Fatty liver risk assessment and prediction model based on artificial intelligence

Abstract

A fatty liver risk assessment and prediction model based on artificial intelligence comprises the following steps: data collection: selecting hospital physical examination crowds of all large hospitals in all regions of the country from January 2020 to January 2023, screening queue research volunteers according to a selection standard, performing reexamination once every year, and performing queue follow-up visit for three years in total, measuring indexes: obtaining research data by performing questionnaire survey, physique measurement and biochemical index detection on a research object, screening out six NPC patient death risk factors such as age, race, gender, StageM, StageT and StageN by combining AI algorithms such as LASSO, XGBoost, DT and RF on the basis of an SEER database and clinical data of local NPC patients, successfully constructing an NPC patient death risk prediction model, and determining the NPC patient death risk. And the model is subjected to internal evaluation, internal verification and external verification, so that a relatively good risk prediction capability is shown, and the method has important clinical significance for evaluating the prognosis of the NPC patient, and relatively high popularization and application values are shown.

Description

Technical Field

[0001] The present invention relates to the field of fatty liver prediction, and specifically to an artificial intelligence-based fatty liver risk assessment and prediction model. Background Art

[0002] Fatty liver, also known as fatty liver disease, is a disease caused by various reasons, such as genetic susceptibility, environmental factors and metabolic stress, with the basic pathological feature of liver cell fat change. It can be divided into alcoholic fatty liver disease, non-alcoholic fatty liver disease and special types of fatty liver, among which non-alcoholic liver disease is the most common. The patients are mainly obese, metabolic syndrome, type 2 diabetes and long-term excessive drinkers.

[0003] The symptoms of fatty liver depend mainly on the severity of the disease. Most chronic fatty liver diseases, such as alcoholic or non-alcoholic liver disease, are asymptomatic at first, but as the disease progresses, symptoms of liver enlargement or even cirrhosis may appear. Acute fatty liver disease presents symptoms similar to acute severe hepatitis and liver failure.

[0004] With the improvement of people's living standards and the emergence of various diseases, people are paying more and more attention to their own health. Now there is no good method to complete the risk assessment and prediction of fatty liver, which leads to great uncertainty and harm in the occurrence of fatty liver, which is not conducive to people's health. Summary of the invention

[0005] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a fatty liver risk assessment and prediction model based on artificial intelligence, which effectively solves the problem that there is no good method to complete the risk assessment and prediction of fatty liver, which leads to great uncertainty and harm in the occurrence of fatty liver, which is not conducive to people's health.

[0006] To achieve the above object, the present invention provides the following technical solution: The present invention comprises the following steps:

[0007] Step 1: Data collection: Select people who have undergone physical examinations in major hospitals across the country from January 2020 to January 2023, screen volunteers for cohort studies according to the inclusion criteria, conduct reexaminations once a year, and conduct a total of 3 years of cohort follow-up;

[0008] Step 2, measurement indicators: The data of this study were obtained by conducting questionnaire surveys, physical measurements, and biochemical index tests on the subjects. Physical measurement indicators included height, body mass index, diastolic blood pressure, and systolic blood pressure. Laboratory indicators included triglycerides, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, alanine aminotransferase, urea nitrogen, and creatinine.

[0009] Step 3: Data sorting: Delete samples with serious missing data in the SEER database (missing data>10%), and fill in samples with fewer missing values. Continuous variables are filled with the mean; categorical variables are filled with the mode. Finally, the clinical data of 3000 NPC patients were obtained;

[0010] Step 4: Random grouping: Based on the clinical data of 3,000 NPC patients in the SEER database (2020-2023), the data were processed using SPSS25.0 software, and randomly divided into a modeling group (2,100 cases) and an internal validation group (900 cases) at a ratio of 7:3. The two groups of data were tested for balance, and six machine learning models including Catboost, XGBoost, KNN, logistic, SVM, and NaiveBayes were used;

