A Deep Survival Analysis Based On Ranking For Prognosis Nasopharyngeal Carcinoma

Nasopharyngeal carcinoma(NPC)is a cancer disease which is highly prevalent in southern China and Southeast Asia.The highest incidence rate is 50 cases of every 100 000 people.Radiotherapy(RT)is the primary treatment,and concurrent chemoradiotherapy with or without adjuvant chemotherapy is the standard of care for advanced NPC.However,20% to 30% of patients will develop local or systemic recurrences,most of which occur within the first two years after treatment.Because 10% to 20% of patients with local or systemic recurrence may be cured with additional treatment,it is necessary to identify patients with a high risk of recurrence earlier.Therefore,there is an urgent need for an accurate prognostic tool to help clinicians diagnose high-risk patients.At present,the TNM staging system based on anatomic information is used to establish a baseline for the treatment of nasopharyngeal cancer patients,but the clinical practice has found that the staging system is not sufficient to predict recurrence.On the other hand,other risk factors in clinical practice can also affect the recurrence of nasopharyngeal carcinoma,but few tools integrate these risk factors.In order to solve the above two types of problems,a survival analysis mo del that can integrate multiple risk factors and accurately relapse of nasopharyngeal carcinoma is very much anticipated.The traditional survival analysis model can use clinical risk factors to analyze the relationship between disease and risk factors and will also use these factors to predict the survival and recurrence of patients.Cox proportional hazards model is a commonly used survival analysis method in medical field.It can integrate TNM staging and other risk factors to predict recurrence of nasopharyngeal carcinoma.However,the prediction of nasopharyngeal carcinoma by Cox proportional hazard model requires clinical experts to conduct Feature Engineering for clinical data.In addition,the Cox proportional hazards model assumes that the risk function of any two patients is proportional,and the hazard function is a linear combination of variables,and these situations are rarely present in reality.Subsequently,the researchers used the output of the neural network as a risk function of the Cox proportional hazards model to avoid the above problems.However,the early development of neural network technology is not mature,and the performance of Cox model based on neural network expansion is not as good as that of traditional Cox proportional hazards model.Recently,benefiting from the current development of deep learning technologies,researchers have used new neural network techniques to improve the Cox proportional hazards model and achieved better predictive results.In order to further improve the accuracy of predicting recurrence of nasopharyngeal carcinoma,we use a more mature neural network to study a deep survival analysis model based on ranking and achieved the best results.This model compares three traditional survival analysis models and shows better predictive results on all four publicly available data.In the prognosis of nasopharyngeal carcinoma,the proposed method uses nine clinical variables to achieve the best predictor of 0.766,which is 0.036 higher than the 0.730 obtained by the clinical expert using the Cox proportional hazards model.Such performance improvement can help clinicians to have a good prognosis for patients,which has important clinical significance.In this paper,we apply a deep survival analysis model based on ranking to analyze the prognosis of nasopharyngeal carcinoma.The main research contents are as follows:1)In this paper,we introduce the related knowledge and theory of survival analysis.2)This paper proposes a deep survival analysis model based on ranking and verifies the performance of the new model in the open data set and analyzes the model.3)This paper applies the deep survival analysis based on ranking to the prognosis of nasopharyngeal carcinoma,and research the effect of single clinical factor EBV DNA on recurrence of nasopharyngeal carcinoma.