Cognitive dysfunction prediction system based on gastrointestinal electrical signals and construction method

Abstract

The present invention relates to the field of disease prediction, and in particular to a system and method for predicting cognitive impairment based on gastrointestinal electrical signals. In one aspect, the present invention provides a cognitive impairment prediction system, comprising: a database for storing data, the types of data including gastrointestinal electrical signal data and clinical data, the gastrointestinal electrical signal data being postprandial gastrointestinal electrical signal data; the postprandial gastrointestinal electrical signal data including postprandial gastric dominant frequency, postprandial gastric electrical rhythm disruption rate, postprandial gastric dominant power ratio, and postprandial intestinal lead time difference; a data acquisition module; a model training module; and a prediction module for predicting the probability of cognitive impairment in a subject. In another aspect, the present invention also provides a method for constructing the cognitive impairment prediction system.

Description

[Technical Field]

[0001] This application claims priority to a patent application entitled "System and Construction Method for Predicting Cognitive Dysfunction Based on Gastrointestinal Electrical Signals," China Patent Application Publication No. 2022115029929, filed on November 25, 2022. The entire contents of this priority invention patent application are incorporated by reference.

[0002] The present invention relates to the field of disease prediction, and in particular to a system and method for predicting cognitive dysfunction based on gastrointestinal electrical signals. [Background technology]

[0003] Early detection of disease is extremely important. It is believed that the earlier a disease is diagnosed, the greater the chance that the disease will be cured (or successfully managed) and the better the patient's prognosis. Early screening and treatment of disease can prevent or delay further progression of the disease and improve the effectiveness of treatment (e.g., extending the patient's lifespan and improving the patient's quality of life).

[0004] In recent years, the number of patients with chronic neurological diseases has been steadily increasing in China. However, some chronic neurological diseases (e.g., cognitive dysfunction, sleep disorders, anxiety, or depression) are often not diagnosed until they have progressed to a stage where intervention or treatment is relatively difficult, due to factors such as inconspicuous early symptoms, low patient awareness, and complex and costly testing procedures. Therefore, mass screening of target populations (e.g., middle-aged and elderly people) is necessary for early disease detection. However, mass screening requires processing a huge amount of data and usually relies on manual analysis. The entire process is time-consuming, laborious, and costly. Furthermore, the results of data analysis are highly subjective, complex, and difficult to quantify. Therefore, it is difficult to popularize mass screening for many diseases, especially for the early detection of cognitive dysfunction.

[0005] Cognitive impairment is a condition or disorder characterized by impaired cognitive function. Cognitive impairment involves one or more impairments in memory, learning, language, executive function, visuospatial function, calculation, and comprehension / judgment, significantly impacting patients' activities of daily living and social and occupational functioning. Patients with cognitive impairment may also experience mental, behavioral, and personality disorders (e.g., anxiety, depression, and physical disabilities) at some stage in the course of the disease, placing a significant burden on their families and society. Furthermore, statistics show that 15% to 20% of patients with mild cognitive impairment progress to Alzheimer's disease each year.

[0006] Conventional diagnostic methods for cognitive impairment-related disorders include basic clinical symptom assessment, neuropsychological status assessment, imaging tests such as computed tomography (CT), magnetic resonance imaging (MRI), and single-photon emission computed tomography (SPECT), and bodily fluid tests such as regular blood tests, cerebrospinal fluid tests, and antibody tests. Specifically, when a patient exhibits symptoms of cognitive impairment, the patient or their family visits a hospital. During the interview, doctors typically provide the patient with a paper scale (combination) and perform an initial assessment based on the patient's test results, which are then combined with other test results to identify patients suspected of having cognitive impairment. Therefore, the entire process is complex, cumbersome, time-consuming, labor-intensive, inefficient, and dependent on the doctor's personal experience. In addition, patients may have low awareness of cognitive impairment-like diseases or a low level of cooperation with diagnosis, or may be unable to communicate with doctors or complete measures independently due to reasons such as impaired hearing, vision, comprehension, or low literacy levels, and may rely on family members to convey information. This makes early detection of cognitive impairment (especially through mass screening at the grassroots level or in broad populations) and timely, targeted treatment and prevention relatively difficult.

[0007] Although there are existing internet-based cognitive impairment screening and assessment systems, they are complex and time-consuming, requiring steps such as filling out online screening scales and remote consultations with doctors. Furthermore, they are not suitable for people with low literacy levels, poor comprehension, or who are unfamiliar with electronic devices. As a result, their scope of application remains limited. Summary of the Invention

[0008] In a first aspect, the present invention provides a system for predicting cognitive impairment based on gastrointestinal electrical signals, comprising:

[0009] a database for storing data, the types of data including gastrointestinal electrical signal data and clinical data, the gastrointestinal electrical signal data being postprandial gastrointestinal electrical signal data; the postprandial gastrointestinal electrical signal data including postprandial gastric dominant frequency, postprandial gastric electrical rhythm disruption rate, postprandial gastric dominant power ratio, and postprandial intestinal lead time difference, and the clinical data including age, body mass index, high density lipoprotein, and low density lipoprotein; and the data including sample data from a sample population and subject data from a subject;

[0010] a data acquisition module for acquiring the data and storing it in the database;

[0011] a model training module that constructs a cognitive impairment prediction model by training and learning the sample data using a machine learning algorithm;

[0012] a prediction module that acquires the subject data through the data acquisition module and predicts the probability of the subject developing cognitive impairment by calling the cognitive impairment prediction model and analyzing the subject data.

