters and the cluster centers. It can accurately determine the key attributes when facing a dataset with a large number of relevant attributes, improve the disease diagnosis efficiency; can reduce redundant features for the subsequent inference model, reduce the amount of calculation, and improve the model accuracy; can help doctors more comprehensively understand the key features of rheumatic diseases, and contribute to improving the diagnosis efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flowchart of a method for preprocessing rheumatic immune disease data based on CL-FCM.

[0040] Figure 2 It is a schematic structural diagram of a device for preprocessing rheumatic immune disease data based on CL-FCM. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the protection scope of the present invention.

[0042] According toFigure 1 The flowchart of the data preprocessing method for rheumatic immune diseases shown in the figure specifically includes the following steps:

[0043] Step 1, collect information of patients with rheumatic immune diseases.

[0044] The present invention collects patient information from different departments of the rheumatology and immunology department of a certain tertiary hospital in Xi'an, and screens 289 admission records with relatively complete information as the original data set. The patient information is saved in the form of plain text, mainly including the patient's basic information, chief complaint, current medical history, past medical history, marital and childbearing history, family history, specialist conditions, laboratory test results, as well as the doctor's diagnosis results, treatment plans, and ward round records, etc.

[0045] Step 2, determine the attributes related to disease diagnosis.

[0046] Based on Step 1, through literature analysis, communication with professional rheumatology and immunology doctors, and analysis of existing medical data, the present invention determines the medical information related to diagnosis, including basic information, family history, specialist conditions, laboratory test results, and diagnosis results, as the main research attributes, specifically as follows:

[0047] Basic information: gender and age;

[0048] Family history: whether suffering from diseases similar to the patient;

[0049] Specialist conditions: morning stiffness >= 30 min, involved joints, number of involved joints, degree of joint pain, whether joint movement is restricted, whether the joint is swollen, and bone friction sound;

[0050] Laboratory test results: erythrocyte sedimentation rate (ESR), C-reactive protein (CPR), antistreptolysin O (ASO), rheumatoid factor (RF), uric acid (UA), anti-cyclic citrullinated peptide antibody (anti-CPP), and antinuclear antibody (ANA);

[0051] Diagnosis results: rheumatoid arthritis, osteoarthritis, gout, ankylosing spondylitis, and systemic lupus erythematosus.

[0052] Step 3, screen the available information of patients in the collected information, remove redundant information, and form a data set.

[0053] The data set used in the present invention is from three hospitals in Xi'an. Since there is a large amount of unstructured information in the original data set and it is stored in text form, it is necessary to screen the available information according to the attributes selected in Step 2, including text data and numerical data, as follows:

[0054] Chief complaint: polyarticular swelling and pain for 1 month

[0055] Present illness history: One month ago, without obvious cause, swelling and pain occurred in the interphalangeal joint of the left great toe. Subsequently, it gradually affected the interphalangeal joint of the left fifth toe, the interphalangeal joint of the right great toe, and both ankles, with the left side being more prominent. There was no rash, no fever, no chills, no dry mouth, no dry eyes, no hair loss, no photosensitivity, no recurrent oral ulcers, and no Raynaud's phenomenon. Therefore, the patient went to the local hospital for treatment, and was diagnosed with "bursitis". Chinese patent medicine was taken orally, and the patient felt slightly improved. More than two weeks ago, swelling and pain occurred in the proximal interphalangeal joint of the second finger of the left hand and pain in the right wrist, accompanied by morning stiffness of both hands for more than half an hour. Two weeks ago, pain occurred in the left shoulder and the left upper arm, accompanied by limited movement. Therefore, the patient went to the local hospital for treatment, underwent small needle-knife treatment, and took Chinese patent medicine orally. The patient felt that the symptoms improved after treatment. Three days ago, pain in the left ankle and left knee occurred again. For further diagnosis and treatment, the patient came to our hospital. Blood routine examination: Hb 83g / L; erythrocyte sedimentation rate: 8mm / 1h; three rheumatism items: RF 754IU / ml, C-reactive protein 2.​34mg / L; anti-CCP: 81.​9IU / ml; anti-mutated citrullinated vimentin 457.​44IU / ml; liver and kidney functions showed no obvious abnormalities; 2019-nCoV novel coronavirus nucleic acid test: negative (-) x200 copies / ml. The patient was admitted to the hospital with the diagnosis of "rheumatoid arthritis" in the outpatient department. Since the onset of the disease, the patient's spirit has been okay, nocturnal rest has been okay, appetite has been okay, bowel movements and urination have been normal, and the weight has not changed significantly.

