Computer-aided diagnosis for knee meniscus tears in magnetic resonance imaging

Abstract

Magnetic Resonance images are used to identify tears in the meniscus, a debilitating condition of the knee. In this paper, a two-staged computerized system for diagnosing meniscus tears is proposed. In the first stage, we apply an Active Contour with Level Sets model to calculate the location and shape of the meniscus area. In the second stage, 180 features, consisting of textural information in the spatial and spectral domains, were extracted from each suspected meniscus tear. Sequential floating forward selection (SFFS) was applied to select the relevant features. The feature vectors were then input to a support vector machine (SVM) classifier to detect meniscus tears. The Az value of the receiver operating characteristic (ROC) curve were used to evaluate the classification performance. Experimental results show that the dimension of the feature vector was reduced from 180 to 64 after the SFFS. Meanwhile, the SVM classifier combined with the SFFS feature selection (Az = 0.9123) outperformed the SVM classifier without feature selection (Az = 0.7273).