Enhanced Ultrasound Classification of Microemboli Using Convolutional Neural Network

Abstract

Classification of microemboli is important in predicting clinical complications. In this study, we suggest a deep learning-based approach using convolutional neural network (CNN) and backscattered radio-frequency (RF) signals for classifying microemboli. The RF signals are converted into two-dimensional (2D) spectrograms which are exploited as inputs for the CNN. To confirm the usefulness of RF ultrasound signals in the classification of microemboli, two in vitro setups are developed. For the two setups, a contrast agent consisting of microbubbles is used to imitate the acoustic behavior of gaseous microemboli. In order to imitate the acoustic behavior of solid microemboli, the tissue mimicking material surrounding the tube is used for the first setup. However, for the second setup, a Doppler fluid containing particles with scattering characteristics comparable to the red blood cells is used. Results have shown that the suggested approach achieved better classification rates compared to the results obtained in previous studies.