[0056] Past history: General health status: General. Lumbar disc herniation for more than 2 years; cholelithiasis for more than 2 years; denying a history of hypertension. Denying a history of diabetes. Denying a history of coronary heart disease. Denying a history of hepatitis. Denying a history of tuberculosis. Other disease history: None. Infectious disease history: No infectious disease history, no history of infectious disease contact. Surgical trauma history: Surgical history: Yes, underwent "cesarean section" 17 years ago, trauma history: None. Blood transfusion history: None. Allergy history: Drug allergy history: None; food allergy history: None. Vaccination history: Unknown.

[0057] Marital and childbearing history: Married, age at marriage: 23 years old, spouse is healthy, has given birth, has 1 son, and the health status of the child: Good

[0058] Family history: Father: Alive, Mother: Alive, Siblings: Alive. There is no disease similar to the patient's, and no disease with a family genetic tendency.

[0059] Specialist examination: The spine is without deformity, showing a physiological curvature, without tenderness, and with normal mobility. There is no tenderness or swelling in both shoulders, and they can be lifted; there is no tenderness or swelling in both elbows, and the extension is not limited; there is no swelling or tenderness in both wrists, and the dorsiflexion and flexion are not limited; there is no tenderness or swelling in the proximal interphalangeal joints and metacarpophalangeal joints of both hands, and the fist clenching is not limited; there is no percussion pain in both hips, the pelvic compression test is negative, the adduction, internal rotation, abduction, and external rotation of both hips are not limited, and the bilateral figure-of-four sign is negative; there is tenderness in the left knee, without swelling, and the flexion and extension of both knees are normal; there is tenderness and swelling in the left ankle; there is tenderness in the bilateral fifth interphalangeal joints.

[0060] Diagnosis: Rheumatoid arthritis

[0061] Step 4: Encode and transform the multi-select text attributes in the dataset.

[0062] After encoding and transforming the data filtered in Step 3, a final dataset is formed. Specifically, for the filtered numerical attributes such as erythrocyte sedimentation rate (ESR), C-reactive protein (CPR), anti-streptolysin O (ASO), and rheumatoid factor (RF), they can be directly stored in the dataset; for the multi-select text attributes, the optional values are encoded and then stored in the dataset. The following are the encodings of some multi-select text attributes:

[0063] Basic information: 1 - male, 0 - female;

[0064] Family history: 1 - yes, 0 - no;

[0065] Specialist conditions: Morning stiffness >= 30 min (1 - yes, 0 - no), affected joints (0 - metacarpophalangeal; 1 - proximal interphalangeal joint; 2 - distal interphalangeal joint; 3 - wrist, knee; 4 - first metatarsophalangeal joint, etc.), degree of joint pain (0 - painless; 1 - 3 - mild pain; 4 - 6 - moderate pain; 7 - 9 - severe pain; 10 - excruciating pain), whether joint movement is restricted (1 - yes, 0 - no), whether the joint is swollen (1 - yes, 0 - no), bone friction sound (1 - positive, 0 - negative);

[0066] Diagnosis result: 0 - rheumatoid arthritis, 1 - osteoarthritis, 2 - gout, 3 - ankylosing spondylitis, 4 - systemic lupus erythematosus.

[0067] Step 5: Handle missing values for the incompleteness of the dataset.

[0068] The dataset after encoding and transformation may have incomplete data. The method of filling missing values is used for processing, that is, for the attribute features without data values, according to the meaning of the attribute, a missing null value is defined and filled to obtain a complete dataset.

[0069] Filling missing values is a common processing method, which can retain the samples where the missing values are located and continue to utilize other features at the same time. Common filling methods include filling with mean, median or mode, interpolation method, random sampling, etc.

[0070] The present invention uses the method of filling with mean to process the dataset after encoding conversion, that is, using the mean of the features to fill the missing values, which can maintain the data distribution and has high efficiency.

