Analysis and Classification of Vibroarthrographic (VAG) Signals using Statistical Features

Abstract

Human Body joints are vital for normal body movements. Early diagnosis of any problem in joints helps the doctors for the timely treatment of the diseases. To identify these problems, many diagnosis methods are available but the easiest and most effective way is through Vibroarthrography. Vibroarthrography is the method of detecting the vibration signals from the knee joint to diagnose any disorders in it. Researchers are studying the usage of vibration signals from the human knee joints, known as Vibroarthrographic (VAG) Signals, for the diagnosis of the condition of the knee joint. There are various types of features and classifiers used for the classification of VAG signals into normal and abnormal signals. In this research, different types of features of the time domain and spectral domain are explored, and studied the combination of these features. These features include statistical features, Auto-Encoder Based features, and Continuous Wavelet Transform based features. The features are then selected by correlation coefficients and fed into classifiers models. Different classifiers are examined but the best results have been achieved by using the Decision Tree Classifier. The accuracy achieved using the Decision Tree Classifier is 93.26%. We have concluded that the proposed methodology performed very well in other performance evaluation parameters as well. We achieved the Sensitivity of 86.84%, Specificity of 98.04%, PPV of 97.06%, NPV of 90.91%, and a Matthews Correlation Coefficient (MCC) score of 0.8641. The proposed method has Area under the Curve of ROC approximately equal to 0.91. The proposed methodology gives us more accurate results, as compared to previous researches, without going into the Deep learning methods that are complex and time-consuming.