Application Of Artificial Neural Network In Constructing Prognostic Model Of Childhood Acute Myeloid Leukemia

Background:Acute Myeloid Leukemia(AML)is a malignant disease originating from myeloid hematopoietic stem/progenitor cells.It is characterized by abnormal cloning of primitive and immature cells in bone marrow and peripheral blood,which inhibits normal hematopoietic activity and infiltrates other organs and tissues.Most of the cases are in critical condition with a dangerous prognosis and high mortality.The incidence of AML ranks 5th among childhood malignancies,but the mortality rate ranks 2th.Over the past few decades,the 5-year survival rate for pediatric AML has increased to 70% with continued intensive treatment and optimized stratification,but about 30% of the patients will relapse and 5-10% will die.so more methods are still needed to improve survival.At present,prognosis is mainly determined by risk stratification based on patient cytogenetics and molecular biology.For children with AML,risk stratification is mainly dependent on cytogenetics,molecular biology and minimal residual disease after induction therapy.However,other clinical characteristics of patients,such as age,ethnicity,post-induction efficacy,leukocyte,and targeted therapy,and so on,also affect the outcome of children with AML.Therefore,the traditional application of common chromosome and molecular biology in AML to determine the prognosis is still relatively simple,and how to synthesize various factors to accurately judge AML prognosis of patients is particularly important.According to this,the study is designed to provide important reference value for children's stratified treatment of AML.Objective:The influencing factors related to the prognosis of children with acute myeloid leukemia were analyzed.The Artificial Neural Network(ANN)model was used toestablish the prognosis model for predicting the 5-year survival status of children with AML,and the predictive performance of the model was compared with the traditional logistic regression model.Methods:Download the Clinical data of pediatric AML patients from the TARGET platform,and select 589 cases were screened with age,gender,bone marrow protocell ratio,FAB typing,chromosome karyotype,gene mutation,targeted therapy and other important clinical data.Kaplan-meier method,and independent sample T test were used for single factor analysis to screen the influencing factors that affect the prognosis,and establish an ANN prediction model.Then,binary logistic multivariate analysis was used to screen the independent factors influencing the prognosis and establish the logistic regression model.The ROC curve was used as an evaluation indicator to compare the performance of the two methods in evaluating the 5-year prognostic model of childhood AML patients.Result:1.589 cases of patients were retrospectively analyzed,single factor analysis results can be obtained: age,race,FAB classification,chloroma,event free survival,chromosome karyotype risk stratification,complexity of karyotype group,gene fusion,CR status at end of course 1,CR status at end of course 2,these 10 items were the influencing factors of 5-year prognosis in children AML patients(P<0.05).Gender,leukocyte,Bone marrow leukemic blast percentage,Peripheral blasts,CNSL,FLT3/ITD positive,NPM1 mutations,CEBPA mutation,WT1 mutation,MRD at end of course 1,MRD at end of course 2,gemtuzumab ozogamicin treatment,these 12 items were not significantly associated with 5-year prognosis in children with AML(P>0.05).2.Event free survival(EFS),CR status at end of course 2,and risk stratification were independent influencing factors for 5-year survival of childhood AML patients.3.The accuracy of the Logistic regression model was 81.3%,the sensitivity was82.10%,the specificity was 81.9%,and the AUC value is 0.82.The accuracy of ANNmodel is 91.8%,the sensitivity was 98.64%,the specificity was 82.6%,and the AUC value was 0.901.In general,it was determined that the Ann model was better at predicting the survival of children with AML.Conclusions:1.Event free survival,cytogenetic prognostic stratification,and CR status at end of course 2 were independent influencing factors affecting the 5-year prognosis of children with AML.2.In this study,ANN model was successfully used to establish a 5-year prediction model for the prognosis of childhood AML patients,and compared with the Logistic regression model,the prediction performance is better than the traditional logistic regression model.3.The artificial neural network model integrates can synthesize the influencing factors related to the prognosis of patients,and makes individualized prediction for the patient's prognosis,which is beneficial to accurate treatment.