Research On Small-for-gestational-age Infants Prediction Based On Deep Learning

Small for Gestational Age(SGA)is a common disease in newborns.The birth weight of the infant with the disease is less than two standard deviations of the average weight of the normal newborn.In China,especially in underdeveloped areas,the disease seriously affects the fetus’ s physical and mental health.Therefore,researchers have been continuously exploring the causes of gestational age in children to try to intervene in the development of the disease as early as possible,so that newborns get more effective treatment.At present,the prediction of the disease requires more tedious medical tests and requires high medical equipment,which is not easy for developing countries and regions.On the problem of SGA predictions,researchers from the original single check-up index prediction and then to traditional machine learning which synthesize variety of prediction results to predict disease,neither can well solve the problem that the time of predict SGA is long and the accuracy is not high.In terms of predicting Small for Gestational Age With the development of science and technology,deep learning has become one of the most effective ways in the field of disease prediction.Deep learning can effectively abstract the underlying features into high-level features,avoiding the inaccuracies and time-consuming defects of artificial methods.Therefore,this paper proposes a method which is based on deep learning to predict the disease of children under gestational ageFirst,this paper proposes an improved Denoising Auto-Encoder.By using this method,the robustness of the model extraction features and the ability to restore features can be improved.The improved Denoising Auto-Encoder takes parental detection data as input.Then this paper predict the disease based on text features.In this paper,the short text feature of the data is segmented and word embedding is used as the input of the neural network and merged with the previous physical examination data to train the neural network model.The experimental results show that the recurrent neuron network can extract short text features better,making the classifier more efficient and accurate.Finally,in order to better improve the effect of disease prediction,some of the trained models obtained in this paper are integrated by the method of stacking and bagging.The final prediction results have greatly improved.The results of this experiment prove the effectiveness of the proposed method.The data set of the paper is based on data from the national pre-pregnancy testing project from April 2010 to December 2013.In this paper,AUC(Area Under the Receiver Operating Characteristic Curve)was used to evaluate the results,and the AUC reached 87.3%.It is shown that deep learning can be applied to predict disease of infants which are under gestational age.