Detection of Weak Doppler Microembolic Signature using Two-Dimensional-Adaptive Time-Frequency Threshold from Spectrogram

摘要

Prevention of cerebrovascular accidents (CVA) can be achieved by detecting their related precursor signs. A new generation of transcranial Doppler (TCD) systems is presented for detecting the smallest possible microemboli. However, many artefacts appear with these mono-gated Holter TCDs. Thus, the aim of the method becomes achieving microembolic detection while rejecting artefacts. For the clinicians' procedure, the detection proposed is based on an adaptive thresholding applied on the spectrogram of the Doppler signal. The method required achieving three steps. First, the beginning of each cardiac cycle is assessed from the spectrogram of the Doppler signal. Second, by assuming that the Doppler signal is pseudo-cyclostationary, the spectrogram are segmented and time-normalised into sub-spectrograms for each cardiac cycles. Two two-dimensional-adaptive (2D-adaptive) thresholds of detection for microemboli and artefacts were statistically adjusted in both time and frequency. Third, the microembolus detection consists in both detecting the over-intensities in the sub-spectrograms and checking if the detected signatures are not artefacts. The ROC curve results show that the performances are 3.6 times higher compared to those of the standard detection. The detection rate can be increased by 22% compared to standard detection. Besides, the false alarm rate can be reduced by 28%. Using an 2D-adaptive threshold adjusted in both time and frequency, microemboli of weaker intensity can be detected. The analysis of a long acquisition could be possible, and better support of high-risk asymptomatic patient could be considered. (C) 2019 Elsevier Ltd. All rights reserved.