Machine learning-based MCI and MCI progress-to-dementia prediction system

Abstract

The invention discloses a machine learning-based MCI (Mild Cognitive Impairment) and a system for predicting the progress of MCI into dementia, and the machine learning-based mild cognitive impairment prediction system comprises the following steps: screening candidate factors to obtain prediction factors in people with known mild cognitive impairment states; the method comprises the following steps: taking people with known mild cognitive impairment states as a data set, and dividing the data set into a training set and a test set; inputting the training set into all machine learning models, and training the models to obtain the machine learning models after preliminary training; inputting the test set into the preliminarily trained machine learning model, testing the model, and taking the machine learning model of which the test index value exceeds a set threshold value as a final trained mild cognitive impairment machine learning model; and inputting a plurality of prediction factors of a to-be-detected object into the finally trained mild cognitive impairment machine learning model to obtain a prediction result of whether the to-be-detected object suffers from the mild cognitive impairment or not.

Description

Technical Field

[0001] The present invention relates to the technical field of disease prediction, and in particular to a prediction system for MCI and the progression of MCI to dementia based on machine learning. Background Art

[0002] Mild cognitive impairment (MCI) is a condition that affects cognitive abilities such as memory, language, and attention, and is often considered a precursor to dementia. As the population ages, the prevalence of MCI and dementia is expected to rise, so early detection and intervention are critical. About 38% of MCI patients progress to Alzheimer's disease after 5 years or more of follow-up. Early detection of MCI and accurate prediction of its progression to dementia can provide an opportunity for intervention to slow the progression of cognitive decline and delay the onset of dementia.

[0003] Previous models for predicting MCI have problems such as small sample size, unresolved data imbalance, and lack of feature importance ranking. In addition, the outcome evaluation of MCI uses screening tools to identify high-risk groups rather than hospital diagnosis data. To address this issue, it is necessary to consider the proportion of MCI patients and non-patients diagnosed by doctors in a large sample population, and construct a prediction model after processing the data imbalance. In addition, the feature importance of the predictors is ranked to enhance the interpretability of the model.

[0004] Previous models for predicting the progression of mild cognitive impairment (MCI) to dementia were mainly based on magnetic resonance imaging (MRI) data or a combination of MRI imaging data and cerebrospinal fluid biomarker data. However, models using imaging or cerebrospinal fluid biomarker data are not suitable for predicting the risk of MCI developing dementia in community populations because it is challenging to obtain such data in community settings. Summary of the invention

[0005] In order to address the deficiencies of the prior art, the present invention provides a prediction system for MCI and MCI progression to dementia based on machine learning; alternative methods are considered to predict which MCI patients in the community have a higher risk of dementia. For example, other types of data can be used, such as cognitive assessment, family history, medical history, genetic markers, sociodemographic characteristics, and lifestyle factors. These data are more easily obtained in a community setting, which helps to more accurately predict the risk of MCI patients progressing to dementia, thereby providing more effective intervention measures.

[0006] On the one hand, a mild cognitive impairment (MCI) prediction system based on machine learning is provided, comprising:

[0007] The prediction factor screening module is configured to: obtain prediction factors by screening candidate factors among a population with known mild cognitive impairment, wherein the screening process includes: preliminarily screening candidate factors related to mild cognitive impairment from a number of candidate factors of the mild cognitive impairment population; then, performing a second screening using a variance inflation factor from the preliminarily screened candidate factors; and finally, performing a third screening using a Lasso regression algorithm from the second screened candidate factors to obtain a number of prediction factors;

[0008] A data set construction module is configured to: take a population with known mild cognitive impairment as a data set, and divide the data set into a training set and a test set according to a set ratio;

[0009] The model training and testing module is configured to: input the training set into all machine learning models to train the models and obtain the machine learning models after preliminary training; input the test set into the machine learning models after preliminary training to test the models, and use the machine learning models whose test index values ​​exceed the set threshold as the final trained mild cognitive impairment machine learning models;

[0010] The prediction module is configured to: input several prediction factors of the subject to be detected into the finally trained mild cognitive impairment machine learning model to obtain a prediction result of whether the subject to be detected suffers from mild cognitive impairment.