[0013] In some embodiments, the sample data is split into a training set and a validation set in a 7:3 ratio.

[0014] In some embodiments, the gastrointestinal electrical signal data is collected simultaneously by leads located in the gastric body, gastric antrum, lesser curvature, greater curvature, ascending colon, transverse colon, descending colon, and rectum.

[0015] In some embodiments, the system further comprises a validation module that uses the validation set to evaluate the accuracy of the cognitive impairment prediction model.

[0016] In some embodiments, the evaluation metrics include one or more of calibration, discrimination, and clinical utility.

[0017] In some embodiments, the cognitive impairment is mild cognitive impairment.

[0018] In a second aspect, the present invention provides a method for constructing a cognitive impairment prediction system based on gastrointestinal electrical signals, wherein the cognitive impairment prediction system based on gastrointestinal electrical signals includes a cognitive impairment prediction model, and the method includes the following steps:

[0019] acquiring sample data from an S1 sample population, wherein the sample data types include gastrointestinal electrical signal data and clinical data;

[0020] S2: Pre-training the sample data to select predictive variables, the selected predictive variables including postprandial gastric dominant frequency, postprandial gastric electrical rhythm disruption rate, postprandial gastric dominant power ratio, postprandial intestinal lead time difference, age, body mass index, high density lipoprotein, and low density lipoprotein;

[0021] S3: constructing the cognitive impairment prediction model by training and learning the sample data using a machine learning algorithm based on the selected predictive variables, and further constructing a cognitive impairment prediction system based on the gastrointestinal electrical signals.

[0022] In some embodiments, the sample data is split into a training set and a validation set in a 7:3 ratio.

[0023] In some embodiments, the pre-training comprises a first round of variable selection and a second round of variable selection; the first round of variable selection comprises LASSO regression analysis, and the second round of variable selection comprises logistic regression analysis and stepwise multiple regression analysis.

[0024] In some embodiments, the method further comprises evaluating the accuracy of the cognitive impairment prediction model using the S4 validation set, wherein the evaluation metrics include one or more of calibration, discrimination, and clinical utility.

[0025] In some embodiments, the pre-training comprises a ridge regression or random forest model. [Effects of the Invention]

[0026] The present invention has the following advantageous effects compared to the conventional techniques.

[0027] The present invention provides a system and method for predicting cognitive impairment based on gastrointestinal electrical signals. The present invention performs pre-training (feature selection) on 46 feature variables in sample data through LASSO regression analysis, logistic regression analysis, and stepwise multiple regression analysis, ultimately retaining eight predictor variables: postprandial gastric dominant frequency, postprandial gastric electrical rhythm disruption rate, postprandial gastric dominant power ratio, postprandial intestinal lead time difference, age, body mass index (BMI), high-density lipoprotein (HDL), and low-density lipoprotein (LDLP). A cognitive impairment prediction model is constructed based on these eight predictor variables.

[0028] Conventional techniques often diagnose cognitive impairment through scales, memory tests, and imaging tests. Patients with mild cognitive impairment exhibit declines in memory, language, and other cognitive functions, but may not exhibit abnormalities in daily life or social functioning. Because early symptoms are often undetectable, the optimal time for intervention may be missed. By the time patients or those around them realize that they may be suffering from cognitive impairment, the patient may have already progressed to a stage where intervention or treatment is relatively difficult, or may have even progressed to irreversible Alzheimer's disease. Based on this, the cognitive impairment prediction model underlying the cognitive impairment prediction system and method provided by the present invention requires only some gastrointestinal electrical indicators and basic clinical information (e.g., age, BMI, high-density lipoprotein, low-density lipoprotein) of the subject, and can predict the subject's risk of developing cognitive impairment without the need for various tests or scales. This contributes to early screening of cognitive impairment and reducing the risk of progression to other diseases (e.g., Alzheimer's disease).

[0029] The cognitive impairment prediction system and method provided by the present invention do not require expensive (and sometimes invasive) tests such as body fluid tests or imaging tests, and do not require subjects to fill out scales, making the procedure simple and inexpensive, thereby increasing subject acceptance and cooperation. The cognitive impairment prediction system and method provided by the present invention are not limited by age or literacy level, and are not affected by communication or comprehension difficulties during the examination process or the physician's personal experience. Therefore, they are capable of analyzing and predicting subject data relatively objectively and non-invasively, making them particularly applicable to early screening and early detection of cognitive impairment in large populations (e.g., communities and health checkup centers). As described above, the cognitive impairment prediction system and method provided by the present invention are useful not only for supporting clinical evaluation but also for individualized prediction, making them applicable to a wide range of application scenarios (e.g., grassroots medical institutions, homes, hospitals, health checkup centers) and populations.