[0071] Step 6: jointly determine the optimal number of clusters using the elbow method, cross entropy, and Xie-Beni index.

[0072] The present invention uses the fuzzy C-means (FCM) algorithm to cluster the complete data set. Common methods for determining the optimal number of clusters in the FCM clustering algorithm include the elbow method, the Xie-Beni index, cross-entropy, etc. The present invention selects the elbow method, the Xie-Beni index, and cross-entropy to jointly determine the optimal number of clusters, that is, by comparing the optimal number of clusters obtained by the three methods. If similar results are obtained, it is determined as the optimal number of clusters; otherwise, a weighted method is adopted for comprehensive consideration.

[0073] Among them, the elbow method selects the point where the decline rate of the clustering error (objective function value) significantly slows down as the optimal number of clusters by plotting the relationship diagram between the number of clusters and the clustering error.

[0074] The Xie-Beni index balances the compactness and separability of the clustering results. It is obtained by calculating the square of the distance between each sample in the clustering result and its affiliated cluster center, and then dividing by the product of the number of samples and the number of clusters. Usually, the number of clusters with the smallest Xie-Beni index is selected as the optimal number of clusters.

[0075] Cross-Entropy is usually used as a loss function in machine learning to measure the difference between the model prediction and the actual situation. By minimizing the cross-entropy, the predicted probability distribution of the model will be as close as possible to the real situation, thereby improving the accuracy and generalization ability of the model.

[0076] Step 7: Based on the optimal number of clusters, perform data augmentation on single or multiple attributes to form multiple view clustering results, and calculate the contrast loss and the clustering loss.

[0077] Step 7.1: Perform data augmentation on single or multiple attributes respectively.

[0078] Data augmentation technology is a method of generating new data samples by transforming, expanding, or processing the original data. It is often used in training machine learning and deep learning models. It can improve the generalization performance of the model, suppress overfitting, and enhance the robustness of the model without increasing the data acquisition cost. The present invention uses Random Perturbation, Gaussian Noise, and feature combination methods to perform data augmentation on it.

[0079] Among them, Random Perturbation performs random small-scale addition and subtraction operations on numerical features, such as adding a small random value or multiplying by a small random ratio; Gaussian Noise adds noise that follows a Gaussian distribution to numerical features, which can simulate the uncertainty of the data; Feature combination generates new features based on the combination of existing features, increasing the diversity of features.

[0080] As an example, for the attributes of stiffness ≥ 30 min, the number of affected joints, and ESR, data augmentation or deletion is performed respectively, and then data augmentation or deletion is performed on the two attribute columns of CPR and PF, and so on, to form multiple view clustering results;

[0081] Step 7.2, calculate the contrastive loss and the clustering loss.

[0082] The goal of the contrastive loss is to make samples from the same category closer in the feature space and samples from different categories farther away. The calculation formula is as follows:

[0083]

[0084] Among them, a is the view after enhancing the attribute feature (single or multiple), n is the number of samples, h i is all the features of the i-th sample, is all the features after enhancing the i-th sample, and τ is the temperature coefficient;

[0085] The goal of the clustering loss is to prompt the model to divide the data into clusters or clusters with similar features. In the present invention, the enhanced view and the original view are clustered to obtain multiple clustering results, and n samples in the multiple clustering results are calculated to compare the similarity between the clustering results and the true labels. The calculation formula is as follows:

[0086]

[0087] Among them, C is the view of the clustering result of the original dataset, C a is the view of the clustering result after enhancing the attribute a, and N[i][j] is the number of samples that are truly i and divided into j in C a The proportion of samples with the true category i in the total samples, and Q(j) is the proportion of the true category j in the total samples;

[0088] Step 7.3, calculate the overall loss, and backpropagate to update the original feature representation. The overall loss is calculated as follows:

[0089]

[0090] Step 8, use the FCM algorithm to cluster the rheumatic immune disease dataset.