[0011] Step 5. Evaluation model construction process Fatty liver prognosis evaluation model Evaluation model construction process Data collection and preprocessing, collect clinical data and laboratory test results, including patient age, gender, body mass index, and liver function indicators; clean and preprocess the collected data to remove outliers and missing values; divide the data into a modeling group and an internal validation group to ensure the reliability and generalization ability of the evaluation model; feature selection and engineering; select features that are meaningful for fatty liver prognosis evaluation through correlation analysis and feature importance evaluation; apply feature engineering techniques, such as continuous variable discretization and categorical variable encoding, to improve the adaptability of the model; evaluation model construction process Model selection and parameter adjustment, compare the performance of different machine learning algorithms, select a model suitable for fatty liver prognosis evaluation, and optimize the model parameters through cross-validation and grid search to improve the prediction accuracy of the model;

[0012] Step 6: The model was internally evaluated using C-index, decision curve analysis (DCA), calibration curve (CC), and area under the receiver operating characteristic (ROC) curve (AUC). At the same time, the internal validation group data was used for internal validation. In addition, the clinical data of NPC patients collected in the region were used for external validation.

[0013] Step seven: SPSS25.0 was used to organize the data, and a univariate χ2 test was performed to compare the differences between the groups; R4.0.4 was used to draw the risk prediction model, C-index, DCA, CC and area under the ROC curve AUC; P < 0.05 was considered statistically significant.

[0014] According to the above technical solution: the selection criteria of step 1 are:

[0015] (1) Complete three physical examinations in three years;

[0016] (2) No NSFLD and negative hepatitis virus markers at the time of inclusion;

[0017] (3) Men’s alcohol intake is less than 140g per week, and women’s alcohol intake is less than 70g per week;

[0018] (4) Patients who had not suffered from viral hepatitis, drug-induced liver disease, or other diseases that cause fatty liver during the 3-year period, and who had no other types of liver diseases such as primary cirrhosis and autoimmune liver disease. According to the above inclusion criteria, those who were lost to follow-up during the follow-up period and those with incomplete or missing data were excluded.

[0019] According to the above technical scheme: FLI is used for preliminary prediction in the measurement index of step 2, and the prediction accuracy of FLI for hepatic steatosis is good, with an AUC of 0.840; the prediction factors include waist circumference, body mass index (BMI), triglyceride (TG) and glutamyl transpeptidase (GGT).

[0020] According to the above technical scheme: the clinical significance of FLI is: FLI value range is 0~100, FLI<30 excludes hepatic steatosis, with a sensitivity of 87.0%; FLI≥60 is determined as hepatic steatosis, with a specificity of 86.0%. The FLI calculation formula is as follows: FLI=e〔0.953×ln(TG)+0.139×BMI+0.718×ln(GGT)+0.053×WC-15.745〕 / 〔1+e0.953×ln(TG)+0.139×BMI+0.718×ln(GGT)+0.053×WC-15.745〕×100.

[0021] According to the above technical solution: the construction of the risk prediction model in step seven is based on NPC modeling data, and the screened clinical characteristics are incorporated into the risk nomogram prediction model to construct a risk nomogram prognostic model for NPC patients. The results show that the risk of death in male NPC patients is higher than that in female patients; the risk of death in elderly NPC patients is higher than that in younger patients; the risk of death in blacks is higher than that in other races; and as the tumor stage (TNM) increases, the risk of death in NPC patients gradually increases.

[0022] According to the above technical scheme: the data is divided into a modeling group and an internal validation group with a ratio of 7:3, and brought into six machine learning models of Catboost, XGBoost, KNN, logistic, SVM, and NaiveBayes to construct a fatty liver prediction model. The prediction performance of the model is evaluated and compared based on the precision, accuracy, recall rate, F1 value, area under the curve (AUC), and calibration curve, and finally the model constructed by Catboost is determined to be the optimal one (AUC=0.985).