[0030] In order to more clearly describe the technical solutions in the embodiments of the present invention or the prior art, the following briefly describes the drawings that need to be used to describe the embodiments or the prior art. In all the drawings, similar elements or parts are generally denoted by similar reference numerals. In the drawings, each element or part is not necessarily drawn to actual scale. Obviously, the drawings in the following description are some embodiments of the present invention, and those skilled in the art can also obtain other drawings from these drawings without creative efforts. [Brief explanation of the drawings]

[0031] [Figure 1a] FIG. 1a is a schematic diagram illustrating a LASSO regression model according to a first embodiment of the present invention.

[0032] [Figure 1b] FIG. 1b is a schematic diagram showing the 10-fold cross-validation result of the first embodiment of the present invention.

[0033] [Figure 2a] FIG. 2a is a schematic diagram showing the nomograph of the optimal logistic regression model of embodiment 1 of the present invention.

[0034] [Figure 2b] FIG. 2b is a schematic diagram showing the dynamic nomograph of the optimal logistic regression model of embodiment 1 of the present invention.

[0035] [Figure 3a] FIG. 3a is a schematic diagram showing the prediction model ROC curve of the test set of embodiment 1 of the present invention.

[0036] [Figure 3b] FIG. 3b is a schematic diagram showing the prediction model ROC curve of the validation set of embodiment 1 of the present invention.

[0037] [Figure 4a] FIG. 4a is a schematic diagram showing the calibration curve of the test set of embodiment 1 of the present invention.

[0038] [Figure 4b] FIG. 4b is a schematic diagram showing the calibration curve of the validation set according to embodiment 1 of the present invention.

[0039] [Figure 5a] FIG. 5a is a schematic diagram showing the decision curve of the test set of embodiment 1 of the present invention.

[0040] [Figure 5b] FIG. 5b is a schematic diagram showing the decision curve of the validation set of embodiment 1 of the present invention.

[0041] [Figure 6] FIG. 6 is a schematic diagram showing lead placement locations in an embodiment of the present invention.

[0042] [Figure 7] FIG. 7 is a schematic diagram showing the configuration of a prediction system provided by an embodiment of the present invention.

[0043] [Figure 8] FIG. 8 is a schematic block diagram illustrating a prediction system provided by an embodiment of the present invention.

[0044] [Figure 9] FIG. 9 is a diagram summarizing the 46 related feature variables incorporated in the first embodiment of the present invention.

[0045] [Figure 10] FIG. 10 is a diagram showing the results of the first variable selection in the first embodiment of the present invention.

[0046] [Figure 11] FIG. 11 is a diagram showing the results of the second variable selection in the first embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0047] In order to make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It is obvious that the described embodiments are only a part of the embodiments of the present invention, and are not all of them. All other embodiments that can be obtained by those skilled in the art based on the embodiments of the present invention without any creative ingenuity are included in the protection scope of the present invention.

[0048] As used herein, "and / or" includes any and all combinations of one or more of the associated listed items.

[0049] As used herein, "plurality" means two or more, ie, including two, three, four, five, etc.

[0050] It should be noted that, as used herein, the terms "comprise," "include," or any other variation thereof, are intended to cover a non-exclusive inclusion, and thus a process, method, article, or apparatus comprising a set of elements includes not only those elements but also other elements not expressly listed or elements inherent in such process, method, article, or apparatus. An element qualified by the phrase "comprising a" does not exclude the presence of other identical elements in a process, method, article, or apparatus that includes that element, unless further limited.

[0051] As used herein, the term "about" is typically + / - 5% of the stated value, more typically + / - 4% of the stated value, more typically + / - 3% of the stated value, more typically + / - 2% of the stated value, even more typically + / - 1% of the stated value, and even more typically + / - 0.5% of the stated value.

[0052] Certain embodiments may be disclosed herein in a range format. It should be understood that this "within a range" description is for convenience and brevity only and should not be construed as a rigid limitation on the disclosed range. Accordingly, the description of a range should be considered to have specifically disclosed all possible subranges and individual numerical values ​​within that range. For example, description of a range of 1 to 6 should be considered to have specifically disclosed subranges of 1 to 3, 1 to 4, 1 to 5, 2 to 4, 2 to 6, 3 to 6, etc., as well as individual numerical values ​​within that range of 1, 2, 3, 4, 5, 6, etc. The above rule applies regardless of the breadth of the range.

[0053] Embodiment 1: Gastrointestinal electrical signals and cognitive dysfunction prediction model

[0054] 1.1 Method

[0055] Participants: Community residents aged 40 years or older were recruited from 60 communities in the western region of China. They volunteered to participate in the study and signed informed consent. General information, including age, sex, marital status, literacy level, lifestyle, and dietary habits, was collected, and participants underwent the Mini-Mental State Examination (MMSE) and electrogastroenterology (EGG) testing. Exclusion criteria included the following: hearing and visual impairments; those diagnosed with gastrointestinal disorders such as gastritis or gastric ulcers within the past 6 months; those with severe cardiac, hepatic, renal, or other major organ failures; or those with metabolic disorders such as diabetes; and those with severe mental illnesses. To mitigate the effects of medication on EGG testing, participants were excluded from the study.

