Survival Risk Prediction of Esophageal Cancer Based on Self-Organizing Maps Clustering and Support Vector Machine Ensembles

Junwei Sun,Yanfeng Wang,Yuli Yang,Xin Song,Xueke Zhao,Lidong Wang

doi:10.1109/access.2020.3007785

Abstract

This article provides a method based on self-organizing maps (SOM) neural network clustering and support vector machine (SVM) ensembles to predict the survival risk levels of esophageal cancer. Nine blood indexes related to patient survival are found by using SOM clustering method. Two critical thresholds for survival are found by plotting the receiver operating characteristic (ROC) curve twice, and the lifetime is divided into three risk levels. Using the SVM method, patients' risk levels are predicted and assessed. Four kernel functions of SVM are compared, and the prediction effect of RBF kernel function is better than other kernel functions. The parameters of SVM are optimized by using genetic algorithm (GA), particle swarm algorithm (PSO) and artificial bee colony (ABC) algorithm. Experimental results show that the prediction accuracies are improved by using optimization algorithms. After comparison, ABC-SVM has better prediction results than GA-SVM and PSO-SVM with a high prediction rate and fast running time.

Highlights

Enormous social and economic burdens have been caused by all kinds of tumors, which have been one of the leading causes of death in the whole world
self-organizing maps (SOM) neural network and support vector machine (SVM) ensembles are used to find blood indexes that are significantly related to patient survival, and to predict patients’ risk levels
Main contributions of this paper are summarized as follows: (i) Based on the SOM clustering method, nine blood index combinations that are significantly related to the survival of esophageal cancer patients are found

Summary

INTRODUCTION

Enormous social and economic burdens have been caused by all kinds of tumors, which have been one of the leading causes of death in the whole world. J. Sun et al.: Survival Risk Prediction of Esophageal Cancer Based on SOM Clustering and SVM Ensembles. A new method based on blood index information of patients to predict the risk levels of esophageal cancer is proposed. SOM neural network and SVM ensembles are used to find blood indexes that are significantly related to patient survival, and to predict patients’ risk levels. The purpose of this article is to study the survival risk prediction of esophageal cancer patients based on information of blood indicators. By using SOM clustering, ROC, SVM, GA, PSO, and ABC algorithm, a new method for predicting the survival risk of esophageal cancer is provided. (i) Based on the SOM clustering method, nine blood index combinations that are significantly related to the survival of esophageal cancer patients are found. Sometimes the initial weight vector of a neuron is too far from the input vector, which will cause it not to win the competition, and never learn and become useless neurons

COX REGRESSION ANALYSIS TO VERIFY INDEX CORRELATION

SVM BASED ON PARAMETER OPTIMIZATION

Findings

CONCLUSION