[0023] Beneficial effects: The fatty liver assessment model of the present invention can provide doctors with more accurate and predictive assessments, which is helpful to formulate personalized treatment plans. The application of the model can improve medical efficiency and reduce unnecessary waste of medical resources. With the continuous optimization of the model, its application in clinical practice will become more and more extensive. The model can be applied to health management for long-term health monitoring and prediction of patients with fatty liver. Through the monitoring and prediction of the model, more accurate health advice and intervention measures can be provided to patients. The application of the model helps to improve the health level and quality of life of patients with fatty liver. This study is based on the SEER database and the clinical data of local NPC patients, combined with AI algorithms such as LASSO, XGBoost, DT and RF to screen out 6 NPC patient death risk factors such as age, race, gender, Stage_M, Stage_T and Stage_N, and successfully constructed a NPC patient death risk prediction model, and conducted internal evaluation, internal verification and external verification on the model, showing good risk prediction ability, which is of great clinical significance for evaluating the prognosis of NPC patients, and shows a high value of promotion and application. DETAILED DESCRIPTION

[0024] The specific implementation modes of the present invention are described in further detail below.

[0025] Embodiment 1, the present invention provides a fatty liver risk assessment and prediction model based on artificial intelligence, comprising the following steps:

[0026] Step 1: Data collection: Select people who have undergone physical examinations in major hospitals across the country from January 2020 to January 2023, screen volunteers for cohort studies according to the inclusion criteria, conduct reexaminations once a year, and conduct a total of 3 years of cohort follow-up;

[0027] Step 2, measurement indicators: The data of this study were obtained by conducting questionnaire surveys, physical measurements, and biochemical index tests on the subjects. Physical measurement indicators included height, body mass index, diastolic blood pressure, and systolic blood pressure. Laboratory indicators included triglycerides, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, alanine aminotransferase, urea nitrogen, and creatinine.

[0028] Step 3: Data sorting: Delete samples with serious missing data in the SEER database (missing data>10%), and fill in samples with fewer missing values. Continuous variables are filled with the mean; categorical variables are filled with the mode. Finally, the clinical data of 3000 NPC patients were obtained;

[0029] Step 4: Random grouping: Based on the clinical data of 3,000 NPC patients in the SEER database (2020-2023), the data were processed using SPSS25.0 software, and randomly divided into a modeling group (2,100 cases) and an internal validation group (900 cases) at a ratio of 7:3. The two groups of data were tested for balance, and six machine learning models including Catboost, XGBoost, KNN, logistic, SVM, and NaiveBayes were used;

[0030] Step 5. Evaluation model construction process Fatty liver prognosis evaluation model Evaluation model construction process Data collection and preprocessing, collect clinical data and laboratory test results, including patient age, gender, body mass index, and liver function indicators; clean and preprocess the collected data to remove outliers and missing values; divide the data into a modeling group and an internal validation group to ensure the reliability and generalization ability of the evaluation model; feature selection and engineering; select features that are meaningful for fatty liver prognosis evaluation through correlation analysis and feature importance evaluation; apply feature engineering techniques, such as continuous variable discretization and categorical variable encoding, to improve the adaptability of the model; evaluation model construction process Model selection and parameter adjustment, compare the performance of different machine learning algorithms, select a model suitable for fatty liver prognosis evaluation, and optimize the model parameters through cross-validation and grid search to improve the prediction accuracy of the model;

[0031] Step 6: The model was internally evaluated using C-index, decision curve analysis (DCA), calibration curve (CC), and area under the receiver operating characteristic (ROC) curve (AUC). At the same time, the internal validation group data was used for internal validation. In addition, the clinical data of NPC patients collected in the region were used for external validation.

[0032] Step seven: SPSS25.0 was used to organize the data, and a univariate χ2 test was performed to compare the differences between the groups; R4.0.4 was used to draw the risk prediction model, C-index, DCA, CC and area under the ROC curve AUC; P < 0.05 was considered statistically significant.

[0033] The inclusion criteria for step 1 are:

[0034] (1) Complete three physical examinations in three years;

[0035] (2) No NSFLD and negative hepatitis virus markers at the time of inclusion;

[0036] (3) Men’s alcohol intake is less than 140g per week, and women’s alcohol intake is less than 70g per week;

[0037] (4) Patients who had not suffered from viral hepatitis, drug-induced liver disease, or other diseases that cause fatty liver during the 3-year period, and who had no other types of liver diseases such as primary cirrhosis and autoimmune liver disease. According to the above inclusion criteria, those who were lost to follow-up during the follow-up period and those with incomplete or missing data were excluded.