[0011] On the other hand, a prediction system for the progression of MCI to dementia based on machine learning is provided, including:

[0012] A predictor screening module is configured to: obtain predictors by screening candidate factors among a population of people whose mild cognitive impairment is known to progress to a dementia state or not, wherein the screening process includes: preliminarily screening candidate factors related to the progression of mild cognitive impairment to dementia from a number of candidate factors known to progress mild cognitive impairment to a dementia state; then, performing a second screening using a variance inflation factor from the preliminarily screened factors; and finally, performing a third screening using a Boruta algorithm from the second screened factors to obtain predictors;

[0013] A data set construction module is configured to: take a population whose mild cognitive impairment is known to progress to dementia as a data set, and divide the data set into a training set and a test set according to a set ratio;

[0014] The model training and testing module is configured to: input the training set into all machine learning models, train the models, and obtain the machine learning models after preliminary training; input the test set into the machine learning models after preliminary training, test the models, and use the machine learning models whose test index values ​​exceed the set threshold as the machine learning models of mild cognitive impairment progressing to dementia after final training;

[0015] The prediction module is configured to: input several prediction factors of the subject to be tested who has suffered from mild cognitive impairment into the final trained mild cognitive impairment progression to dementia machine learning model, and obtain the prediction result of the subject to be tested progressing from mild cognitive impairment to dementia.

[0016] The above technical solution has the following advantages or beneficial effects:

[0017] The present invention obtains a final trained mild cognitive impairment machine learning model by performing deep learning on the training set, and based on the final trained mild cognitive impairment machine learning model, can give an accurate prediction result of whether the subject to be tested suffers from mild cognitive impairment.

[0018] The present invention uses data from a large population cohort study, UK Biobank (UKB), to construct a prediction model for the progression of MCI to dementia based on non-brain imaging and non-cerebrospinal fluid biomarker indicators of the community population. By combining epidemiological nested case-control design, time series causal inference of directed acyclic graphs, and machine learning algorithms, predictors of the progression of MCI to dementia are identified and a prediction model is developed. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0020] Figure 1 This is a flow chart of prediction factor screening in Example 1;

[0021] Figure 2 A schematic diagram of data plan points in Example 1;

[0022] FIG3( a ) is a receiver operating characteristic (ROC) curve for predicting MCI in the training set of Example 1;

[0023] FIG3( b ) is a ROC curve for predicting MCI in the cross-validation set of Example 1;

[0024] FIG3( c ) is a ROC curve of the CatBoost model predicting MCI in the test set of Example 1;

[0025] FIG3( d ) is a calibration curve diagram of the CatBoost model predicting MCI in Example 1;

[0026] FIG3( e ) shows the evaluation indicators of each model for predicting MCI in Example 1;

[0027] FIG3( f ) is a SHAP summary diagram of the first 10 features of the CatBoost model predicting MCI in Example 1;

[0028] FIG4( a ) is a receiver operating characteristic (ROC) curve for predicting the progression of MCI to dementia in the training set of Example 1;

[0029] FIG4( b ) is a ROC curve for predicting the progression of MCI to dementia in the cross-validation set of Example 1;

[0030] FIG4( c ) is a ROC curve of the RF model in the test set of Example 1 for predicting the progression of MCI to dementia;

[0031] FIG4( d ) is a calibration curve diagram of the RF model of Example 1 for predicting the progression of MCI to dementia;

[0032] FIG4( e ) shows the evaluation indicators of each model for predicting the progression of MCI to dementia in Example 1;

[0033] Figure 4(f) is a SHAP summary diagram of the top 10 features of the RF model for predicting the progression of MCI to dementia in Example 1; *Each data point in the SHAP diagram represents a participant, and the color indicates the size of the predictor. The horizontal range represents the overall predictive power of each predictor (the larger the range, the greater the contribution to the model). The size and direction on the x-axis can be used to determine the likelihood of an outcome event occurring (the right side indicates a greater likelihood of suffering an outcome event).

[0034] FIG5( a ) is an example of an application platform for predicting MCI based on the CatBoost model in Embodiment 1;

[0035] FIG5( b ) is an example of an application platform for predicting MCI based on the CatBoost model of Embodiment 1;

[0036] FIG5( c ) is an example of an application platform for predicting MCI based on the CatBoost model of Embodiment 1;

[0037] FIG6( a ) is an example of an application platform for predicting the progression of MCI to dementia based on the Random Forest model in Example 1;

[0038] FIG6( b ) is an example of an application platform for predicting the progression of MCI to dementia based on the Random Forest model of Example 1;

[0039] FIG6( c ) is an example of an application platform for predicting the progression of MCI to dementia based on the Random Forest model of the first embodiment. DETAILED DESCRIPTION

[0040] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0041] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0042] In this embodiment, all data is obtained in compliance with laws and regulations and based on the user's consent, and is used legally.

