Diagnosis Model Of Pancreatic Cancer Based On Machine Learning Algorithm

With the rapid development of computer network technology,since the 1990 s,in the medical,health and hygiene field,A large number of clinical information about the patient's history,diagnosis,examination and treatment is stored.From the vast amount of medical data,the data mining technique can extract the trend of diseases and discover the valuable diagnostic rules.Then,using machine learning method to build prediction model.,training and testing the model,get the appropriate algorithm model,The algorithm model can diagnose,predict and monitor quickly and accurately,and help doctors determine the best diagnosis and treatment plan..For the diagnosis of pancreatic cancer,we mainly go through five stages: stage 1: examination of history,risk factors and physical;Stage 2: examination of serum tumor markers;Stage 3: it is decided whether to conduct imaging examination on the basis of tumor markers examination results;Stage 4: pathological examination.Stage 5: clinical staging.And pathological examination is the only "gold standard" for the diagnosis of pancreatic cancer.Based on the understanding and the research status of the computer technology applied in the diagnosis of pancreatic cancer,we found that the existing methods by means of computer or machine learning to assist in the diagnosis of pancreatic cancer model,mostly in the third stage based on the results of imaging examination,and machine learning algorithm was applied to the first stage of the characteristics of clinical symptoms,risk factors and tumor markers results of the second phase of the model is few and far between.Now,most of the researches on tumor markers in the second stage are made by ELISA or electrochemiluminescence.So this paper presents a view that using machine learning algorithm to the diagnosis of pancreatic cancer in the first and the second stage,using the classification of traditional machine learning algorithms mainly,analysising the clinical medical data of patients with pancreatic cancer and the data of general surgery of patients with no-pancreatic cancer of a hospital in changchun.first,data preprocessing and feature selection,and then,training and testing data,set up the early auxiliary diagnosis model based on machine learning algorithms,and analyzed the performance of the pancreatic cancer diagnosis model,at last,using majority voting strategy to fusion the single model.The result through the above processes,compared with the test results by using ELISA method and the electrochemical luminescence method,the accuracy rate and recall rate are significantly increased.This article also use metric learning algorithms: Large-Margin Nearest Neighbors method LMNN and Neighbourhood Components Analysis method NCA,training the kernel function of the support vector machines(SVM),on the basis of the improved algorithm,we trained the pancreatic cancer diagnosis model,it improved the accuracy rate of data classification,especially the support vector machine(SVM)model improved by NCA method,the accurate rate achieved 92% and the recall rate has reached to 90%,the average performance reaches to 90.1%,compared with the diagnosis model established by traditional support vector machine(SVM),respectively increased by 10%,6% and 9.9%.The research of this paper is based on machine learning algorithm,characterized by the typical symptoms and risk factors of pancreatic cancer combination with serum tumor marker test results,established diagnosis Model of Pancreatic Cancer,achieved the purpose of early pancreatic cancer early diagnosis;improved the selfidentification awareness of the high-risk group of pancreatic tumors,assisted the doctor to prediagnosis of pancreatic cancer through clinical non-specific symptoms,and reduced the misdiagnosis rate due to similar symptoms.