Fig 1.
Block diagram of the proposed approach.
Fig 2.
EMG data from five selected muscles before and after preprocessing for the KOA class.
Fig 3.
The various higher-order spectra for a deterministic signal.
F [.] denotes the k-dimensional Fourier Transform.
Fig 4.
Samples of a Bicoherence representation for EMG signal from five muscles: (a) The Normal EMG from Biceps Femoris; (b) The Normal EMG from Medial Gastroc; (c) The Normal EMG from Rectus Femoris; (d) The Normal EMG from Semitendinosus; (e) The Normal EMG from Vastus Medialis; (f) The Normal EMG from average muscles; (g) The KOA EMG from Biceps Femoris; (h) The KOA EMG from Medial Gastroc; (i) The KOA EMG from Rectus Femoris; (j) The KOA EMG from Semitendinosus (k) The KOA EMG from Vastus Medialis; (l) The KOA EMG from average muscles.
Fig 5.
The architecture of ResNet101.
Fig 6.
(a) Precision and (b) sensitivity for prediction the knee osteoarthritis (KOA) and normal subjects using EMG signals from different muscles. Where biceps femoris (BF), medial gastrocnemius (MG), rectus femoris (RF), semitendinosus (ST), vastus medialis (VM), and average of all five muscles (ALL) groups. Data are presented as mean with standard deviations (mean ±STD). *P<0.5, significantly higher than ALL group.
Fig 7.
(a) Specificity and (b) F-measure for prediction the knee osteoarthritis (KOA) and normal subjects using EMG signals from different muscles. Data are presented as mean ±STD. *P<0.5, significantly higher than ALL group.
Table 1.
Area Under the Curve (AUC), Matthews correlation coefficient (MCC) and accuracy using different muscles to predict knee osteoarthritis (KOA) and normal subjects using EMG signals.
Table 2.
Comparison of the proposed method’s performance with other studies.