Feature selection based on RF signals and KNN Rule: Application to microemboli classification

In the human body, emboli can produce severe damage like stroke or heart attack. Commonly used Doppler detection techniques have shown their limits in the determination of the embolus nature. An alternative approach would be to examine Radio Frequency (RF) signal instead of Doppler signals. Under specific conditions of the ultrasound excitation wave, gaseous bubbles show a nonlinear behavior exploited to distinguish gaseous from solid microemboli. Fundamental and second harmonic signals amplitudes and bandwidths are selected for input parameters. Moreover, fundamental and second harmonic spectral components have been approximated by Gaussian functions. In this paper, we propose a new approach for feature selection based on the K-Nearest Neighbors Rule (KNNR). The technique proved an effective improving classification. Feature selection and extraction not only indicate a good ability to find the most relevant set of inputs that result in higher classification accuracy but also to reduce the size of feature vector.