[0056] Cognitive assessment: The Mini-Mental State Examination (MMSE) was used to assess cognitive function. This scale has good reliability and validity in testing cognitive impairment. The scale was completed by two experts in a quiet environment, with guidance from the subjects, using standardized terminology.

[0057] EGEG recording: Gastrointestinal myoelectric activity signals were measured and collected using an 8-channel electrogastrointestinal (EGG) device (XDJ-S8, Hefei Kaili Co., Hefei, China). All subjects were instructed to avoid alcohol consumption and spicy or pungent foods for three days and to fast for at least six hours before the test. Measurements were performed in the supine position. Four gastric electrodes (leads placed at the gastric body 601, lesser curvature 602, greater curvature 603, and gastric antrum 604, respectively) and four intestinal electrodes (leads placed at the ascending colon 605, transverse colon 606, descending colon 607, and rectum 608, respectively) were placed on the abdominal skin (Hanjie Co. Ltd., Shanghai, China) (Figure 6). Subjects were instructed to refrain from moving or talking during the test. After 6 minutes of pre-prandial EGEG recording, a meal-induced functional challenge test was performed. After a standard meal of approximately 200 kcal, postprandial gastrointestinal electrical signals were recorded for 6 minutes. The lead placements are shown in Figure 6: gastric body 601: 3-5 cm leftward and 1 cm upward from the midpoint of the line connecting the xiphoid process and the umbilicus; gastric antrum 604: 2-4 cm rightward from the midpoint of the line connecting the xiphoid process and the umbilicus; lesser curvature 602: 1 / 2 upward from the midpoint of the line connecting the xiphoid process and the umbilicus; greater curvature 603: 1 / 2 downward from the midpoint of the line connecting the xiphoid process and the umbilicus; ascending colon 605: 2-4 cm rightward at the same level as the umbilicus; transverse colon 606: 1 cm below the umbilicus; descending colon 607: 2-4 cm leftward at the same level as the umbilicus; and rectum 608: below the coccyx on the back.

[0058] Gastrointestinal electrical indices: The EGEG sampling frequency was 1 Hz, and the filtering frequency was 0.008 Hz to 0.1 Hz to remove background noise, including heartbeat. After artifact detection, the raw EGEG data was calculated using the device's software. Spectral analysis was performed using the software to derive the following parameters for each of the eight leads: (1) mean waveform amplitude, (2) mean waveform frequency, (3) gastric (intestinal) electrical rhythm disruption rate, (4) waveform response area, (5) lead time difference, (6) dominant frequency, (7) dominant power ratio, (8) normal slow wave rate, and (9) coupling rate.

[0059] Other indicators: Each subject underwent measurements of blood glucose and lipid indices, including glucose, triglycerides, cholesterol, high-density lipoprotein, and low-density lipoprotein. Basic personal characteristics, including gender, age, smoking history, drinking history, and BMI, were also collected from each subject.

[0060] Prediction model construction: First, a first round of variable selection was performed using LASSO regression analysis, a regularization algorithm, to obtain a subset of predictors. Furthermore, 10-fold cross-validation was performed in the LASSO regression analysis, and the included variables were centered and normalized. "lambda.min" was selected as the optimal performance. Subjects were randomly divided into a training set and a validation set in a 7:3 ratio. Next, a second round of variable selection was performed using stepwise multivariate logistic regression analysis on the predictors selected in the LASSO regression model. A prediction model was constructed using the retained statistically significant predictors (in this specification, "predictors" and "predictor variables" have the same meaning). Finally, the constructed prediction model was applied to predict the risk of cognitive impairment and to construct a nomographic prediction model. It should be understood that other suitable algorithms known in the art, such as random forests, other regularization methods (e.g., ridge regression), neural networks, etc., can also be used.

[0061] Additionally, we employed several validation methods to evaluate the accuracy of the risk prediction model using the training and validation data. These validation methods included receiver operating characteristic (ROC) curves, in which the area under the ROC curve was used to distinguish true positives from false positives (i.e., discrimination), calibration curves to perform the Hosmer-Lemeshow test while assessing the calibration of the cognitive impairment risk nomograph, and decision curve analysis to determine the clinical utility of the cognitive impairment risk nomograph based on net benefit under different threshold probabilities in natural population cohorts. All analyses were performed using the software packages glmnet and rms (R4.1.3), with a significance level of two-sided α < 0.1.

[0062] 1.2 Results

[0063] Subject Data Information:

[0064] A total of 886 subjects, including 273 men and 613 women, completed all relevant tests, of which 106 (71 men and 35 women) were diagnosed with mild cognitive impairment (MCI). The subjects' gastrointestinal electrical indices were obtained by averaging the parameterized index data from the eight leads before and after a meal, representing the pre-meal gastric lead signal index, the post-meal gastric lead signal index, the pre-meal intestinal lead signal index, and the post-meal intestinal lead signal index, respectively. By employing a method of simultaneously collecting signals from multiple leads at multiple locations and then averaging them, the overall motility pattern of the stomach and intestine can be better captured, enabling more effective acquisition of signals that can reflect the true overall state of the stomach and intestine. Furthermore, preliminary experimental tests of the present invention showed that the signal indices obtained by multi-point signal collection using the above method were relatively stable, contributing to the construction of a model, and the constructed model was found to have high versatility even for large populations.