[0043] Embodiment 1

[0044] This embodiment provides an MCI prediction system based on machine learning;

[0045] like Figure 1 As shown, the mild cognitive impairment (MCI) prediction system based on machine learning includes:

[0046] The prediction factor screening module is configured to: obtain prediction factors by screening candidate factors among a population with known mild cognitive impairment, wherein the screening process includes: preliminarily screening candidate factors related to mild cognitive impairment from a number of candidate factors of a population with mild cognitive impairment; then, performing a second screening using a variance inflation factor from the preliminarily screened candidate factors; and finally, performing a third screening using a Lasso regression algorithm from the second screened candidate factors to obtain a number of prediction factors;

[0047] A data set construction module is configured to: take a population with known mild cognitive impairment as a data set, and divide the data set into a training set and a test set according to a set ratio;

[0048] The model training and testing module is configured to: input the training set into all machine learning models to train the models and obtain the machine learning models after preliminary training; input the test set into the machine learning models after preliminary training to test the models, and use the machine learning models whose test index values ​​exceed the set threshold as the final trained mild cognitive impairment machine learning models;

[0049] The prediction module is configured to: input several prediction factors of the subject to be detected into the finally trained mild cognitive impairment machine learning model to obtain a prediction result of whether the subject to be detected suffers from mild cognitive impairment.

[0050] Furthermore, several candidate factors for the mild cognitive impairment population include:

[0051] Category 1: Genetic factors, including: APOE e4 allele carrier number and ABO blood type;

[0052] Category II: Early life factors, including birth weight and breastfeeding;

[0053] Category 3: Demographic factors, including age, gender, race, education level, income level, Townsend deprivation index;

[0054] The fourth category: midlife environmental factors, including: PM 2.5 Pollution, PM 10 Pollution, NO2 pollution, noise pollution (L den );

[0055] The fifth category: middle-aged lifestyle factors, including smoking, drinking, physical activity, sedentary time, sleep time, and diet;

[0056] Category 6: middle-aged biological factors, including body mass index (BMI), grip strength, diastolic blood pressure, systolic blood pressure, low-density lipoprotein, glycated hemoglobin level, insulin use, and vitamin D level;

[0057] Category 7: Social and psychological factors in middle age, including depression, tension, anxiety, loneliness, and social isolation;

[0058] Category 8: Medical history; the medical history includes: disability, cardiovascular disease, metabolic diseases, respiratory system diseases, digestive system diseases, musculoskeletal system diseases and cancer; the disability includes the use of hearing aids, hearing loss, and physical weakness; the cardiovascular diseases include coronary heart disease, stroke, heart failure, and hypertension; the metabolic diseases include diabetes, obesity, and dyslipidemia; the respiratory system diseases include influenza and pneumonia, chronic obstructive pulmonary disease, and asthma; the digestive system diseases include inflammatory bowel disease, gastroesophageal reflux disease, peptic ulcer disease, and irritable bowel syndrome; the musculoskeletal system diseases include gout, osteoarthritis, and rheumatoid arthritis; the cancers include lymphohematopoietic system cancer, digestive system cancer, endocrine system cancer, reproductive system cancer, respiratory system cancer, and urinary system cancer.

[0059] Furthermore, the candidate factors related to mild cognitive impairment are initially screened out from among several candidate factors of people who have suffered from mild cognitive impairment, including:

[0060] A matched nested case-control study was used to examine the relationship between candidate factors and mild cognitive impairment.

[0061] Furthermore, the matched nested case-control study was used to examine the relationship between candidate factors and mild cognitive impairment, including:

[0062] The mild cognitive impairment case and control samples were from the same cohort, and the matching date was defined as the date of onset of mild cognitive impairment in the current case (index date).

[0063] The control sample was defined as individuals who did not experience mild cognitive impairment on the index date.

[0064] Greedy matching was used, with each case matched to a maximum of 10 controls, based on the principle that the age of the controls was within the range of ±2 years and the age of the cases was the same sex. Greedy matching is a commonly used matching method in case-control studies, which aims to control confounding variables and improve the validity of research results. Greedy matching matches 10 control samples for each case by gradually selecting the control samples that are most similar to each case, calculating the distance, and ensuring that the matched controls are not selected repeatedly. The advantage of this method is that it can reduce confounding bias, improve statistical power, and enhance the interpretability of the results.

[0065] For the nested case-control study, multivariable conditional logistic regression models were used to estimate the strength of association between candidate characteristics and the risk of mild cognitive impairment (ORs).

