REAL-TIME AUTOMATED COMPUTERIZED DETECTION OF VENOUS AIR EMBOLI IN DOGS

Objective. My objective was to develop a real-time pattern recognition system to monitor the precordial Doppler and end-tidal CO2 for charac teristic changes of venous air emboli. The system also must check the adequacy of the input signals, to allow for unattended operation. The sensitivity of the precordial Doppler monitoring of the resulting syst em was the focus of this study. Methods. The computerized system elect ronically sampled systolic sounds, the amplitude envelope of Doppler p ulsations, and, optionally, end-tidal CO2. Features were defined and c alculated from the samples, the means and standard deviations of which were also calculated. During real-time test administrations of intrav enous air in anesthetized dogs, each new sample was compared with prev ious statistics and, when parameters changed beyond calculated limits, an alarm was activated. Results. The sensitivity of the on-line syste m to an intravenous air injection of 0.025 ml/kg was 33%; to 0.05 ml/k g, 73%; to 0.1 ml/kg, 90%; and to 0.2 ml/kg, 100%. A confounding facto r, air lodging in the veins, was detected in the smaller injections; w hen this was corrected, the sensitivity of the system improved beyond these results. Conclusion. An on-line, real-time system, developed for continuous observation of precordial Doppler, has a sensitivity compa rable to human observers. This system may improve clinical monitoring particularly in situations where the occurrence of a venous air emboli sm is not a high probability and, therefore, monitoring is not current ly used because of its requirement for human observation. Systems such as the one described may allow many more patients to be monitored for this complication.