An enhanced algorithm for knee joint sound classification using feature extraction based on time-frequency analysis

Feature Computing and Signal Classifications

“…The logistic regression analysis technique was employed to perform the signal classifications (accuracy: 68.9 %, sensitivity: 0.56, specificity: 0.78) [18]. Kim et al [16] extracted the EP, ESP, FP, FSP features from the enhanced TFD of VAG signal segments, and then used a back-propagation neural network to implement pattern classifications (accuracy: 95.4 %, sensitivity: 0.92, specificity: 0.9868). Umapathy et al [36] used to Daubechies db4 wavelets constructe the local discriminant base (LDB) tree to measure the dissimilarity between the wavelet basis coefficients.…”

Averaging of nonlinearly aligned signal cycles for noise suppression

“…This similarity measure was developed for spoken words recognition [29,22] and later applied in a variety of pattern recognition problems, for instance to ECG beats clustering [27], to knee joint sound classification [11], to sleep stage classification [18] and in [33] more generally to different biomedical time series clustering. In this paper, we focus on application of the DTW technique to noise suppression.…”

Acoustic Analysis and Assessment of the Knee in Osteoarthritis During Walking

“…Significant work was directed to the development of Vibroarthrography (VAG) as an alternative to PAG which relies on accelerometers which are sensitive at frequencies below 1 kHz, to pick up mechanical vibrations. Several algorithms were proposed for classifying the knee VAG signals, according to pathological conditions, using linear prediction modelling [6,7], time-frequency analysis [8] and wavelet decomposition [9]. Features used include waveform variability parameters, spectrogram features, statistical features [10], fundamental frequency, mean amplitude of pitches and their jitter and shimmer [11,12].…”

“…Features used include waveform variability parameters, spectrogram features, statistical features [10], fundamental frequency, mean amplitude of pitches and their jitter and shimmer [11,12]. Various classifiers were considered, from early neural network architectures [8,13] to maximal posterior probability decision criterion [14], bagging ensemble and multiple classifier system based on adaptive weighted fusion [9]. Recently the use of Acoustic Emission (AE) in the ultrasound frequencies for assessing knee joints was explored [15].…”