Automated Cerebral Emboli Detection Using Adaptive Threshold and Adaptive Neuro-Fuzzy Inference System

Abstract

This work proposes an automated algorithm based on adaptive threshold and adaptive neuro-fuzzy inference system (ANFIS) to couple with transcranial Doppler ultrasound in detecting cerebral embolic signal (ES). Our main objective is to support practical stroke risk monitoring in interventional procedures. Suspected ESs are captured in real time using adaptive thresholds based on 1) standard deviation, to capture suspected ESs of long duration and 2) median absolute deviation, to capture the shorts, which proved to be the key contribution of this paper. For classification using ANFIS, handcrafted feature extraction is performed and the resulting features are classified as embolic or non-embolic. The effectiveness of the classifier was evaluated over 19 subjects going under procedures generating emboli and compared with the Euclidean matrix-based indexing high-dimensional model representation system. The ANFIS-based system yielded in average of 91.5% sensitivity, 90.0% specificity, and 90.5% accuracy significantly outperformed the HDMR system and the hybrid of HDMR system and the proposed features in both detection accuracy [F(2,57) = 10623.05, p < 0.0001] and sensitivity [F(2,57) = 10572.12, p < 0.0001] at 90.0% specificity. The system using adaptive threshold to capture suspected intervals and ANFIS to identify ES has promising potential as a medical decision support in various clinical settings, e.g., real-time monitoring of cerebral emboli in carotid artery stenting procedures.