[0038] In the step 2, FLI is used for preliminary prediction. The prediction accuracy of FLI for hepatic steatosis is good, with an AUC of 0.840. The prediction factors include waist circumference, body mass index (BMI), triglyceride (TG) and glutamyl transpeptidase (GGT).

[0039] The clinical significance of FLI is as follows: FLI value range is 0~100, FLI<30 excludes hepatic steatosis, with a sensitivity of 87.0%; FLI≥60 is determined as hepatic steatosis, with a specificity of 86.0%. The FLI calculation formula is as follows: FLI=e〔0.953×ln(TG)+0.139×BMI+0.718×ln(GGT)+0.053×WC-15.745〕 / 〔1+e0.953×ln(TG)+0.139×BMI+0.718×ln(GGT)+0.053×WC-15.745〕×100.

[0040] The construction of the risk prediction model in step seven is based on NPC modeling data, and the screened clinical characteristics are incorporated into the risk nomogram prediction model to construct a risk nomogram prognostic model for NPC patients. The results show that the risk of death in male NPC patients is higher than that in female patients; the risk of death in elderly NPC patients is higher than that in younger patients; the risk of death in blacks is higher than that in other races; and as the tumor stage (TNM) increases, the risk of death in NPC patients gradually increases.

[0041] The data were divided into a modeling group and an internal validation group in a ratio of 7:3, and were introduced into six machine learning models, namely Catboost, XGBoost, KNN, logistic, SVM, and NaiveBayes, to construct a fatty liver prediction model. The prediction performance of the models was evaluated and compared based on the precision, accuracy, recall, F1 value, area under the curve (AUC), and calibration curve, and finally the model constructed by Catboost was determined to be the optimal one (AUC=0.985).

[0042] Beneficial effects: The fatty liver assessment model of the present invention can provide doctors with more accurate and predictive assessments, which is helpful to formulate personalized treatment plans. The application of the model can improve medical efficiency and reduce unnecessary waste of medical resources. With the continuous optimization of the model, its application in clinical practice will become more and more extensive. The model can be applied to health management for long-term health monitoring and prediction of patients with fatty liver. Through the monitoring and prediction of the model, more accurate health advice and intervention measures can be provided to patients. The application of the model helps to improve the health level and quality of life of patients with fatty liver. This study is based on the SEER database and the clinical data of local NPC patients, combined with AI algorithms such as LASSO, XGBoost, DT and RF to screen out 6 NPC patient death risk factors such as age, race, gender, Stage_M, Stage_T and Stage_N, and successfully constructed a NPC patient death risk prediction model, and conducted internal evaluation, internal verification and external verification on the model, showing good risk prediction ability, which is of great clinical significance for evaluating the prognosis of NPC patients, and shows a high value of promotion and application.

[0043] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. 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. An artificial intelligence-based fatty liver risk assessment and prediction model, characterized by: The following steps are included: Step 1: Data collection: Select people who have undergone physical examinations in major hospitals across the country from January 2020 to January 2023, screen volunteers for cohort studies according to the inclusion criteria, conduct reexaminations once a year, and conduct a total of 3 years of cohort follow-up; Step 2, measurement indicators: The data of this study were obtained by conducting questionnaire surveys, physical measurements, and biochemical index tests on the subjects. Physical measurement indicators included height, body mass index, diastolic blood pressure, and systolic blood pressure. Laboratory indicators included triglycerides, total cholesterol, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, alanine aminotransferase, urea nitrogen, and creatinine. Step 3, data sorting: Delete samples with serious missing data in the SEER database (missing data>10%), and fill in samples with fewer missing values. Continuous variables are filled with mean values; Categorical variables were imputed using mode, and finally clinical data of 3000 NPC patients were obtained; Step 4: Random grouping: Based on the clinical data of 3,000 NPC patients in the SEER database (2020-2023), the data were processed using SPSS25.0 software, and randomly divided into a modeling group (2,100 cases) and an internal validation group (900 cases) at a ratio of 7:

