A new method for denoising knee joint vibroarthrographic signals

Analysis of knee joint vibration (VAG) signals using signal processing, feature extraction and machine learning algorithms has shown promise for performing non-invasive diagnosis of knee joint disorders. However for such techniques to yield reliable results, the extracted features must only reflect the essential information contained in the signal. The signals must hence be accurately denoised and free from artifacts. This paper proposes a new method for denoising VAG signals using ensemble empirical mode decomposition, followed by wiener-entropy based thresholding and filtering to accurately remove noise contained in the signal. Standard metrics : mean squared error, mean absolute error, signal to noise ratio, peak signal to noise ratio and cross-correlation and CPU consumption time were used to assess the performance of this method. Metric normalized root mean squared error has been used to evaluate the effectiveness of our method in denoising synthetic signals with additive noise. In both scenarios, the proposed method yielded a superior performance in terms of accuracy of denoising in comparison to wavelet-soft thresholding but with increased computational cost.