[0091] According to Step 7, use the FCM algorithm to cluster the updated dataset. FCM is a soft clustering method that divides the data into different clusters so that the data in each cluster has similar features; when the input data and the desired number of clusters are given, the center of each cluster is determined, and each data point is assigned to a specific cluster. The steps are as follows:

[0092] Step 8.1, calculate the minimum distance from each sample point to each cluster center, as shown in the following formula:

[0093]

[0094] where N is the number of sample points, C is the number of clusters, m is a hyperparameter, and u ij is the relationship value between the sample point X i and the cluster center C j ;

[0095] Step 8.2, design the number of clusters and the hyperparameter m, and initialize the relationship matrix U;

[0096] Step 8.3, update the cluster center C,

[0097]

[0098] Step 8.4, update the relationship matrix U,

[0099]

[0100] Step 8.5, repeat the steps of updating the cluster center C and updating the relationship matrix U until the difference between the relationship matrices U obtained from two calculations is less than the set value or remains unchanged;

[0101] So far, the preprocessing method ends. The obtained clustering results are sent into the downstream prediction model for disease prediction, and the improvement in the prediction results reflects the effectiveness of this data preprocessing technology.

[0102] In summary, the present invention proposes a data preprocessing method for rheumatic immune diseases based on CL-FCM. Based on the information of rheumatic immune disease patients and under the guidance of rheumatic immune disease clinicians and professional field experts, the attributes related to the disease are determined; the available information is screened, the redundant information is removed, and each attribute is encoded and transformed and the missing value is processed; the cross entropy and elbow method are used to determine the optimal number of clusters; data augmentation is performed on single or multiple attributes to form multiple view clustering results, the overall loss is calculated, and the original feature representation is updated; the FCM algorithm is used to cluster the updated data set, and the obtained clustering results can be directly used in the downstream prediction model for disease prediction. The present invention can accurately enhance the key attributes and improve the disease diagnosis efficiency when facing a data set with a large number of relevant attributes; it can reduce the redundant features and the amount of calculation for the subsequent inference model, and can help doctors understand the key features of rheumatic diseases more comprehensively.

[0103] Based on the technical concept of the method, the present invention provides a data preprocessing system for rheumatic immune diseases based on CL-FCM, including a data extraction module, a comparison module, and a clustering module;

[0104] The data extraction module is used to screen the information of patients with rheumatic immune diseases, extract the attributes related to the disease diagnosis, screen and collect the available information of the patients in the information, including text data and numerical data, remove redundant information, and form a data set;

[0105] The comparison module is used to perform encoding transformation and missing value processing on each attribute based on the data set to obtain a complete data set; use multiple indicators such as the elbow method, Xie-Beni index, and cross-entropy to determine the optimal number of clusters;

[0106] The clustering module, based on the optimal number of clusters, performs data augmentation on single or multiple attributes to form multiple view clustering results, calculates the contrast loss and clustering loss, and updates the original feature representation through backpropagation to obtain the data set after updating the original feature representation; uses the FCM algorithm to perform the final clustering on the data set after updating the original feature representation.

[0107] Figure 2 It is a structural diagram of a rheumatic immune disease data preprocessing system based on CL-FCM, including a data extraction module, a comparison module, and a clustering module:

[0108] The data extraction module is used to screen the electronic medical history information of rheumatic patients, extract medical information related to the diagnosis, including the basic information of the patients, family history, specialist conditions, laboratory test results, and diagnosis results, etc., and perform encoding transformation and normalization processing on the extracted medical information;

[0109] The comparison module performs data augmentation on the original data set using different strategies, forms multiple views with the original data set, compares the feature representations in different views, and calculates the contrast loss, so that the enhanced data can retain the characteristics of the original data set while being distinguishable;

[0110] The clustering module is used to cluster the original data set, the data views formed after data augmentation, and the updated data set to obtain multiple view clustering results, which can reflect patient samples with similar symptoms and are used for disease prediction in the downstream prediction model to improve the accuracy of the prediction model.

[0111] In addition, the present invention can also provide a computer device, including a processor and a memory. The memory is used to store computer-executable programs. The processor reads part or all of the computer-executable programs from the memory and executes them. When the processor executes part or all of the computer-executable programs, it can implement the method for identifying the characteristics of rheumatic immune diseases based on fuzzy logic described in the present invention.