3. The two groups of data were tested for balance, and six machine learning models including Catboost, XGBoost, KNN, logistic, SVM, and NaiveBayes were used; Step 5, the construction process of the evaluation model Fatty liver prognosis evaluation model The construction process of the evaluation model Data collection and preprocessing, collect clinical data and laboratory test results, including patient age, gender, body mass index, and liver function indicators; clean and preprocess the collected data to remove outliers and missing values; divide the data into a modeling group and an internal validation group to ensure the reliability and generalization ability of the evaluation model; feature selection and engineering; select features that are meaningful for fatty liver prognosis evaluation through correlation analysis and feature importance evaluation; Apply feature engineering techniques, such as discretization of continuous variables and encoding of categorical variables, to improve the adaptability of the model; evaluate the model selection and parameter adjustment of the model construction process, compare the performance of different machine learning algorithms, select a model suitable for the prognosis assessment of fatty liver disease, and optimize the model parameters through cross-validation and grid search to improve the prediction accuracy of the model; Step 6: The model was internally evaluated using C-index, decision curve analysis (DCA), calibration curve (CC), and area under the receiver operating characteristic (ROC) curve (AUC). At the same time, the internal validation group data was used for internal validation. In addition, the clinical data of NPC patients collected in the region were used for external validation. Step seven: SPSS25.0 was used to organize the data, and a univariate χ2 test was performed to compare the differences between the groups; R4.0.4 was used to draw the risk prediction model, C-index, DCA, CC and area under the ROC curve AUC; P < 0.05 was considered statistically significant.

2. The artificial intelligence-based fatty liver risk assessment and prediction model according to claim 1, characterized in that: The inclusion criteria for step 1 are: (1) Complete three physical examinations in three years; (2) No NSFLD and negative hepatitis virus markers at the time of inclusion; (3) Men’s alcohol intake is less than 140g per week, and women’s alcohol intake is less than 70g per week; (4) Patients who had not suffered from viral hepatitis, drug-induced liver disease, or other diseases that cause fatty liver during the 3-year period, and who had no other types of liver diseases such as primary cirrhosis and autoimmune liver disease. According to the above inclusion criteria, those who were lost to follow-up during the follow-up period and those with incomplete or missing data were excluded.

3. The artificial intelligence-based fatty liver risk assessment and prediction model according to claim 1, characterized in that: In the step 2, FLI is used for preliminary prediction. The prediction accuracy of FLI for hepatic steatosis is good, with an AUC of 0.

840. The prediction factors include waist circumference, body mass index (BMI), triglyceride (TG) and glutamyl transpeptidase (GGT).

4. The artificial intelligence-based fatty liver risk assessment and prediction model according to claim 3, characterized in that: The clinical significance of FLI is as follows: FLI value range is 0~100, FLI<30 excludes hepatic steatosis, with a sensitivity of 87.0%; FLI≥60 is determined as hepatic steatosis, with a specificity of 86.0%. The FLI calculation formula is as follows: FLI=e〔0.953×ln(TG)+0.139×BMI+0.718×ln(GGT)+0.053×WC-15.745〕 / 〔1+e0.953×ln(TG)+0.139×BMI+0.718×ln(GGT)+0.053×WC-15.745〕×100.

5. The artificial intelligence-based fatty liver risk assessment and prediction model according to claim 1, characterized in that: The risk prediction model in step seven is constructed based on NPC modeling data, and the screened clinical characteristics are incorporated into the risk nomogram prediction model, so as to construct a risk nomogram prognostic model for NPC patients. The results show that the risk of death in male NPC patients is higher than that in female patients; the risk of death in elderly NPC patients is higher than that in younger patients; The risk of death in blacks is higher than that in other races; and as the tumor stage (TNM) increases, the risk of death in NPC patients gradually increases.

6. The artificial intelligence-based fatty liver risk assessment and prediction model according to claim 1, characterized in that: The data were divided into a modeling group and an internal validation group in a ratio of 7:3, and were introduced into six machine learning models, namely Catboost, XGBoost, KNN, logistic, SVM, and NaiveBayes, to construct a fatty liver prediction model. The prediction performance of the models was evaluated and compared based on the precision, accuracy, recall, F1 value, area under the curve (AUC), and calibration curve, and finally the model constructed by Catboost was determined to be the optimal one (AUC=0.985).