Data for: Surface EMG muscle activation patterns of the lower extremities during gait in individuals with and without a knee injury history
Description
This data set was used to investigate lower extremity muscle activation patterns during treadmill walking in young adults with and without a previous intra-articular knee injury 3-12 years ago. The hypothesis was that systematic adaptations in time-frequency activation patterns of lower extremity muscles during gait persist more than three years post-injury. Bipolar surface EMG signals and heel accelerations (1D) were recorded at 2400 Hz from 71 individuals with and without a history of knee injury 3-12 years prior during one minute of treadmill walking (4.5 km/h). EMG signals were obtained bilaterally from five muscles: Vastus lateralis (VL), biceps femoris (BF), medial hamstrings (MH), gastrocnemius lateralis (GL), gastrocnemius medialis (GM). Recordings from 61 individuals were used for further analysis. Raw EMG data and the time points (sample numbers) of heel strike for the right and left leg of each individual can be found in the .mat file 'Raw_Data'. Raw EMG signals were transformed into time-frequency space using a wavelet analysis (for details, see: von Tscharner, 2000, "Intensity analysis in time-frequency space of surface myoelectric signals by wavelets of specified resolution"). EMG intensities were resampled to 250 time points, summarized in three frequency bands: Low 30-60Hz, Mid 60-100 Hz, High 100-300Hz, and amplitude-normalized to the overall, mean EMG intensity across all 49 gait cycles. Normalized EMG intensities were then averaged across the 49 gait cycles, yielding one EMG intensity pattern for each of the five muscles, representing 3 frequency bands and 250 time points (5x3x250 = 3750 elements per pattern). Thus, there are 122 MMPs, one for each leg of the 61 participants that were appended row-wise in matrix M in file 'M.mat'. A PCA was applied to the data matrix M (for details, see: von Tscharner et al., 2018, "A wavelet based time frequency analysis of electromyograms to group steps of runners into clusters that contain similar muscle activation patterns"). The residual mean (PC 0) and the first 10 PC-vectors (70% of the original variance) are saved in the file 'PC-vectors-features.mat' and the corresponding weights in the matrix W are saved in 'W_unwhitened.mat'. W was whitened and used as an input to a support vector machine analysis to investigate systematic differences in lower extremity muscle activation patterns between legs that had sustained a previous injury vs. legs with no injury history. The results showed that trained classifiers could successfully recognize whether muscle activation patterns belonged to the affected or unaffected leg of previously injured individuals. In contrast, it was not possible to discriminate between patterns belonging to the previously injured legs or dominant legs of control subjects. In conclusion, systematic knee injury effects on the neuromuscular control of the knee during gait were present 3 to 12 years later.
Files
Steps to reproduce
Details related to the EMG pattern and SVM analysis are described in detail in the manuscript "Mohr M, von Tscharner V, Emery CA, Nigg BM. (2018). Classification of gait muscle activation patterns according to knee injury history using a support vector machine approach" that is currently under review in the journal 'Human Movement Sciences'. Once this manuscript is accepted, more detailed information on the methodology can be gathered from the manuscript. Raw EMG signals were transformed into time-frequency space using a wavelet analysis (for details, see: von Tscharner, 2000, "Intensity analysis in time-frequency space of surface myoelectric signals by wavelets of specified resolution"). Specifically, the raw EMG signals were convolved with 30 non-linearly scaled wavelets (scale = 1.6 in von Tscharner, 2000) spanning center frequencies between 1 and 500 Hz to extract the EMG power in 30 frequency bands as a function of time. The square root of the power yields the EMG intensity, which was used as a measure of muscle excitation. A time window of EMG intensities was selected starting 30% of gait cycle duration before the current heel strike and ending 70% of gait cycle duration after the heel strike. Within this window the EMG intensities were resampled to 250 time points. EMG intensities were summarized in three frequency bands: Low 30-60Hz, Mid 60-100 Hz, High 100-300Hz and amplitude-normalized to the overall, mean EMG intensity across all 49 gait cycles. Normalized EMG intensities were then averaged across the 49 gait cycles, yielding one EMG intensity pattern for each of the five muscles, representing 3 frequency bands and 250 time points (5x3x250 = 3750 elements per pattern). Thus, there are 122 MMPs, one for each leg of the 61 participants that were appended row-wise in matrix M. M is can be found in the file 'M.mat'. A PCA was applied to the data matrix M (for details, see: von Tscharner et al., 2018, "A wavelet based time frequency analysis of electromyograms to group steps of runners into clusters that contain similar muscle activation patterns"). The residual mean (PC 0) and the first 10 PC-vectors (70% of the original variance) are saved in the file 'PC-vectors-features.mat' and the corresponding weights in the matrix W are saved in 'W_unwhitened.mat'. All uploaded files have the following structure: Each row corresponds to one leg of one individual, right legs have the label 1 and left legs have the label 2. The rows in 'Raw_Data' represent the same legs as in 'M' and in 'W'. 'Anonymized_Data_Identifiers' indicates injury history, leg dominance, sex and anthropometrics corresponding to each row in 'Raw_Data', 'M', and 'W'. The PCA output was used as an input to a support vector machine analysis described in the manuscript currently under review.