Study On The Diagnosis And Outcome Prediction Model Of Intrahepatic Cholestasis Of Pregnancy Based On BP-NN

Medical health has always been a topic of concern,especially the health of mothers and infants.It is greatly related to the happiness of a family.Intrahepatic cholestasis of pregnancy(ICP)is a serious hazard during pregnancy.Maternal and child health complications,the highest incidence rate can reach 12%.At present,its pathogenesis has not been fully known and the diagnostic indicators are single,which poses a great challenge to the timely diagnosis of ICP.With machine learning technology in the medical field,With rapid development,researchers can use machine learning to mine information from data and design diagnostic models.This study focused on intrahepatic cholestasis of pregnancy as the main research object,and explored the diagnosis and outcome prediction model of ICP.Firstly,the newly proposed SC-ReliefF algorithm is used for feature selection;then the improved BP algorithm is used to improve the BP algorithm;finally,a diagnosis and outcome prediction model of intrahepatic cholestasis of pregnancy based on BP-NN is designed.The specific work is as follows:(1)Aiming at the high redundancy and unbalance of the original ICP data,a new feature selection algorithm SC-ReliefF is proposed and applied to the feature selection of ICP data.SC-ReliefF algorithm is based on ReliefF.On the one hand,a new sample selection method is proposed based on the average distance within the class,which can well adapt to the characteristics of ICP data imbalance.The algorithm introduces cosine similarity as a measure of feature redundancy,and proposes a de-redundancy method that can obtain a smaller feature subset.The clinical data experiments provided in Wuxi Maternal and Child Health Hospital show that the SC-ReliefF algorithm has smaller ICP feature subsets than ReliefF,mRMR,and RS-ReliefF,and has better classification on SVM and BP-NN.Effect,can improve the efficiency of learning.(2)Aiming at the shortcoming that BP neural network is easy to fall into local minimum when modeling,a new neural network training algorithm SGWO-BP is proposed.This algorithm introduces the Wolf Pack Algorithm(GWO)into the initial value assignment of the BP neural network,and uses the global optimization capabilities of the GWO algorithm to set the initial weight and threshold of the BP neural network as close to the optimal value as possible.At the same time,adaptive improvements were made to the wolf pack algorithm during the introduction process.On the one hand,the position update formula for optimizing wolves was optimized,and a walking factor was added to reduce the predation of wolves locating in the local optimization process.On the other hand,for the problem of artificial wolf attack and the constant step length during the siege process,which caused the algorithm to converge slowly in the later period,the position update formula for artificial wolf attack and siege was adjusted,so that the attack and siege step length can be changed with The dynamic adjustment of the number of iterations of the algorithm makes the optimization result more accurate.Finally,experimentally compare the performance of SGWO-BP algorithm with GA-BP,PSO-BP,and SVM on five common UCI datasets,which shows that SGWO-BP can obtain higher accuracy when the convergence speed is slightly improved.degree.(3)Aiming at the problem of different diagnosis costs for diagnosing a patient as a healthy person and a healthy person as a patient in the ICP diagnosis,the loss function of the neural network is improved,and a new cross-entropy loss function is proposed.The new loss function introduces the misjudgment factor into the cross-entropy loss function,which makes the training process of BP-NN cost-sensitive and reduces the probability of patients being misdiagnosed as normal people.