Assessment of knee joint abnormality using Acoustic Emission sensors

The aim of this project is to distinguish the knee joint condition between normal and osteoarthritis subject by using acoustic signals. There were 17 subjects participated in this study and 8 of them are normal subject, while others are osteoarthritis subject. Acoustic Emission (AE) wave was produced, when the stress bone is subjected to external forces and cause the friction between the cartilages. A data acquisition protocol was developed to obtain the AE signal from subjects. Sit-stand-sit and swing the leg movements were used to record AE signals. The recorded signals were decomposed up to level 4 using Wavelet Packet Transform (WPT) with mother wavelet function of Daubechies (db) 44. Skewness and kurtosis were extracted from the each decomposed signal. The dimension of extracted features was reduced using Principal Component Analysis (PCA). The features before and after PCA were classified using Feed Forward Neural Network (FFNN) and Support Vector Machine (SVM) and obtain the highest mean accuracy of 83.6% (sit-stand-sit) and 85.76% (swing the leg) using FFNN for dataset after PCA.