Introduction: Recently, the use of synergy analysis to describe dynamic motor control in cerebral palsy (CP) has become increasingly popular, particularly because synergies have been shown to be a strong predictor of treatment outcome . However, several studies have shown that technical issues such as the matrix factorization algorithms used  and EMG processing [3,4] can influence the outcome and interpretation of synergies. Therefore, it is important to study how processing steps can influence synergy outcome measures in CP, especially before and after treatment. Research question: What is the effect of different choices in the EMG analysis (filtering, normalization, and stride selection) on the outcome of synergy analysis in CP, and could this affect the comparison of synergies before and after botulinum neurotoxin (BoNT-A) treatment? Methods: 68 ambulant children with CP (age 9.3 AE 3.6, GMFCS level I–IV) underwent multilevel BoNT-A treatment, and received pre-and 73 AE 25 days post-treatment gait analysis, with EMG measurement of 5 muscles (rectus fem., vastus lat., semitendinosus, tibialis ant. and gastrocnemius med.). EMG data of the most affected leg were high-pass filtered at 20 Hz, rectified, and low-pass filtered at either 2, 3, 10 or 25 Hz. EMG data were either not normalized, normalized to their own maximum, or normalized to the pre-treatment maximum. Finally, EMG was either averaged over all strides (N = 10.8 AE 4.2), or taken from one random stride, or concatenated over all strides. EMG with each combination of filter, normalization, and stride selection was input to a non-negative matrix factorization algorithm (NNMF). An ANOVA for repeated measures was run on the variance accounted for (VAF) by one synergy, to analyse BoNT-A treatment, filtering, normalization, and stride selection effects, as well as their interactions. Results: VAF decreased strongly with increasing filter frequency (Fig. A, p <0.001); and when including individual strides rather than the mean (Fig. B, p <0.001). No main effect of normalization was found. VAF increased slightly after BoNT-A treatment (Effect size 0.010; p = 0.044), but only when higher filter frequencies were used (Fig. A, interaction p <0.001). Discussion: Choices in EMG processing have a large effect on synergy outcomes, which can lead to conflicting findings in terms of treatment outcome. The decrease in dynamic motor control (increased VAF) after BoNT-A treatment, which was largest at higher filtering frequencies, may be due to lower amplitudes and hence reduced signal-to-noise ratio after treatment due to denervated muscles, rather than an actual change in dynamic motor control. This study shows that synergy outcomes strongly depend on EMG processing method, and may help in making deliberate choices during data analysis.
|Number of pages||1|
|Publication status||Published - 2016|
|Event||ESMAC 2016 - NH Collection Hotel, Seville, Spain|
Duration: 29 Sep 2016 → 1 Oct 2016
Conference number: 25
|Abbreviated title||ESMAC 2016|
|Period||29/09/16 → 1/10/16|