[0065] All subjects were randomly assigned to the training and validation sets using a 7:3 ratio, with 620 and 266 subjects assigned to the training and validation sets, respectively. In both datasets, highly statistically significant differences (p<0.01) were observed between mild cognitive impairment (MCI) and non-MCI subjects in terms of age, high-density lipoprotein (HDL), and postprandial gastric main frequency (SMFA).

[0066] Screening for independent risk factors:

[0067] The technical solution of the present invention employs a LASSO regression-based non-zero coefficient feature variable selection method to select and retain 13 of the 46 relevant feature variables (also referred to as independent variables (IVs) in the present invention) (FIG. 9) as potential predictor variables for the artificial intelligence model (FIGS. 1a and 1b, detailed description of which follows). These feature variables are used to predict the response variable (also referred to as dependent variables (DVs) in the present invention) (i.e., the risk of developing cognitive impairment). These include alcohol consumption, age, BMI, high-density lipoprotein (HDLP), low-density lipoprotein (LDLP), mean postprandial gastric frequency, dominant postprandial gastric frequency, postprandial gastric electrical rhythm disruption rate, postprandial gastric power ratio, postprandial bowel electrical rhythm disruption rate, postprandial bowel lead time difference, preprandial gastric power ratio, and preprandial gastric normal slow wave rate (FIG. 10).

[0068] The LASSO (Least Absolute Shrinkage and Selection Operator) regression analysis adopted in the technical solution of the present invention is a method for shrinking a linear regression model and selecting characteristic variables. To obtain a predictor subset, LASSO regression analysis minimizes the prediction error of the response variable by imposing constraints on model parameters to shrink the regression coefficients (abbreviated as "coefficients" in Figure 1a) of some characteristic variables toward zero. After the shrinkage process, characteristic variables with regression coefficients equal to zero are excluded from the model, while characteristic variables with non-zero regression coefficients have the strongest correlation with the response variable. The parameter λ is used to adjust the complexity of LASSO regression. Specifically, the larger λ is, the stronger the penalty for linear regression models with many characteristic variables. Therefore, a model with fewer characteristic variables, all of which are strongly correlated with the response variable (i.e., a model with optimal predictive performance) is finally obtained. Specifically, as shown in Figure 1a, each curve in the figure represents the change trajectory of the regression coefficient of the corresponding characteristic variable, where the vertical axis represents the value of the regression coefficient, the lower horizontal axis represents Log(λ), and the upper horizontal axis represents the number of non-zero regression coefficients in the model. Specifically, for example, the independent variable m in Figure 1a has a non-zero regression coefficient when the value of λ is large and continues to increase as the value of λ decreases. In other words, in the first round of characteristic variable selection, we mainly excluded characteristic variables from the 46 related characteristic variables whose regression coefficients tend to shrink to zero, but retained the above 13 characteristic variables as predictor variables for the prediction model.

[0069] Furthermore, to more accurately evaluate the performance of the predictive model based on the above 13 predictor variables, the technical solution of the present invention uses a log-likelihood function (-2log-likelihood) and a binary dependent variable (which can be understood as a "Yes / No" variable) (i.e., the target parameter to be minimized during model selection by cross-validation) to perform 10-fold cross-validation using LASSO regression analysis, centering and normalizing the 46 included feature variables, and then selecting the optimal λ value. As shown in Figure 1b, through cross-validation, for each λ value, the black dot represents the mean value of the target parameter, the solid lines above and below the black dot represent the confidence interval of the target parameter, and the two dashed lines represent two specific λ values ​​(i.e., Lambda.min and Lambda.lse), respectively. Any λ value between these two λ values ​​is considered appropriate. The model constructed using Lambda.1se (Lambda.1se represents the λ value that obtains the simplest model within one variance range of Lambda.min) is the simplest (i.e., uses the fewest number of predictor variables), while the model constructed using Lambda.min (Lambda.min represents the smallest mean value of the target parameter among all λ values) has higher accuracy. Therefore, the technical solution of the present invention used "Lambda.min" to construct a predictive model with optimal performance and highest accuracy.

[0070] Developing predictive models:

[0071] The technical solution of the present invention uses stepwise multivariate logistic regression analysis to build a prediction model by introducing feature variables selected from the LASSO regression model, and then analyzes the statistical significance level of the selected feature variables by introducing them, and uses some statistically significant feature variables as predictor variables / predictors to build a cognitive impairment risk prediction model.

