Evaluation And Research Of CTG Signal Quality Based On SVM And Deep Learning

Fetal heart rate monitoring,as a routine examination item in obstetric clinics,is of great significance for fetal health monitoring during perinatal period.Ultrasound transducer is the most commonly used monitoring method.It transmits an ultrasound beam to the fetal heart,gets the Doppler shift from the reflected beam,then processes the Doppler shift and calculates the instantaneous fetal heart rate.Therefore,the quality of ultrasound Doppler signal is very vital for extracting fetal heart rate accurately.The success of deep learning in various classification tasks promotes the application of deep learning in time series classification tasks.The end-to-end deep learning model,which does not require complex preprocessing of the original data,occupies a place in the deep learning model of times series classification and brings more possibilities for the classification of ultrasonic Doppler signal quality.Existing ultrasound Doppler signal quality classifications are based on extracting features,such as sample entropy that quantify the complexity of the time series.They do not take the fetal heart rate extraction algorithm of ultrasound Doppler signal into account and require complicated data preprocessing.This paper proposes two methods to solve the above two problems respectively:(1)The work combines the fetal heart rate extraction algorithm and extracts features from the average magnitude difference function curve,which relate to the quality of ultrasound Doppler signal,then uses support vector machine(SVM)to classify the quality.Thus,the fetal heart rate and the quality of ultrasound Doppler signal are closely linked.(2)In this paper,the author proposes an end-to-end deep learning model,Inc-ALSTM-CNN,which simplifies the complexity of data preprocessing and avoids the participation of artificial feature engineering.The Inception structure enhances the depth and width of feature extraction,and enriches the hierarchy of features;meanwhile,the ALSTM captures the long-term dependencies in the sequence,and extracts more critical and important information to improve the classification effect.The experimental results show that the SVM-based feature classification model can effectively classify the ultrasonic Doppler signal quality with an accuracy rate of 97.16%,which is about 3% higher than the sample entropy-based classification model,and the performance of the features extracted from the average magnitude difference function curve is superior to sample entropy on the quality classification of ultrasonic Doppler signal.The Inc-ALSTM-CNN model proposed in this paper has a better classification effect on ultrasound Doppler signal quality classification,with an accuracy rate of 96.51%.In this paper,the ultrasound Doppler signal quality classification model is integrated into the obstetric monitoring workstation software system.During the fetal monitoring process,the signal quality of each fetal heart rate is marked,it provides auxiliary information about the quality of the original signal for obstetricians to analyze CTG diagram,and provides reference for obstetricians to interpret CTG charts.