[0112] On the other hand, the present invention provides a computer-readable storage medium. A computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, it can implement the method for identifying the characteristics of rheumatic immune diseases based on fuzzy logic described in the present invention.

[0113] The computer device may be a laptop computer, a desktop computer or a workstation.

[0114] The processor may be a central processing unit (CPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

[0115] For the memory described in the present invention, it may be an internal storage unit of a laptop computer, a desktop computer or a workstation, such as a memory or a hard disk; or an external storage unit may also be adopted, such as a mobile hard disk or a flash card.

[0116] The computer-readable storage medium may include a computer storage medium and a communication medium. The computer storage medium includes volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), solid state drives (SSD) or optical discs, etc. Among them, the random access memory may include resistive random access memory (ReRAM) and dynamic random access memory (DRAM).

Claims

1. A data preprocessing method for rheumatic immune diseases based on CL-FCM, characterized in that, It includes the following steps: Determine the attributes related to disease diagnosis based on the clinical data of rheumatic immune diseases and the expert database; Based on the attributes related to disease diagnosis, screen and collect the available information of patients in the collected information, including text data and numerical data, and remove redundant information to form a data set; Based on the data set, perform encoding transformation and missing value processing on each attribute to obtain a complete data set; Use the elbow method, Xie-Beni index, and cross-entropy index to determine the optimal number of clusters for the complete data set; Based on the optimal number of clusters, perform data augmentation on single or multiple attributes to form multiple view clustering results. According to the multiple view clustering results, calculate the contrast loss and clustering loss, and backpropagate to update the original feature representation to obtain the data set after updating the original feature representation; When performing data augmentation on single or multiple attributes, use random perturbation, Gaussian noise, and feature combination methods. The use of random perturbation, Gaussian noise, and feature combination methods specifically includes: The goal of the contrast loss is to make the samples from the same category closer in the feature space and the samples from different categories farther away. The calculation formula is as follows: Among them, a is the view after enhancing one or more attribute features, n is the number of samples, are all the features of the i-th sample, are all the features after enhancement of the i-th sample, is the temperature coefficient; The goal of the clustering loss is to prompt the model to divide the data into clusters or clusters with similar features. Cluster the enhanced views and the original views, calculate for n samples in multiple clustering results, and compare the similarity between the clustering results and the true labels. The calculation formula is as follows: Among them, C is the clustering result view of the original data set, is the clustering result view after enhancing attribute a, and N[i][j] is the number of samples that are truly i and classified as j in ; is the proportion of samples with the true category i in the total samples, is the proportion of the true category j in the total samples; Calculate the overall loss and backpropagate to update the original feature representation. The overall loss is calculated as follows: ； Use the FCM algorithm to perform final clustering on the data set after updating the original feature representation to obtain the final number of clusters and the cluster centers.

2. The method for preprocessing data of rheumatic immune diseases based on CL-FCM according to claim 1, wherein Determine the attributes related to disease diagnosis based on the clinical data of rheumatic immune diseases and the expert database. Adopt literature analysis, communicate with professional rheumatologists, and analyze existing medical data, and select basic information, family history, specialist conditions, laboratory test results, and diagnosis results as the attributes related to disease diagnosis. The attributes related to disease diagnosis are specifically as follows, Basic information: gender and age; Family history: whether suffering from diseases similar to the patient; Specialist conditions: morning stiffness >= 30 minutes, involved joints, number of involved joints, degree of joint pain, whether joint movement is restricted, whether the joint is swollen, and bone friction sound; Laboratory test results: erythrocyte sedimentation rate, C-reactive protein, anti-streptolysin O, rheumatoid factor, blood uric acid, anti-cyclic citrullinated peptide antibody, and anti-nuclear antibody; Diagnosis results: rheumatoid arthritis, osteoarthritis, gout, ankylosing spondylitis, systemic lupus erythematosus.

3. The method for preprocessing rheumatic immune disease data based on CL-FCM according to claim 1, wherein, Based on the attributes related to disease diagnosis, screen and collect the available information of patients in the collected information, remove redundant information, and the data set formed includes: There is unstructured information in the original data set. According to the attributes related to disease diagnosis, screen the available information, including text data and numerical data.

