A Study On Automatic Identification Of Neonatal CFS Based On Traditional Features And Deep Features

In modern clinical medicine,amplitude-integrated electroencephalography(aEEG)and electroencephalography(EEG)play an increasingly important role in the identification of neonatal cerebral functional status.In clinical practice,doctors usually combine aEEG and EEG to identify neonatal cerebral functional status.However,most studies on the automatic analysis of cerebral functional status are based on aEEG or EEG alone.Referring to the doctor's clinical application,the paper uses aEEG and EEG combined methods to identify the neonatal cerebral functional status.The paper proposes an automatic identification method of neonatal cerebral function combining aEEG with EEG and combining traditional features and deep features.The main tasks include:1.Referring to the doctor's clinical diagnosis of neonatal cerebral functional status,the paper proposes an automatic recognition model of neonatal cerebral function combined with aEEG and EEG.Traditional feature engineering and deep learning techniques were used to extract aEEG and EEG signal features,and they were merged.Then a classification model was constructed to achieve the automatic identification of neonatal cerebral functional status,the final result is better than the previous algorithm.2.A method of extracting aEEG time series signal features based on LSTM is proposed.Using the processing timing signals advantage of LSTM,an LSTM model was constructed to extract the aEEG time series signal features.The results of the actual aEEG signal analysis show that the aEEG time series signal feature extracted based on the LSTM model constructed by the paper is superior to the traditionally designed linear feature.The accuracy of using this feature to identify the cerebral function status is better than that of the linear function.The recognition accuracy is 2% higher.3.A method of extracting EEG segmented time series signal features based on CNN is proposed.Using the processing one-dimensional signals advantage of CNN,a CNN model was constructed to extract the EEG segmented time series signal features.The results of the actual EEG signal analysis show that this feature is superior to the combination of the linear features and the complexity features extracted by traditional feature engineering methods.The accuracy of using this feature to identify brain function status is 2% higher than that of using traditional combination feature to identify brain functional status,reaching 96.69%.Seizure as an abnormal condition of neonatal brain function status,the paper finally combined with the traditional features and deep features to identify neonatal cerebral functional status with the reference to seizures,reaching 97.68%.