A Method for the Automated Detection of Venous Gas Bubbles in Humans Using Empirical Mode Decomposition

Abstract

Doppler ultrasound signals are widely used to grade the quantity of circulating venous bubbles in divers. Current techniques rely on trained observers, making the grading process both time-consuming and subjective. The automated detection of bubbles, however, is confounded by the presence of other signals, primarily those arising from blood motion. Empirical Mode Decomposition was used here to calculate the intrinsic mode functions (IMFs) of a number of Doppler ultrasound signals from recreational divers, post-decompression. The IMFs provide a basis set for signal decomposition, each IMF corresponding to a different timescale in the signal. Each signal was found to comprise approximately 20 IMFs: the precise number being dependent upon the nature of the signal. A method is presented to detect bubbles using the IMF; features are first identified in the individual heart cycles, these having been previously determined using a robust peak detection method, by examining deviations from the ensemble averaged IMF. Bubbles are then identified as features appearing in more than one IMF, with significant energy in the original signal. This method has been applied to a subset of the available database and appears to perform with good sensitivity even when the signal has variable signal strength.