[0066] The expression of the multivariate conditional logistic regression model is as follows:

[0067]

[0068] Y ij represents the outcome variable for the jth observation of the ith case, Y ij =1 indicates an event occurs (such as a case), Y ij =0 means the event did not occur (such as control). kij represents the kth independent variable (covariate) in the jth observation of the i-th case, which is used to explain the change in the outcome variable. The independent variable can be continuous or categorical. β0 represents the intercept term, which means that when all independent variables X kij When β is zero, the logarithmic probability of the event occurring. k represents the regression coefficient of the kth independent variable, representing the independent variable X kijThe change in the log odds of an event occurring per unit increase in the probability of occurrence.

[0069] Specifically, exp(β k ) represents the change in relative risk of an event occurring when the independent variable increases by one unit.

[0070] The formula for calculating OR is:

[0071] OR = (a / b) / (c / d);

[0072] Among them, a represents the number of cases in the exposed group, b represents the number of controls in the exposed group, c represents the number of cases in the non-exposed group, and d represents the number of controls in the non-exposed group.

[0073] If the confidence interval of the OR value does not include 1, it means that the candidate factor is related to mild cognitive impairment; if the confidence interval of the OR value includes 1, it means that the candidate factor is not related to mild cognitive impairment.

[0074] Furthermore, the second screening using variance inflation factors from the candidate factors screened initially includes:

[0075] For all candidate factors that are initially screened, the variance inflation factor (VIF) is calculated. If the VIF value of the current candidate factor is greater than the set threshold, the current candidate factor is deleted. If the VIF value of the current candidate factor is less than or equal to the set threshold, the current candidate factor is retained.

[0076] For example, to avoid multicollinearity, the variance inflation factor (VIF) is used to detect multicollinearity, and variables with VIF values ​​greater than 10 are eliminated.

[0077] Furthermore, the Lasso regression algorithm is used to perform a third screening on the candidate factors screened for the second time to obtain several predictive factors, including:

[0078] Sparsity is achieved by adding a penalty term (L1 regularization) to the optimization objective function, so that the weights of many features in the coefficient vector become 0. By selecting the features corresponding to non-zero coefficients, the features with the greatest predictive power for the target variable can be screened out, thereby simplifying the model, improving the generalization ability of the model, and obtaining several prediction factors.

[0079]

[0080] Among them, Minβ represents the minimization target, which is a function of coefficient β, N is the sample size, p is the number of parameters, represents L1 regularization (penalty term); under the condition that the sum of the absolute values ​​of the L1 penalty coefficients is less than a constant, some coefficients are forced to 0; y i is the actual value of the i-th sample, x ij is the jth eigenvalue of the i-th sample, β0 is the intercept term; β j is the coefficient of feature j, and λ is the regularization parameter, which controls the weight of the penalty term.

[0081] The flowchart of predictor screening is as follows Figure 1 shown.

[0082] Furthermore, the population with known mild cognitive impairment is used as a data set, and the data set is divided into a training set and a test set according to a set ratio. When performing MCI risk prediction, 501,172 MCI and non-MCI patients are divided into a training group, a cross-validation group, and a test group. First, a training set (80% of the population) and a test set (20% of the population) are created, and then the training set is divided into 10 partitions of uniform size to form 10 different training sets and cross-validation sets.

[0083] Furthermore, all the machine learning models include: Multi-Layer Perceptron (MLP), Categorical Boosting (CatBoost), Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Gradient Boosting Decision Tree (GBDT), Decision Tree (Decision Tree), Adaptive Boosting (AdaBoost), Logistic Regression (LG), Linear Support Vector Classification (LinearSVC), and eXtreme Gradient Boosting (XGBoost).

[0084] It should be understood that 1 deep learning model (MLP) and 9 machine learning models (CatBoost, RF, LightGBM, GBDT, Decision Tree, AdaBoost, LG, LinearSVC, XGBoost) were selected as candidate prediction models. Models that require hyperparameter tuning are tuned using random search with cross validation.

[0085] In data with imbalanced classes, many machine learning methods "bias" towards the class with a larger proportion. In the prediction of MCI, the ratio of MCI to non-MCI patients is about 1:50, so the EasyEnsemble algorithm is used to solve the extremely imbalanced dataset.

[0086] Furthermore, the test set is input into the machine learning model after preliminary training, the model is tested, and the machine learning model whose test index value exceeds the set threshold is used as the final trained mild cognitive impairment machine learning model, wherein the test index value includes:

[0087] The area under the ROC (Receiver operating characteristic, ROC) curve (Area Under the Curve, AUC), accuracy, precision, recall and F1 score indicators.

