Key Biomarker Mining Based On Survival Prediction Of Patients With Colorectal Cancer

Colorectal cancer is a common malignant tumor with high morbidity and mortality,and the incidence is on the rise.Therefore,the research of survival prediction and prognostic factors in patients with colorectal cancer are also becoming more and more important.Biomarkers can accurately and sensitively evaluate early damage.Effective biomarkers play an important role in early diagnosis and survival prediction of patients with colorectal cancer.Although traditional univariate and multivariate analysis methods validate the importance of biomarkers,however,they do not take the interactions between biomarkers into account.In the background,this paper studies key biomarker mining algorithms based on survival prediction of colorectal cancer patients.This paper first introduces the research status and significance of survival prediction in colorectal cancer patients.Based on the current research,a key biomarker mining model based on patient survival prediction was designed and applied on a specific dataset.Finally,a prototype system was designed.Main research contents include:(1)Design of key biomarker mining model for colorectal cancer: the key biomarker mining model framework based on feature interaction is proposed through the analysis of hybrid feature selection technology and random search feature selection method.(2)Key biomarker mining based on patient discrete survival Prediction: according to the characteristics of the dataset of clinical data characteristics,treatment characteristics and biomarker data collection,which is from the patients of the oncology department of Link?ping University in Sweden,this paper comparatively analyzes the classifier model.Aims at the discrete survival prediction of colorectal cancer patients,a marker mining model(WPSO-NB)which combined particle swarm optimization algorithm with naive Bayesian was proposed.Firstly,the significance of biomarker characteristics was verified.Then,the 5-year and 3-5-year survival predictions of intestinal cancer patients were taken as the target,and the model was verified by experiments.The classification accuracy respectively exceeded 90% and 75%,which proved the feasibility and effectiveness of the model.(3)Key biomarker mining based on patient continuous survival prediction: the combination particle swarm optimization algorithm and BP neural network constructs the mining model(WPSO-BP),which takes the prediction of continuous survival(months)of patients with colorectal cancer as the goal and was used in the mining experiment.The experimental results show that the prediction error of the optimal feature subsets mining by the model is much smaller than the prediction error of all features,which verifies the feasibility of the WPSO-BP model.(4)Key biomarker mining prototype system based on cancer patient survival prediction: a key marker mining system based on patient survival prediction was built with the help of hybrid mining model designed by the thesis,which realized the data fitting,survival prediction and key markers mining.