Semiautomatic detection of microbubble ultrasound contrast agent destruction applied to definity® using support vector machines

For different applications such as imaging, drug delivery, and tissue perfusion measurement, it is necessary to know the inertial cavitation (IC) threshold of ultrasonic contrast agent (UCA) microbubbles. Even though the influence of the incident acoustical pressure, frequency and pulse duration (PD) in the regime of the microbubble's response is well established, the investigation of the IC threshold is essential for the accuracy of some measurement techniques and for ultrasound safety. The goal of our work was to find the IC threshold for the FDA-approved UCA Definity®. The dependency of the threshold on the peak rarefactional pressure and PD of an incident toneburst was investigated. The experiments performed to estimate IC thresholds yield a large amount of data to be classified in the five following classes: Noise, Oscillation, Collapse, Multiple Bubbles and Unknown. A reduction of the manually to classified data was reduced by using a semiautomatic algorithm in order to achieve a low variance in the IC estimates. Further more significant features to distinguish between classes were found and tested. The development of a heuristic algorithm to detect events of thee class Collapse was not successful due to the fact that the classes were overlapping and some signals could not be classified to a single class. Therefore, a semiautomatic algorithm using support vector machines was developed.