Validation of the Automated Electronic Microemboli Detection System in Patients Undergoing Carotid Endarterectomy

Purpose To assess the diagnostic value of automatic embolus detection software (AEDS) in transcranial Doppler (TCD) monitoring for the detection of solid microemboli in patients at risk for perioperative stroke during carotid endarterectomy (CEA).

Materials and Methods In 50 patients undergoing CEA, perioperative TCD registration was recorded. All recorded events, identified and saved by the AEDS, were analyzed off-line doubly by two human experts (HEs) within a time frame of > 4 months. The inter- and intraobserver variability was assessed. The overall agreement with the HEs, the sensitivity, specificity, negative and positive predictive values (NPV and PPV) of the AEDS were computed for different cut-offs (patient displaying perioperative 5, 10, 20, 25, or 50 microemboli).

Results 77 233 events were analyzed. The inter- and intraobserver variability was good (min κ = 0.72, max κ = 0.79). AEDS and the HEs identified 760 and 470 solid emboli, respectively. The agreement between AEDS and the HEs for solid emboli detection was poor (κ = 0.24, SE = 0.016). The specificity and NPV were high (99.2 % and 99.6 %) but the sensitivity and PPV were low (30.6 % and 19.8 %). Applying a threshold of > 20 microemboli resulted in the best sensitivity (100.0 %), specificity (84.4 %), PPV (42.7 %), NPV (100.0 %) and area under the curve (0.898). However, 58.3 % of the patients were false positive as classified by AEDS.

Conclusion In this validation cohort, AEDS has insufficient agreement with HEs in the identification of solid emboli. AEDS and HEs disagree with respect to the identification of specific patients at risk. Therefore, AEDS cannot be used as a standalone system to identify patients at risk for perioperative stroke during CEA.