Research On The Key Technologies Of Hierarchical Deep Learning Based Symptoms Monitoring During Severe Pneumonia In Children

Children severe pneumonia is a conceptual diagnostic term,in a narrow sense,refer to whose clinical symptoms observed in pediatric intensive care unit and neonatal intensive care unit,complicated with respiratory failure,pneumonia shock,heart failure and other complications.Precision tracking of patients’ body reactivity and organ functional status is a direct factor in clinical treatment during severe pneumonia.With the advent of the digital age,it is possible to accurately capture clinical symptoms or changes through artificial intelligence technology.This thesis takes the children severe pneumonia as a breakthrough point to discuss two basic problems in medical data processing:data completion and multi-sequence processing,based on the priori hierarchical data structure and multi-source time series in clinical medicine.Firstly,a novel hierarchical deep matrix completion model is proposed for data completion task based on the data feature with tree structure.Then,a novel hierarchical multi-sequence time-regular network is proposed to detect abnormal symptoms in children pneumonia intensive care under heterogeneous multi sources data.Finally,an intelligent monitoring and management prototype system for children’s severe pneumonia is designed and implemented.(1)Data completion is an essential part of data preprocessing.The Low rank matrix completion based on the low rank structure of data for recovering missing items,is a prevalent technique in data completion.We proposed a novel matrix completion model called hierarchical deep matrix completion(HDMC),where we assume that the variables lie in hierarchically organized groups.HDMC explicitly expresses either shallow or high-level hierarchical structures,by embedding a series of so-called structured sparsity penalties in a framework to encourage hierarchical relations between compact representations and reconstructed data.Moreover,HDMC considers the group-level sparsity of neurons in a neural network to obtain a pruning effect and compact architecture by enhancing the relevance of within group neurons while neglecting the between group neurons.Since the optimization of HDMC is a nonconvex problem,to avoid converting the framework of the HDMC models into separate optimized formulations,we unify a generic optimization by applying a smoothing proximal gradient strategy in dual space.HDMC is compared with state-of-the-art matrix completion methods on applications with simulated data,MRI image datasets,and gene expression datasets.The experimental results verify that HDMC achieves higher matrix completion accuracy.(2)For the diagnosis and treatment of severe pneumonia in clinical critical stage,patients rely on ICU respiratory circulation.Patients’ body reactivity is reflected by a wide range of multi-source/channel data,in addition,their clinical symptoms are differ from the individual baseline of children’s respiratory and immune system development stages.To solve the above problems,a novel hierarchical multi-sequence time-regular network(HMTNet)is proposed for analyzing the multi-channel data and demographic jointly.The architecture of HMTNet is constructed based on multi-head attention mechanism with a double-layer LSTM,which we called hierarchical time-regular LSTM.The hierarchical time-regular LSTM extract time unified of context,direct dependence and hierarchical dependence representation jointly from multisource/channel data,which constructing time-regular embedding layer in the first layer by using heuristic time decay function,and hierarchical embedding layer in the second with tree structural connecting.Finally,the multi-head attention mechanism is employing in the network,to capture the interdependence between dynamic and static information,and attend on the channel related to the underlying task.Experiment results show that,compared with the state-of-the-art network,our HMTNet improve the detection accuracy in clinical symptoms significantly,which can satisfy critical stage monitoring of children with severe pneumonia.(3)Based on the above research,this paper designs and implements an intelligent monitoring and management prototype system for children with severe pneumonia.The algorithm proposed in this paper is applied to design and implement the detection process of clinical abnormal symptoms.Firstly,HDMC is used to complete the clinical data,and filtering algorithm is used to preprocess the data.Then,HMTNet was used to detect the abnormal symptoms.Finally,statistical analysis and abnormal cause analysis of the symptoms were carried out.