Automated Detection Of Shockable Rhythm Using Adaptive Window

Sudden Cardiac Death(SCD)is a very serious heart disease.In recent years,as the pace of life had accelerated(such as overtime,staying up late,uneven nutrition,etc),the number of deaths caused by SCD had sharply increased.According to statistics,there are approximately three million people died from SCD each year in the world and about 544,000 people died from SCD each year in China,top of the world.After several years of research,it was showed thar vast majority of cases of SCD was due to ventricular fibrillation(VF),ventricular tachycardia(VT),or ventricular flutter(VFL).Defibrillation is considered to be the most effective treatment way,so VF/VT/VFL are collectively called shockable rhythm(SH).The implementation time of defibrillation is very important.If using the electric defibrillation within "Golden 4 minutes",the probability of saving patients’lives is very high.Therefore,accurate detection of shockable rhythm has clinical application value.At present,there are many detection methods for detection shockable rhythm.According to the difference of feature selection ways,they can be divided into classical method based on hand-crafted features and burgeoning method based on convolutional neural network.The classical method analysis electrocardiosignal characteristic by people themselves and then obtain specific feature of the signal within fixed time window by the time series decomposition tools and finally detect shockable rhythm by traditional classifiers such as support vector machines.With the develop of deep learning,convolutional neural network had gradually been applied to detect arrhythmias.The convolutional layer in the network structure can automatically learn the input data to obtain high-dimensional features of the signal within fixed time window and the pooling layer can filter feature to avoid redundant information,so convolutional neural network can simplify the step of detection shockable rhythm.Aimed at addressing the deficiencies and drawbacks of existing shockable rhythm detection algorithms,we proposed a novel method based on adaptive window instead of fixed time window.The important step for obtaining adaptive window is locating R-wave.In this paper,we proposed the peaks and troughs algorithm to locate R-wave.The peaks and troughs algorithm first find electrocardiosignal’s troughs and screen troughs by adaptive amplitude threshold and search-back procedure.And then it finds peak between two successive troughs,similarly screen peaks by adaptive amplitude threshold and search-back procedure,those remaining peaks were regarded as R-waves.The energy,time length,sample entropy of original electrocardiosignal within adaptive window were used as timedomain features.The discrete stationary wavelet transform was used to decompose the electrocardiosignal into number of sub-signals,so the energy,sample entropy of wavelet coefficients D2 to D5 and the correlation coefficients between wavelet coefficients D2 to D5 and original signal were used as frequency-domain features.The above features were used as input to support vector machines,k-nearest neighbors,and random forest for detection shockable rhythm.We used Sensitivity(Se)、Specificity(Sp)、Positive Predictive Value(PPV)、F1-Score(F1)and Balance Error Rate(BER)as performance indexes of algorithm,and achieved the best Se,Sp,PPV,F1 and BER were 98.39%,99.17%,98.58%,98.29%and 1.35%,respectively.In addition,we obtained grayscale time-frequency images and pseudo-color time-frequency images by using continuous wavelet transform.These grayscale images and pseudo-color images were respectively used as input to 8-layer convolutional neural network for detection shockable rhythm.Finally,we achieved the best Se,Sp,PPV,F1 and BER were 98.47%,98.66%,97.73%,98.10%and 1.44%,respectively.The superiority of proposed method is proved by comparison of existing methods for detection shockable rhythm.