| A very important task in analyzing data in the real-world survival category is the study of the relationship and magnitude of the relationship between survival time and outcome and the numerous factors and covariates that affect patients,which is known as survival analysis.Traditional survival analysis(Survival analysis)is the statistical extrapolation of one or more non-negative random variables(the more common case is the analysis of the life span of a technological product or biological or human life span),also known as survival analysis or survival analysis.In early traditional survival analysis problems,the more frequently used methods were the Kaplan-Meier algorithm and the Cox proportional hazards model,but because these traditional survival analysis methods suffer from problems such as not incorporating patient covariates or poor results in practical problems.In recent years,more and more scholars have used deep learning ideas to solve the survival analysis problem,and deep learning has become a powerful tool to study the survival analysis problem.In this paper,we propose two survival analysis methods based on deep learning ideas for processing complex survival data,and the methods mainly use different research ideas and regularization methods to build different survival models.One is to use Resnet residual network for feature extraction of patient covariates to make the model avoid possible problems such as gradient explosion,and to propose a new contrast loss function based on the concept of contrast learning using appropriate changes and fusion changes with ranking losses.The other is the use of RNN network structure in a shared subnetwork,using the property that RNN networks can learn historical information about patients,which merges historical information about each patient,i.e.,longitudinal time series data and updates the risk prediction for a specific reason,and adds a priori loss to regularize the model,which can handle competing risks,i.e.,settings where there is more than one possible event of interest.In this paper,two different research ideas of Dscm model and Drscm model are designed and used to deal with complex survival data,and the results of each of the two models are compared with other models used to deal with complex survival analysis problems by performance metrics.Following the proposed approach to the survival analysis problem,a neural network is trained to directly learn the joint distribution probabilities of time and the estimated events of interest,and the data are processed and analyzed,and the results show that the proposed approach has better performance than several other approaches in an environment with competing risks. |
