Study On The Treatment Of Neonatal Hypoxic-Ischemic Encephalopathy Based On Temporal Feature Learning

Neonatal hypoxic ischemic encephalopathy is one of the common causes of neurological disability in the neonatal period,seriously endangering the brain development of newborns.After the patient is diagnosed,intensive care treatment will be carried out immediately,and the continuous physiological signals of the patient during this treatment period will be recorded simultaneously by the equipment.The early stage of hypoxic ischemic encephalopathy in newborns is crucial for determining the treatment plan in the future.How to effectively explore the inherent information contained in these physiological signals,achieve classification and recognition of treatment effects based on potential states,promote disease progression tracking,and predict patient future trajectories is a key focus in the field of medical signaling.This article has conducted relevant analysis and research on this,and the main content is as follows:1)Preprocessing of raw physiological indicator dataProvide a preprocessing method for continuous time series physiological signals.The original data is continuous data recorded by instruments during the patient’s low-temperature treatment period.Due to on-site collection factors affecting the original data,there is a certain amount of noise that is not conducive to subsequent processing.This article adopts a weighted sliding mean filter based on one-dimensional convolution for signal smoothing and denoising to achieve data enhancement,which is convenient for subsequent feature extraction work.2)Coherence acquisition method between time series physiological signalsA feature acquisition method for coherence between two temporal physiological signals is proposed.In the field of medical brain diseases,the condition of a patient’s illness is determined based on whether their brain tissue can function normally,and the self-regulation ability of the brain is commonly used to measure the state of the brain tissue.The average arterial pressure and brain tissue oxygen saturation,two physiological indicators,are strongly correlated with the self-regulation ability of the brain.Therefore,extracting effective features from these physiological indicators of patients is crucial for studying the treatment effect of hypoxic ischemic brain lesions in newborns.In this thesis,we propose a method to obtain the coherence of indicators,which can achieve coherence feature extraction between physiological signal data.Based on this method,we can classify and recognize the target data set of neonatal hypoxicischemic encephalopathy,and the accuracy can reach 82.41%.3)Classification and Recognition Method Based on Time Neighborhood CodingProvide a neighborhood encoding temporal signal classification and recognition method based on comparative learning.In the field of medicine,the dynamic modeling ability of time series data is valuable for identifying,tracking,and mining potential patient states.However,in reality,due to the complexity of the pathology in the medical field and the uncertainty of the patient’s treatment process,it is difficult to achieve marked data,so it is quite difficult to carry out supervised learning.Therefore,based on the concept of contrastive learning,this article designs a method of contrastive learning through the temporal representation of adjacent regions to achieve the classification,recognition,and prediction of temporal medical physiological signals.The accuracy of this method on ECG,HAR,and HIE temporal physiological signal datasets is 77.79%,88.32%,and 97.52%.