[0088] For example, model evaluation and comparison: AUC is used to compare multiple models. According to the AUC value obtained by cross-validation, the prediction model with the highest performance is selected for testing on the test set. The AUC values ​​are classified as follows: 0.5-0.7, 0.7-0.8, 0.8-0.9, and >0.9 represent acceptable, general, good, and excellent models, respectively. In addition, accuracy, precision, recall, and F1 score are calculated to evaluate the model.

[0089] Furthermore, the system further comprises:

[0090] The model explanation module is configured to: input several predictive factors of people who have suffered from mild cognitive impairment into the interpretable machine learning model SHAP (SHapley Additive Explanations) to obtain the importance value of each predictive factor to the model.

[0091] Furthermore, the several predictors of people with mild cognitive impairment are input into the interpretable machine learning model SHAP to obtain the importance value of each predictor to the model. The SHapley Additiveexplanation (SHAP) method is used to quantify the importance of features in the optimal model and visualize the results.

[0092] In order to enhance the interpretability of the model, the present invention uses the SHAP method to quantify the importance of features, calculates the contribution of each input feature to the MCI and MCI progression to dementia outcomes of each participant, and assigns an importance score (i.e., SHAP value) to each feature. To visualize the results, honeycomb diagrams and bar charts of the predictors are drawn, and the features are sorted according to the absolute value of the SHAP value.

[0093] The model calibration performance was evaluated by comparing the true probability with the predicted probability, and a calibration curve was drawn. The closer to the diagonal line, the better the model performance. The random forest method was used to interpolate variables with a missing percentage of less than 10%.

[0094] like Figure 5(a) to Figure 5(c) As shown, the embodiments of the present application also develop an online platform so that medical institutions and researchers can easily use these models to predict the risk of MCI.

[0095] By solving the above technical problems, the present invention finally determined that CatBoost is the best model for predicting MCI, with an AUC of 0.839 and an accuracy of 77.7%.

[0096] People with mild cognitive impairment, and people with mild cognitive impairment progressing to dementia, data source: This invention uses data from a large community-based cohort study (UKB data), which recruited more than 500,000 participants aged 40-70 years from 2006 to 2010. Written informed consent was obtained from participants when collecting questionnaires and biological data. All participant data were linked to hospital data and national death registry records from England, Scotland and Wales to determine the date of the first diagnosis of MCI and dementia after the baseline assessment. UKB has obtained approval from the Northwest Multicenter Research Ethics Committee (https: / / www.ukbiobank.ac.uk / learn-more-about-uk-biobank / about-us / ethics) and supervised this study. First, patients with dementia at baseline were excluded. When constructing the prediction model for MCI, 501,172 people remained after further excluding patients with MCI at baseline. 12,031 MCI patients at baseline and follow-up were included in the construction of the prediction model for the progression of MCI to dementia.

[0097] Outcome definitions: International Classification of Diseases-10th (ICD-10) codes F067, G318, and R41 were used to identify patients with MCI. Patients with dementia were identified within the MCI population using the algorithm outlined in UKB category 47. MCI and dementia outcomes were finalized by the UK Biobank Disease Outcomes Adjudication Panel.

[0098] Existing MCI prediction models are typically based on small samples of dozens to hundreds of people. These predictors typically use routinely collected electronic health records, clinical imaging examinations, or community-acquired survey data, with AUC values ​​ranging from 0.62 to 0.75. Although models that use features from brain imaging tests (such as positron emission tomography) to predict MCI generally perform better, obtaining these data is both expensive and infeasible in community settings. In addition, these models lack a ranking of feature importance, and the interpretability of the models needs to be improved.

[0099] Based on a large sample of 501,172 people, the present invention uses 16 non-brain imaging data predictors (APOE e4 allele carrier number, age, sleep duration, frailty, heart failure, stroke, hypertension, diabetes, dyslipidemia, influenza and pneumonia, chronic obstructive pulmonary disease, gastroesophageal reflux disease, osteoarthritis, lymphatic hematopoietic system cancer, digestive system cancer, respiratory system cancer) to build 10 models to predict the occurrence of MCI, with an AUC value of 0.744-0.859 for the training set and an AUC value of 0.739-0.835 for the cross-validation set. Finally, CatBoost was determined to be the optimal model, with a test set AUC of 0.839, which is higher than the AUC index of previous studies using non-image test data, and the accuracy, precision, recall, and F1 score are 0.780, 0.972, 0.780, and 0.857, respectively. The calibration curve is close to the diagonal, indicating that the predicted probability is close to the true probability. In addition, the SHAP graph showed that age, hypertension, number of APOE e4 allele carriers, diabetes, influenza and pneumonia, osteoarthritis, stroke, dyslipidemia, heart failure and chronic obstructive pulmonary disease were the top 10 predictors that contributed most to the occurrence of MCI. Figure 3(a) to Figure 3(f) shown.

