International Journal of Artificial Intelligence and Neural Networks
Author(s) : DZATI ATHIAR RAMLI, HARYATI JAAFAR, NAJAH GHAZALI
Occurrence of embolism from patients who suffer from carotid artery stenosis may bring to the onset of stroke if it became severe. In clinical practice, Doppler ultrasound technique is commonly used to detect the emboli in the cerebral circulation. Instead of depending on human observer as a gold standard to detect the emboli, this study proposes an automated embolic identification system based on ultrasound signal analysis. Experimental studies on 1,400 samples from five independent data sets are employed in this study. Two feature extraction methods based on spectral feature i.e. Linear Prediction Coefficient (LPC) and statistical features i.e. combination of Measured Embolus-to-Blood Ratio (MEBR), Peak Embolus-to-Blood Ratio (PEBR), entropy, standard deviation and maximum peak are used to extract the signal. Subsequently, four classifiers based on nearest neighbor approach i.e. k Nearest Neighbor (kNN), Fuzzy k-Nearest neighbor (FkNN), k Nearest Centroid Neighbor (kNCN), and Fuzzy-Based k-Nearest Centroid Neighbor (FkNCN) are used to evaluate the performance of the identification system. The experimental results show that FkNCN with statistical feature outperforms the other classifiers with the performance of 92.45±2.12% is achieved.