Knee osteoarthritis condition
Number of knee joint sound records
Healthy (stage 0)
250
Minor (stage 1)
80
Mild (stage 2)
90
Moderate (stage 3)
120
Severity (stage 4)
110
Total
650
Table 1: The Overall dataset obtained for each class
Roll No
Selected features
Domain of the features
1
Peak2rms
Time domain
2
Snr
3
spectral skewness
Frequency domain
4
Skewness
5
Rms
6
Kurtosis
7
RDE std
8
Spectral roll off point
9
Spectral spread
10
Spectral flux
Table 2: Extracted features using audio feature extractor
Models
Algorithm
Accuracy (%)
Lstm
80.71%
SVM
Quadratic SVM
81.8%
Cubic SVM
88.9%
Fine gaussian SVM
86.1%
KNN
Fine KNN
87.5%
Weighted KNN
89.6%
Ensemble
Bagged Trees
90.00%
Table 3: Comparison between different models
Figure 1: General block diagram of the method used
Figure 2: SG electret microphone
Figure 3: single head stethoscope with its extension (a), the separated stethoscope head (b)
Figure 4: Sensor construction with microphone cable (a), unplugged stethoscope head (b)
Figure 5: Designed electronic stethoscope
Figure 7: sample of knee joint signal at different stage (normal, minor, mild, moderate and severe)
Figure 8: Filtered signal using 4th order Butterworth and notch filter
Figure 9: Training dataset of each class before performing class balance (a), and after performing class balance (b).
Figure 10: Augmented vibroarthrographic signal
Figure 11: Training models of bagging ensemble method [70]
Figure 12: New data on bagging ensemble method
Figure 13: EWT of a signal
Figure 14: Selected features
Figure 15: Confusion matrix of test set of different models
Tables at a glance
Figures at a glance