[0100] Embodiment 2

[0101] This embodiment provides a prediction system for the progression of MCI to dementia based on machine learning, including:

[0102] A predictor screening module is configured to: obtain predictors by screening candidate factors among a population of people whose mild cognitive impairment is known to progress to a dementia state or not, wherein the screening process includes: preliminarily screening candidate factors related to the progression of mild cognitive impairment to dementia from a number of candidate factors known to progress mild cognitive impairment to a dementia state; then, performing a second screening using a variance inflation factor from the preliminarily screened candidate factors; and finally, performing a third screening using a Boruta algorithm from the factors screened for the second time to obtain predictors;

[0103] A data set construction module is configured to: take a population whose mild cognitive impairment is known to progress to dementia as a data set, and divide the data set into a training set and a test set according to a set ratio;

[0104] The model training and testing module is configured to: input the training set into all machine learning models, train the models, and obtain the machine learning models after preliminary training; input the test set into the machine learning models after preliminary training, test the models, and use the machine learning models whose test index values ​​exceed the set threshold as the machine learning models of mild cognitive impairment progressing to dementia after final training;

[0105] The prediction module is configured to: input several prediction factors of the subject to be tested who has suffered from mild cognitive impairment into the final trained mild cognitive impairment progression to dementia machine learning model, and obtain the prediction result of the subject to be tested progressing from mild cognitive impairment to dementia.

[0106] Furthermore, the several candidate factors for the progression of the known mild cognitive impairment to the dementia state are consistent with the several candidate factors of the mild cognitive impairment population in Example 1.

[0107] Furthermore, the factors related to the progression of mild cognitive impairment to dementia are preliminarily screened out from among several candidate factors known to cause mild cognitive impairment to progress to dementia. The preliminary screening process is consistent with the screening process of Example 1.

[0108] Furthermore, the second screening using the variance inflation factor from the initially screened factors is consistent with the screening process of the first embodiment in which the variance inflation factor is used to perform the second screening from the initially screened candidate factors.

[0109] Furthermore, all the machine learning models in Example 2 are consistent with all the machine learning models in Example 1.

[0110] Furthermore, the system also includes: a model interpretation module, which is configured to: input several predictive factors of the progression of mild cognitive impairment to dementia into an interpretable machine learning model SHAP package to obtain the importance value of each predictive factor to the model.

[0111] Furthermore, the Boruta algorithm is used to perform a third screening on the factors screened twice to obtain prediction factors. The Boruta algorithm generates a random forest through multiple self-service resampling, calculates the importance of features, and introduces shadow features (generated by disrupting the original features) for comparison. If the importance of the original feature is significantly higher than that of the shadow feature, it is considered important, otherwise it is marked as irrelevant. This process is iterated for multiple rounds, and finally the truly important features are screened out.

[0112] Furthermore, the data set is divided into a training set and a test set according to a set ratio; when predicting the risk of MCI progressing to dementia, the data set of 12031 MCI patients is divided in the same way as in Example 1. Figure 2 In the prediction of MCI progression to dementia, the ratio of MCI patients who progressed to dementia to those who did not progress to dementia was approximately 1:4, and random oversampling was used to solve the data imbalance problem.

[0113] In order to enhance the interpretability of the model, the present invention uses the SHAP method to quantify the importance of features, calculates the contribution of each input feature to the MCI progression to dementia outcome of each participant, and assigns an importance score (i.e., SHAP value) to each feature. To visualize the results, honeycomb diagrams and bar graphs of the predictors are drawn, and the features are sorted according to the absolute value of the SHAP value.

[0114] In data with imbalanced categories, many machine learning methods "bias" towards the category with a larger proportion. In the prediction of MCI, the ratio of MCI to non-MCI patients is about 1:50, so the EasyEnsemble algorithm is used to solve the extremely imbalanced dataset. In the prediction of MCI progression to dementia, the ratio of MCI patients who progress to dementia to those who do not progress to dementia is about 1:4, and random oversampling is used to solve the data imbalance problem.

