Online automated detection of cerebral embolic signals from a variety of embolic sources.

A major limitation of embolic signal (ES) detection by transcranial Doppler ultrasound is the lack of a reliable automated system. The performance of an automated system needs to be evaluated for different embolic sources on consecutively acquired typical data. We evaluated a new online frequency filtering approach in a total of 565 h of data containing 925 ES from four groups of patients: post carotid endarterectomy (postCEA), symptomatic carotid stenosis (SCS), asymptomatic carotid stenosis (ACS) and atrial fibrillation (AF). The following sensitivities and specificities were achieved: postCEA = sensitivity 95.8%, specificity 88.2%; SCS = sensitivity 98.4%, specificity 88.6%; ACS = sensitivity 85.7%, specificity 13.0%; AF = sensitivity 54.8%, specificity 7.0%. This online automated system performed similarly to the human expert in the postCEA and SCS groups, but less well in patients with AF and ACS. The low ratio of ES to normal data in patients with ACS may have contributed to the lower specificity; further evaluation with a higher number of ES is required. Refinement of the algorithm is required to improve its sensitivity for AF data.