Processing time improvement for automatic embolic signal detection using fuzzy c-mean

Abstract

Transcranial Doppler (TCD), a non-invasive approach to measure blood flow velocities in brain arteries, can be used to detect emboli in cerebral circulation. Classification of a measured TCD as an embolic signal (ES) or artifact is usually performed by a well-trained physician referred to as a gold standard. However, human error and inter-rater reliability among physicians are unavoidable issues. Therefore, an automatic ES detection system is useful as a medical support system especially for the countries where a number of well-trained physicians are limited. However, in clinical application, the computation complexity of the automatic ES detection algorithm should have been considered. As an example, our previous work, the automatic embolic signal detection algorithm using adaptive wavelet packet transform (AWPT) and adaptive neuro-fuzzy inference system (ANFIS) (Lueang-on et al., Proc. of ISC, 2013), could provide impressive sensitivity and specificity, the algorithm is considerable complicated. In this study, we aim to develop further the algorithm that still provides high detection accuracy yet significantly reduces the processing time. To do so, a number of fuzzy rules in the ANFIS model are optimized. Two data sets, training and validation sets composed of 176 ESs and 484 artifacts were used to evaluate the algorithm resulting in a sensitivity of 95.5% and specificity of 95.4%. The processing time for classification can be reduced by 63% compared with our previous algorithm. The results suggested that the algorithm could be used as a medical support system more efficiently.