Study On The Risk Prediction Model Of Liver Cancer Based On Serological Markers

Liver cancer which has high mortality is the most popular malignant tumor.China is a large country with liver cancer,and the number of liver cancer incidence and deaths per capita are higher than the world average.In the diagnosis of liver cancer,ultrasound and AFP are usually used for examination.When suspected liver cancer or abnormalities are found,pathological analysis is carried out by means of puncture biopsy.This method has high accuracy of inspection results,but it will cause excessive burden on patients and inspectors.With the accumulation of medical data,it has become a hot research field to search for potential liver cancer markers with the help of algorithm tools such as machine learning in a large amount of clinical data and establish a predictive model through correlation analysis to improve the efficiency of liver cancer diagnosis.Through the collection and processing of multicenter clinical data,this paper uses machine learning to analyze and model the data for the prediction and diagnosis of liver cancer.The main contents are as follows:(1)The values of AFP,AFP-L3% and DCP in blood samples from 8medical centers were measured experimentally.The background information of the samples was searched through the electronic medical record system of the hospital.Through data cleaning,clinical data from a multicenter cohort were obtained.(2)A liver cancer diagnosis model was established based on seven indicators of gender,age,alpha-fetoprotein,total bilirubin,alphafetoprotein heterogenous body ratio,platelet count and abnormal prothrombin.At a cutoff of-2.316,the accuracy rate,AUC,sensitivity and specificity of the model in the test set are 94.41%,0.954,88.04% and94.85%,which are better than existing models.The model prediction results are converted into risk prediction grades through formulas,which is convenient for clinical application.(3)Based on multi-center data,the prediction ability of six machine learning models including support vector machine,decision tree,logistic regression,naive Bayes,K-nearest neighbor,and neural network was evaluated by nested cross-validation.And based on the idea of ensemble learning,three ensemble methods of Bagging,Boosting and Stacking are used for the base model to establish a model with stronger prediction ability.Finally,four machine learning methods with F1 scores greater than 0.7 were screened and applied in the dataset,showing good performance.This paper builds a model by collecting data from multiple medical centers across China,which is more in line with the situation of Chinese patients.The research results can provide decision support for clinicians in the early screening and diagnosis of liver cancer,reduce the medical burden,increase the detection rate of early liver cancer,and improve the survival rate of patients.