[0115] like Figure 6(a) to Figure 6(c) As shown, the embodiments of the present application also develop an online platform so that medical institutions and researchers can easily use these models to predict the risk of MCI progressing to dementia.

[0116] By solving the above technical problems, the present invention finally determined that Random Forest is the best model for predicting MCI progression to dementia, with an AUC of 0.887 and an accuracy rate of 80.1%.

[0117] In addition, an online platform was developed as a tool to facilitate the application of these models, providing an important reference for early intervention and treatment of MCI progression to dementia. Existing prediction models for MCI progression to dementia are mainly developed based on brain imaging data from MRI examinations in small sample populations of dozens to thousands, and are not suitable for promotion and application in community populations. There are few prediction models based on non-brain imaging data in community environments, and the AUC values ​​range from 0.36 to 0.72. The performance of the models needs to be further improved.

[0118] Based on a large sample of 12,031 people, the present invention uses 17 non-brain imaging data predictors (APOE e4 allele carrier number, glycosylated hemoglobin level, insulin use status, social isolation, income level, smoking status, BMI, grip strength, coronary heart disease, influenza and pneumonia, osteoarthritis, education level, sedentary time, chronic obstructive pulmonary disease, heart failure, hypertension, and diabetes status) to construct 10 models to predict the occurrence of MCI progression to dementia. The AUC value of the training set is between 0.636-0.980, and the AUC value of the cross-validation set is between 0.627-0.876. Finally, the RF model with good performance was determined, and the test set AUC was 0.875, which is higher than the AUC index of previous studies using non-imaging data. The accuracy, precision, recall rate, and F1 score are 0.751, 0.797, 0.793, and 0.792, respectively. The calibration curve is close to the diagonal line, indicating that the predicted probability is close to the true probability. In addition, the SHAP graph showed that the number of APOE e4 allele carriers, income level, smoking status, BMI, coronary heart disease, education level, sedentary time, chronic obstructive pulmonary disease, heart failure, and diabetes status were the top 10 predictors that contributed the most to the prediction of the progression of MCI to dementia. Figure 4(a) to Figure 4(f) shown.

[0119] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A mild cognitive impairment prediction system based on machine learning, characterized by including: The prediction factor screening module is configured to: obtain prediction factors by screening candidate factors among a population with known mild cognitive impairment, wherein the screening process includes: preliminarily screening candidate factors related to mild cognitive impairment from a number of candidate factors of the mild cognitive impairment population; then, performing a second screening using a variance inflation factor from the preliminarily screened candidate factors; and finally, performing a third screening using a Lasso regression algorithm from the second screened candidate factors to obtain a number of prediction factors; A data set construction module is configured to: take a population with known mild cognitive impairment as a data set, and divide the data set into a training set and a test set according to a set ratio; The model training and testing module is configured to: input the training set into all machine learning models to train the models and obtain the machine learning models after preliminary training; input the test set into the machine learning models after preliminary training to test the models, and use the machine learning models whose test index values ​​exceed the set threshold as the final trained mild cognitive impairment machine learning models; The prediction module is configured to: input several prediction factors of the subject to be detected into the finally trained mild cognitive impairment machine learning model to obtain a prediction result of whether the subject to be detected suffers from mild cognitive impairment.

2. The machine learning-based mild cognitive impairment prediction system according to claim 1, characterized in that: The several candidate factors for the mild cognitive impairment population include: the first category: genetic factors; the second category: early life factors; the third category: demographic factors; the fourth category: middle-aged environmental factors; the fifth category: middle-aged lifestyle factors; the sixth category: middle-aged biological factors; the seventh category: middle-aged social and psychological factors; the eighth category: disease history.

3. The machine learning-based mild cognitive impairment prediction system according to claim 1, characterized in that: The method preliminarily screens out candidate factors related to mild cognitive impairment from several candidate factors in the mild cognitive impairment population, including: using a matched nested case-control study to examine the relationship between the candidate factors and mild cognitive impairment.