4. The preprocessing method for rheumatic immune disease data based on CL-FCM according to claim 1, wherein Based on the data set, the encoding transformation and missing value processing of each attribute include: When encoding and transforming each attribute, the filtered data needs to be encoded and transformed to form the final dataset. Specifically, for the filtered numerical attributes, they are directly stored in the dataset; for the multi-select text attributes, the optional values are numerically encoded and then stored in the dataset. Fill in the missing values in the dataset after encoding and transformation. For the attribute features without data values, define a missing null value according to the meaning of the attribute features, and fill it using the mean, median or mode, interpolation method or random sampling method.

5. The preprocessing method for rheumatic immune disease data based on CL-FCM according to claim 1, wherein Use the elbow method, Xie-Beni index and cross-entropy index to determine the optimal number of clusters for the complete dataset: by comparing the optimal number of clusters obtained by the elbow method, cross-entropy and Xie-Beni index, if the same result is obtained, then determine the result as the optimal number of clusters; otherwise, calculate the optimal number of clusters using a weighted method.

6. The method for preprocessing rheumatic immune disease data based on CL-FCM according to claim 1, wherein When performing data augmentation on single or multiple attributes, use random perturbation, Gaussian noise and feature combination methods. The feature combination is based on the combination of existing features to generate new features.

7. The method for preprocessing rheumatic immune disease data based on CL-FCM according to claim 1, wherein Using the FCM algorithm, the final clustering of the dataset after updating the original feature representation includes: For the updated dataset, perform clustering. According to the input data and the expected number of clusters, determine the center of each cluster and assign each data point to a specific cluster. Calculate the minimum distance from each sample point to the centers of each cluster. The training steps of the FCM clustering method are as follows: Design the number of clusters and hyperparameters, and initialize the relationship matrix. Update the cluster centers and update the relationship matrix. Repeat updating the cluster centers and updating the relationship matrix until the difference between the relationship matrices obtained from two calculations is less than the set value or remains unchanged.

8. A data preprocessing system for rheumatic immune diseases based on CL-FCM, characterized in that, It includes a data extraction module, a comparison module and a clustering module. The data extraction module is used to screen the information of patients with rheumatic immune diseases, extract the attributes related to disease diagnosis, screen the available information of patients in the collected information, including text data and numerical data, and remove redundant information to form a dataset. The comparison module is used to perform encoding transformation and missing value processing on each attribute based on the dataset to obtain a complete dataset; use multiple indicators such as the elbow method, Xie-Beni index and cross-entropy to determine the optimal number of clusters. The clustering module, based on the optimal number of clusters, performs data augmentation on single or multiple attributes to form a multi-view clustering result, calculates the contrast loss and clustering loss, and updates the original feature representation through backpropagation to obtain the dataset after updating the original feature representation. Using the FCM algorithm, perform the final clustering on the dataset after updating the original feature representation. When performing data augmentation on single or multiple attributes, use random perturbation, Gaussian noise and feature combination methods, specifically including: The goal of the contrast loss is to make the samples from the same category closer in the feature space and the samples from different categories farther away. The calculation formula is as follows: where a is the view after enhancing one or more attribute features, n is the number of samples, are all the features of the i-th sample, are all the features after enhancement of the i-th sample, is the temperature coefficient; The goal of the clustering loss is to prompt the model to divide the data into clusters or clusters with similar features, perform clustering on the enhanced view and the original view, calculate n samples in multiple clustering results, and compare the similarity between the clustering results and the true labels. The calculation formula is as follows: Among them, C is the clustering result view of the original data set, is the clustering result view after enhancing attribute a, and N[i][j] is the number of samples that are truly i and are divided into j in is the proportion of samples with the true class i in the total samples, is the proportion of the true class j in the total samples; Calculate the overall loss, and update the original feature representation through backpropagation. The overall loss is calculated as follows: 。 9. A computer device, characterized in that, It includes a processor and a memory. The memory is used to store computer-executable programs. The processor reads and executes the computer-executable programs from the memory. When the processor executes the computational executable programs, it can implement the CL-FCM-based data preprocessing method for rheumatic immune diseases described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the CL-FCM-based data preprocessing method for rheumatic immune diseases described in any one of claims 1 to 7.

Images