[0072] The technical solution of the present invention further employs a logistic regression model to analyze the above 13 predictor variables and select the optimal predictor variables through a stepwise method, ultimately retaining eight predictor variables (each with statistical significance at a test level of 0.1). These eight predictor variables are age, BMI, high-density lipoprotein, low-density lipoprotein, postprandial gastric dominant frequency, postprandial gastric electrical rhythm disruption rate, postprandial gastric dominant power ratio, and postprandial intestinal lead time difference (Figure 11). The technical solution of the present invention tests the above eight characteristic variables using multiple statistical methods, particularly analyzing the odds ratio (OR) of these characteristic variables. The OR value is a statistic that quantifies the strength of the association between two events and represents the ratio of the probability of an outcome occurring after exposure (i.e., the tested characteristic variable in this invention, hereinafter the same) to the probability of an outcome occurring in the absence of the same exposure. In the present invention, the OR value can be specifically understood as the strength of the association between cognitive impairment (i.e., the response variable) and exposure, and indicates that the risk of cognitive impairment (which can also be understood as disease risk) in an exposed individual is multiple of the risk of cognitive impairment in an unexposed individual. When the OR value of the tested characteristic variable is greater than 1, the risk of cognitive impairment increases with exposure, indicating a "positive" association between the characteristic variable and cognitive impairment. When the OR value of the tested characteristic variable is less than 1, the risk of cognitive impairment decreases with exposure, indicating a "negative" association between the characteristic variable and cognitive impairment. When the OR value of the tested characteristic variable is 1, there is no association between cognitive impairment and the characteristic variable. The 95% confidence interval (CI) provides an estimate of the precision of the OR value obtained by the test, which indicates that the overall true value may vary within the 95% confidence interval of the OR value obtained by the test, and the smaller the confidence interval, the more precise and robust the OR value obtained by the test. OR (top) and OR (bottom) in Figure 11 represent the confidence intervals (95% CI) of the OR values ​​of the tested characteristic variables.Figure 11 shows the odds ratios (ORs) and their 95% confidence intervals for the eight retained predictor variables, implying that all of these predictor variables have a certain association with cognitive impairment and are therefore applied to the cognitive impairment prediction model of the present invention.

[0073] Also, in Figure 11, the regression coefficient β is the partial regression coefficient that indicates the relationship between the predictor variable and the response variable obtained by logistic regression model analysis, and represents the magnitude and direction of the impact of an increase per unit amount on the response variable (partial regression coefficients become comparable after standardization, and the regression coefficient β of the tested predictor variable is an estimated value). Standard error is the standard error of the regression coefficient β of the tested predictor variable, and represents the precision of the regression coefficient (the larger the standard error, the lower the precision of the tested predictor variable). Z value is the z statistic, i.e., the regression coefficient β of the tested predictor variable divided by its corresponding standard error, and is mainly used to determine the P value of the tested predictor variable. P value is the P value corresponding to the z statistic of the tested predictor variable (the smaller the P value, the more significant the tested predictor variable is with respect to the response variable).

[0074] Based on the above eight predictor variables, this embodiment constructs a cognitive dysfunction risk prediction model and plots a corresponding nomogram to better visualize the constructed cognitive dysfunction prediction model (see Figures 2a and 2b, where *P<0.05 is considered statistically significant and **P<0.01 is considered highly statistically significant). Figures 2a and 2b show different representations of the nomogram of the cognitive dysfunction prediction model constructed by the present invention. In Figures 2a and 2b, the line segments corresponding to each variable (e.g., "age," "body mass index (BMI)," etc.) are marked with scales representing the range of values ​​of the variable, and the length of the line reflects the magnitude of the variable's contribution to the outcome event (i.e., the occurrence of cognitive dysfunction). At different values, each variable can obtain a corresponding single-item score in the "Score" or "Regression Coefficient β" at the top of Figure 2a or 2b. After obtaining the values, the corresponding single-item scores of all variables can be summed to obtain a "total score." Based on the total score, the probability of occurrence of cognitive impairment can be obtained in the "Risk of Mild Cognitive Impairment (MCI)" or "Probability of Mild Cognitive Impairment (MCI)" at the bottom of Figure 2a or Figure 2b. For example, in Figure 2a, if a subject's total score is 300, the risk of occurrence of mild cognitive impairment (MCI) is 0.3 (30%). Furthermore, in the dynamic nomograph shown in Figure 2b, the black dots on the line corresponding to each variable indicate the actual value for a subject, and the waveform diagram above the line indicates the specific distribution of each variable. According to the position of the black dots, the corresponding single item score can be obtained in the "Regression Coefficient β" above, and then the total score can be calculated, and the corresponding probability of mild cognitive impairment (MCI) can be obtained (in the example of Figure 2b, the subject's risk of occurrence of mild cognitive impairment (MCI) is 0.571).

