A NEW METHOD FOR AUTOMATIC DIFFERENTIATION BETWEEN CEREBRAL EMBOLI AND ARTEFACTS
Introduction: Cerebral embolus monitoring systems suitable for routine clinical use must have the ability to automatically recognise and differentiate between artefacts and emboli. This has to date proven to be an extremely difficult problem to solve. Methods: In this study we present a new advancement with regard to the automatic recognition of cerebral emboli and differentiation from artefacts based on a binary decision tree which includes a completely new parameter. This is the 1/4 Doppler shift for the maximum power reflection of an embolic event at 2.5 MHz insonation frequency compared to 2.0 MHz. A new multifrequency transcranial Doppler system together with this software was used in this study of 2000 artefacts and 100 embolic events in one heart valve patient. The level for event recognition was set at 5db above background Doppler power. The artefacts in 2 healthy controls consisted of 200: tapping the probe, 200: tapping the skull, 200: talking (counting), 200: swallowing, 400: coughing, 200: wrinkling the forehead, 200: clenching the teeth, 400: movement of the skin near the probe. Results: Only two (skin movements near the probe) of the 2000 artefacts were recognised as embolic events. All 100 heart valve emboli were detected and 98 (98%) were correctly classified (specificity 98%). Conclusion: This study suggests that a binary decision tree including 1/4 Doppler shift assessment at 2.0 and 2.5 MHz insonation frequencies will greatly improve the ability to carry out automatic differentiation between artefacts and cerebral emboli during monitoring in routine clinical situations.