Feature extraction and classification of knee joint disorders using Hilbert Huang transform

Non-invasive investigation methods along with computer based exploration of vibroarthrography (VAG) signals can contribute compiling indication of human knee-joint deformity. The VAG signals are characterized as non-stationary and aperiodic in nature. As a result, feature extraction technique is challenging for researchers. This paper proposes analysis of VAG signal using Hilbert-Huang transform (HHT). The ensemble empirical mode decomposition (EEMD) decomposes raw VAG signal individual characteristic scales known as intrinsic mode function (IMF). The analytical signal representation of IMFs is attained by implementing Hilbert transform on IMFs. In the z-plane, the fundamental analytic IMFs are plotted which are circular in geometry. Area of these circular curves in the z-plane are computed using the central tendency measure (CTM) and chosen as feature in differentiating between healthy and unhealthy VAG signals. A pattern analysis is carried out using least square support vector machine (LS-SVM) which gives a classification accuracy of 83.12% and area under receiver operating characteristic of 0.6708 were obtained.