In this paper, we report on using pattern recognition techniques for embolic signal(ES) detection based on transcranial doppler ultrasound(TCD) audio data collected via machine EMS-9(from Shenzhen Delicate Electronics...In this paper, we report on using pattern recognition techniques for embolic signal(ES) detection based on transcranial doppler ultrasound(TCD) audio data collected via machine EMS-9(from Shenzhen Delicate Electronics, Co. Ltd).Firstly, we adopted complex discrete fourier transform to get spectra of audio recordings; secondly, we used principal component analysis(PCA) for the visualization of selected signals, which makes it easy and intuitive to verify whether a signal contains an embolic component; finally we designed the classifier with support vector machines(SVM) for detection. With contrast to traditional methods of ES detection systems, the proposed approach considers two channel signals from the audio data collected by single transducer, and there is no predefined features for classification. The primary experimental results on real data are promising.展开更多
基金SZU R/D Fundgrant number:201054+3 种基金Natural Science Foundation of Shenzhengrant number:JC201005280685AKey Program of National Natural Science Foundation of Chinagrant number:61031003
文摘In this paper, we report on using pattern recognition techniques for embolic signal(ES) detection based on transcranial doppler ultrasound(TCD) audio data collected via machine EMS-9(from Shenzhen Delicate Electronics, Co. Ltd).Firstly, we adopted complex discrete fourier transform to get spectra of audio recordings; secondly, we used principal component analysis(PCA) for the visualization of selected signals, which makes it easy and intuitive to verify whether a signal contains an embolic component; finally we designed the classifier with support vector machines(SVM) for detection. With contrast to traditional methods of ES detection systems, the proposed approach considers two channel signals from the audio data collected by single transducer, and there is no predefined features for classification. The primary experimental results on real data are promising.