[0075] Validation of the predictive model:

[0076] The present invention evaluated the sensitivity (which may also be understood as the true positive rate) and specificity (which may also be understood as the true negative rate) of the constructed prediction model by plotting the corresponding subject operating characteristic (ROC) curves using the data from the training set and validation set. In Figures 3a and 3b, the x-axis represents the "false positive rate," i.e., "1-specificity," and the y-axis represents the "true positive rate," i.e., "sensitivity." Analysis of the area under the ROC curve (i.e., the solid line in Figures 3a and 3b) (AUC, i.e., the area enclosed by the ROC curve and the coordinate axes) is used to identify the quality of the risk nomograph and distinguish true positives from false positives. For the constructed prediction models, the areas under the receiver operating characteristic curves (AUCs) of the nomographs were all 0.6 or greater (i.e., greater than the area under the dashed line): 73.75% (95% CI: 67.72%-79.78%) for the training set (Figure 3a) and 72.63% (95% CI: 64.36%-80.9%) for the validation set (Figure 3b). This indicates that the model constructed by the present invention exhibits good robustness. Calibration curves are used to observe whether the predicted probabilities are close to the actual probabilities. The calibration curves of the nomographs for both datasets also showed good agreement (Figures 4a and 4b; the dashed curve (denoted by d) represents the actual observed probability of cognitive impairment, and the solid curve (denoted by c) represents the probability of cognitive impairment predicted by the prediction model). Taking these validation results together, it can be seen that the cognitive impairment prediction model constructed by the present invention has relatively good predictive ability.

[0077] Furthermore, decision curve analysis (DCA) revealed that the threshold range of cognitive impairment risk accurately predicted by the nomograph of this embodiment in the training set was between 11% and 36% ( FIG. 5 a), while the threshold range in the validation set was between 13% and 32% ( FIG. 5 b). This indicates that the threshold probabilities of both datasets overlap between 13% and 32%. The overlap between the training and validation sets in the threshold probability range of the DCA curve can be interpreted as indicating that both the training and validation sets can demonstrate effective net benefit within this overlapping range. That is, if patients with probabilities exceeding the threshold probability are identified as true positives and intervention is performed, a positive net benefit can be achieved in both the training and validation sets. In other words, patients with a predicted probability higher than 0.32 (32%) are considered to have a higher risk of cognitive impairment, which is the clinical reference threshold for a net benefit. This also suggests that setting the probability threshold in this overlapping range provides greater practical value, and the magnitude of the benefit can be compared using a benefit rate.

[0078] Embodiment 2: Predicting the probability of cognitive impairment in a subject using the cognitive impairment prediction model

[0079] Referring to FIG. 7, FIG. 7 is a schematic diagram illustrating an optional configuration of a prediction system provided by an embodiment of the present invention. To support an exemplary application, terminals (illustratively shown as a first terminal 202 and a second terminal 204) are connected to the prediction system via a network. The network according to the present invention may be a wide area network, a local area network, or a combination of both, which realizes data transmission using wireless links. The terminal according to the present invention may be various types of user terminals, such as a smartphone, a tablet computer, or a laptop computer. The terminal may be used to display an interface for inputting subject data and / or sample data and an interface for displaying the prediction results of the prediction system.

[0080] An exemplary configuration of the prediction system is described below. In some embodiments, as shown in FIG. 8, the prediction system 100 includes:

[0081] a database 106 for storing data, the types of data including gastrointestinal electrical signal data and clinical data, the data including sample data from a sample population and subject data from a subject;

[0082] a data acquisition module 102 for acquiring and storing said data in said database 106;

[0083] a model training module 108 that constructs a predictive model (e.g., a cognitive impairment predictive model) by training and learning the sample data using a machine learning algorithm;

[0084] It may also include a prediction module 112 that acquires the subject data through the data acquisition module 102 and predicts the probability of the subject developing cognitive impairment by calling a constructed prediction model (e.g., a cognitive impairment prediction model) to analyze the subject data.

[0085] The prediction system 100 may further include a validation module 110 that evaluates the accuracy of the constructed prediction model (e.g., a cognitive impairment prediction model), and the evaluation indicators of the evaluation include one or more of calibration, discrimination, and clinical utility.

[0086] Here, the database 106 , the model training module 108 , and the validation module 110 may be integrated into a model building module 104 .

[0087] For example, in the cognitive impairment prediction system of embodiment 1, the gastrointestinal electrical signal data specifically includes postprandial gastric dominant frequency, postprandial gastric electrical rhythm disruption rate, postprandial gastric dominant power ratio, and postprandial intestinal lead time difference, and the clinical data specifically includes age, body mass index (BMI), high-density lipoprotein, and low-density lipoprotein.

[0088] Specific application scenarios of the present invention will be described below.

[0089] Community A conducts a large-scale screening activity for cognitive impairment to collect gastrointestinal electrical signal data, blood glucose and lipid data, and personal characteristic data (the latter two are collectively referred to as "clinical data") from a target population (e.g., middle-aged and elderly people) within the community, and inputs the subject data via a first terminal 202. The subject data is transmitted to a data acquisition module 102 of the prediction system 100 via a network 206. The data acquisition module 102 acquires the subject data from the subjects and stores it in a database 106.

[0090] The prediction module 112 obtains the subject data, analyzes the subject data by invoking the constructed cognitive impairment prediction model, and predicts the probability of the subject developing cognitive impairment.

