Doppler embolic signal detection using the adaptive wavelet packet basis and neurofuzzy classification

The transcranial Doppler (TCD) ultrasound technique is widely applied to detect emboli within the middle cerebral artery (MCA). However, due to the interference of certain artifacts including the probe motion and patient movement, etc. the detection results obtained with the TCD are always obscured. In traditional methods, the spectrogram analysis and/or the wavelet transform are performed to represent Doppler ultrasound signals and extract sensitive characteristics. Unfortunately these features do not have ideal specificity and sensitivity because of limitations on the time and frequency resolution and a lack of an adaptive classifier. In this paper, a new method based on the adaptive wavelet packet basis (AWPB) is used to make a sparse representation of Doppler ultrasound blood flow signals. Different from other dimensionality reduction methods, both the approximation coefficients and the decomposition scales are extracted to represent the signal and sent to the Takagi–Sugeno (T–S) neurofuzzy classifier. The T–S fuzzy inference system can easily combine the linguistic expert rules to make a more robust decision score to characterize emboli. Experiments show that within a 15% confidence interval of the decision score, 99.0% and 97.1% detection rates of emboli can be obtained for simulated and in vivo studies, respectively. The proposed AWPB method and neurofuzzy classification quite outperform traditional methods and are suitable to detect Doppler embolic signals with a high performance.