Automated embolus identification using a rule-based expert system.

Transcranial Doppler ultrasound (US) can be used to detect microemboli in the cerebral circulation, but is still limited because it usually relies on "human experts" (HEs) to identify signals corresponding to embolic events. The purpose of this study was to develop an automatic system that could replace the HE and, thus, make the technique more widely applicable and, potentially, more reliable. An expert system, based around a digital signal-processing board, analysed Doppler signal patterns in both the time domain and frequency domain. The system was trained and tested on Doppler signals recorded during the dissection and recovery phases of carotid endarterectomy. It was tested with 74 separate 2.5-min recordings that contained at least 575 artefacts in addition to 253 s of diathermy interference. The results were compared with the results obtained by three HEs. Using a "gold-standard" that classified any event detected by the majority of HEs as an embolus, the automatic system displayed a sensitivity of 94.7% and a specificity of 95.1% for 1151 candidate events 7 dB or more above the clutter (signal-to-clutter ratio, SCR, > or = 7 dB), and 89.6% and 95.3%, respectively, for 2098 candidate events with SCR > or = 5 dB. The system had a very similar performance to individual HEs for SCR > or = 7dB, and was only marginally worse for SCR > or = 5 dB.