[0091] As an output, the prediction module 112 may generate a report presenting the subject's risk of cognitive dysfunction and transmit the prediction result to the first terminal 202 via the network 206. The local community A may preset a risk threshold for cognitive dysfunction (e.g., a 30% probability of cognitive dysfunction). If the predicted risk of a certain patient (e.g., patient B) developing cognitive dysfunction exceeds the set risk threshold (e.g., a 40% probability of cognitive dysfunction), the local community A may issue a warning to patient B or his / her family and recommend that the patient visit a recommended or affiliated hospital. If patient B visits the hospital, the hospital may conduct further detailed examinations on patient B to determine whether the patient suffers from the presented cognitive dysfunction. The doctor may transmit the diagnosis result of patient B to the prediction system 100 via the second terminal 204. Patient B's data (subject data + diagnosis result) may be used as new sample data for further training of the cognitive dysfunction prediction model. Of course, the diagnosis result of patient B may be transmitted to the prediction system 100 via the first terminal 202; in other words, the terminal transmitting the diagnosis result of patient B may be the same or different.

[0092] From the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be realized by a combination of software and a required general-purpose hardware platform, or of course by hardware, and in many cases, the former is a more preferred embodiment. Based on this understanding, the technical solution of the present invention may essentially be embodied in the form of a computer software product, or a portion that contributes to the prior art may be embodied in the form of a computer software product, which is stored in a storage medium (e.g., ROM / RAM, magnetic disk, optical disk) and includes a plurality of instructions for causing a computer terminal (which may be a smartphone, computer, server, network device, etc.) to execute the method according to each embodiment of the present invention.

[0093] Although the embodiments of the present invention have been described with reference to the drawings, the present invention is not limited to the above-mentioned specific embodiments. The above-mentioned embodiments are merely illustrative and not limiting. Those skilled in the art can make many other forms under the teachings of the present invention without departing from the spirit of the present invention and the scope of protection of the claims, and all of these fall within the scope of protection of the present invention. [Explanation of symbols]

[0094] 100 Prediction System 102 Data Acquisition Module 104 Model Building Module 106 databases 108 Model Training Module 110 Verification Module 112 Prediction Module 202 First Terminal 204 Second Terminal 206 Network 601 Body of stomach 602 Kobuki 603 Great Curve 604 Gastric antrum 605 Ascending colon 606 Transverse colon 607 Descending colon 608 Rectum

Claims

1. A cognitive impairment prediction system based on gastrointestinal electrical signals, A database for storing data, wherein the types of data include gastrointestinal electrical signal data and clinical data, the gastrointestinal electrical signal data being postprandial gastrointestinal electrical signal data including the dominant frequency of the stomach after a meal, the percentage of electrical rhythm disturbance of the stomach after a meal, the dominant power ratio of the stomach after a meal, and the lead time difference of the intestine after a meal; the clinical data including age, body mass index, high-density lipoprotein, and low-density lipoprotein; and the database including sample data from a sample population and subject data from subjects; A data acquisition module that acquires the aforementioned data and stores it in the aforementioned database; A model training module that constructs a cognitive impairment prediction model by training and learning on the aforementioned sample data using a machine learning algorithm; A prediction module that acquires subject data through the data acquisition module and predicts the probability of the subject developing cognitive impairment by calling the cognitive impairment prediction model and analyzing the subject data; A system characterized by including

2. The system according to claim 1, characterized in that the sample data is divided into a training set and a validation set in a ratio of 7:

3.

3. The system according to claim 1, characterized in that the gastrointestinal electrical signal data is simultaneously collected by leads located in the body of the stomach, the gastric antrum, the lesser curvature, the greater curvature, the ascending colon, the transverse colon, the descending colon, and the rectum, respectively.

4. The system according to claim 2, further comprising a validation module for evaluating the accuracy of the cognitive impairment prediction model using the validation set, wherein the evaluation indicators for the evaluation include one or more of calibration, discrimination and clinical utility.

5. The system according to claim 1, characterized in that the cognitive impairment is mild cognitive impairment.

6. A method for constructing a cognitive impairment prediction system based on gastrointestinal electrical signals, wherein the cognitive impairment prediction system based on gastrointestinal electrical signals includes a cognitive impairment prediction model, and the method comprises the following steps: S1 A step of obtaining sample data from a sample population, wherein the type of sample data includes gastrointestinal electrical signal data and clinical data; S2 A step of pre-training on the sample data to select predictor variables, wherein the selected predictor variables include the major postprandial gastric frequency, the percentage of postprandial gastric electrical rhythm disturbance, the major postprandial gastric power ratio, the postprandial intestinal lead time difference, age, body mass index, high-density lipoprotein, and low-density lipoprotein; S3 A step of constructing the cognitive impairment prediction model by training and learning the sample data using a machine learning algorithm based on the selected predictor variables, and further constructing the cognitive impairment prediction system based on gastrointestinal electrical signals; A method characterized by including

7. The method according to 6, characterized in that the sample data is divided into a training set and a validation set in a ratio of 7:

3.

8. The method according to 6, characterized in that the pre-training includes a first variable selection and a second variable selection; the first variable selection includes LASSO regression analysis; and the second variable selection includes logistic regression analysis and stepwise multiple regression analysis.

9. The method according to 7, further comprising the step of evaluating the accuracy of the cognitive impairment prediction model using the verification set, wherein the evaluation indicators of the evaluation include one or more of calibration, discrimination and clinical utility.

10. The method according to 6, characterized in that the pre-training includes a ridge regression or random forest model.

Images