Classification of Carotid Plaque Echogenicity by Combining Texture Features and Morphologic Characteristics

    loading  Checking for direct PDF access through Ovid

Abstract

Objectives

Anechoic carotid plaques on sonography have been used to predict future cardiovascular or cerebrovascular events. The purpose of this study was to investigate whether carotid plaque echogenicity could be assessed objectively by combining texture features extracted by MaZda software (Institute of Electronics, Technical University of Lodz, Lodz, Poland) and morphologic characteristics, which may provide a promising method for early prediction of acute cardiovascular disease.

Methods

A total of 268 plaque images were collected from 136 volunteers and classified into 85 hyperechoic, 83 intermediate, and 100 anechoic plaques. About 300 texture features were extracted from histogram, absolute gradient, run-length matrix, gray-level co-occurrence matrix, autoregressive model, and wavelet transform algorithms by MaZda. The morphologic characteristics, including degree of stenosis, maximum plaque intima-media thickness, and maximum plaque length, were measured by B-mode sonography. Statistically significant features were selected by analysis of covariance. The most discriminative features were obtained from statistically significant features by linear discriminant analysis. The K-nearest neighbor classifier was used to classify plaque echogenicity based on statistically significant and most discriminative features.

Results

A total of 30 statistically significant features were selected among the plaques, and 2 most discriminative features were obtained from the statistically significant features. The classification accuracy rates for 3 types of plaques based on statistically significant and most discriminative features were 72.03% (κ= 0.571; P < .001) and 88.14% (κ= 0.820; P < .001), respectively. The receiver operating characteristic curve for identifying anechoic plaques showed an area under the curve of 0.918 when the most discriminative features were used to train the classifier.

Conclusions

It is feasible to classify carotid plaque echogenicity by combining texture features extracted from sonograms by MaZda and morphologic characteristics.

Related Topics

    loading  Loading Related Articles