4. The machine learning-based mild cognitive impairment prediction system according to claim 3, characterized in that: The described matched nested case-control study was used to examine the relationship between candidate factors and mild cognitive impairment, including: mild cognitive impairment cases and control samples were from the same cohort, the matching date was defined as the date of occurrence of mild cognitive impairment in the current case; the control sample was defined as an individual who did not experience mild cognitive impairment on the index date; for the nested case-control study, a multivariate conditional logistic regression model was used to estimate the association strength OR value between the candidate characteristics and the risk of mild cognitive impairment; the expression of the multivariate conditional logistic regression model is as follows: Y ij represents the outcome variable for the jth observation of the ith case, Y ij =1 indicates the event occurs, Y ij =0 means the event did not occur, X kij represents the kth independent variable in the jth observation of the i-th case, which is used to explain the change of the outcome variable; the independent variable is continuous or categorical, β0 represents the intercept term, and β0 represents the sum of all independent variables X kij When β is zero, the logarithmic probability of the event occurring, β k represents the regression coefficient of the kth independent variable, representing the independent variable X kij The change in the logarithmic probability of an event occurring for each unit increase; exp(β k ) represents the change in relative risk of an event when the independent variable increases by one unit. The calculation formula of OR is: OR = (a / b) / (c / d); where a represents the number of cases in the exposed group, b represents the number of controls in the exposed group, c represents the number of cases in the non-exposed group, and d represents the number of controls in the non-exposed group. If the confidence interval of the OR value does not include 1, it means that the candidate factor is related to mild cognitive impairment; if the confidence interval of the OR value includes 1, it means that the candidate factor is not related to mild cognitive impairment.

5. The machine learning-based mild cognitive impairment prediction system according to claim 1, characterized in that: The variance inflation factor is used to conduct a second screening of the candidate factors from the preliminary screening, including: For all candidate factors that are initially screened, the variance inflation factor (VIF) is calculated. If the VIF value of the current candidate factor is greater than the set threshold, the current candidate factor is deleted. If the VIF value of the current candidate factor is less than or equal to the set threshold, the current candidate factor is retained.

6. The machine learning-based mild cognitive impairment prediction system according to claim 1, characterized in that: The Lasso regression algorithm is used to perform a third screening on the candidate factors screened the second time, and several predictive factors are obtained, including: Among them, Minβ represents the minimization target, which is a function of coefficient β, N is the sample size, p is the number of parameters, stands for L1 regularization; under the condition that the sum of the absolute values ​​of the L1 penalty coefficients is less than a constant, some coefficients are forced to 0; y i is the actual value of the i-th sample, x ij is the jth eigenvalue of the i-th sample, β0 is the intercept term; β j is the coefficient of feature j, and λ is the regularization parameter, which controls the weight of the penalty term.

7. The machine learning-based mild cognitive impairment prediction system according to claim 1, characterized in that: All machine learning models described include: multilayer perceptron, class boosting model, random forest model, lightweight gradient boosting machine, gradient boosting decision tree model, decision tree model, adaptive boosting model, logistic regression model, linear support vector classification model or extreme gradient boosting model.

8. The machine learning-based mild cognitive impairment prediction system according to claim 1, characterized in that: The test set is input into the machine learning model after preliminary training to test the model. The machine learning model whose test index value exceeds the set threshold is used as the final trained mild cognitive impairment machine learning model, where the test index values ​​include: the area under the ROC curve, accuracy, precision, recall rate and F1 score indicators.

9. The machine learning-based mild cognitive impairment prediction system according to claim 1, characterized in that: The system also includes: a model interpretation module, which is configured to: input several predictive factors of people who have suffered from mild cognitive impairment into an interpretable machine learning model SHAP to obtain the importance value of each predictive factor to the model.

10. A prediction system for the progression of MCI to dementia based on machine learning, characterized by: A predictor screening module is configured to: obtain predictors by screening candidate factors among a population of people whose mild cognitive impairment is known to progress to a dementia state or not, wherein the screening process includes: preliminarily screening candidate factors related to the progression of mild cognitive impairment to dementia from a number of candidate factors known to progress mild cognitive impairment to a dementia state; then, performing a second screening using a variance inflation factor from the preliminarily screened factors; and finally, performing a third screening using a Boruta algorithm from the second screened factors to obtain predictors; A data set construction module is configured to: take a population whose mild cognitive impairment is known to progress to dementia as a data set, and divide the data set into a training set and a test set according to a set ratio; The model training and testing module is configured to: input the training set into all machine learning models, train the models, and obtain the machine learning models after preliminary training; input the test set into the machine learning models after preliminary training, test the models, and use the machine learning models whose test index values ​​exceed the set threshold as the machine learning models of mild cognitive impairment progressing to dementia after final training; The prediction module is configured to: input several prediction factors of the subject to be tested who has suffered from mild cognitive impairment into the final trained mild cognitive impairment progression to dementia machine learning model, and obtain the prediction result of the subject to be tested progressing from mild cognitive impairment to dementia.

Images