Improved detection of embolic signals using multi scale wavelet filtering, AR and ANN, for TCD ultrasound

Transcranial Doppler ultrasound can be used to detect emboli in blood flow in the brain. The presence of emboli is an indication of high risk of stroke. Embolic signals have characteristic transient chirps suitable for wavelet analysis. We have implemented an on-line intelligent wavelet pre-filter combined with a new frequency based neural network classification system (NFS) to produce a new online detection system. Initial results show an improvement in accuracy compared with the widely used FS-1 system. Our system makes, use of multi-scale wavelet denoising using an adaptive coefficient threshold. The pre-filtering system is combined with a detection system which uses a two layer neural classifier and a new auto-regressive event detector. For conditions such as carotid stenosis an improvement of 20% in detection accuracy was obtained. Our online (real time) intelligent wavelet amplifier and its matrix optimised form uses the matched filter properties of multiple coefficients from multiple wavelets to significantly enhance embolic signals and improve classification performance.