Dear Community,

I am thinking of using some form of matrix decomposition methods to identify muscle modes (aka synergy). This is a question on relative pros and cons of certain methodological factors. I have read important references below, but have not gotten some answers

Issue one:
Oliveira studied multiples of creating the input matrix, 1) one trial, 2) average of many trials (i define trials as one complete cycle e.g. HS to HS), 3) concatenation of multiple trials. The issue then is how can one account for trial to trial variability? (1 and 2) assumes one trial is representative, and (3) accounts for variability at the subject level (assuming the study involve subjects performing multiple gait cycles for walking at one speed).

What I thought would be optimal is at least using a 3-mode array input for decomposition, rather than a common 2 mode matrix. So assuming 10 participants walking for 8 cycles, each time normalized to 101 points, the solution is a subject by time by trial array (10 x 101 x 8). Something like doing a tensorial NMF. are there good reasons why multi-mode methods have not been more common for synergy analysis?

Issue two:

This is a typical synergy experiment in gait. Eg. 10 healthy 10 pathological participants, 5 motor task (walking at 5 speed), each task 8 cycles, each cycle normalized to 101 points

This is a similar issue to above. Synergy seems to be analysed for each subject and for each task. The muscle weights and coefficients are then pooled across participants. I would have thought that a subject by task, by cycle by time (20 x 5 x 8 x 101) array could be used to identify mode, subject weightings for each mode, activation coefficients for each mode, cycle coefficients for each mode. If anyone is wondering what method I am speaking of, they are really simple tensorial extensions of typical algorithms, like PCA, ICA etc

Common question: Are there pros and cons when sticking to traditional matrix input vs an array input when doing matrix decomposition to identify muscle modes? Any input or references would be greatly appreciated. Hope I was clear enough in describing my query.


Motor modules of human locomotion: influence of EMG averaging, concatenation, and number of step cycles.
Oliveira AS, Gizzi L, Farina D, Kersting UG.
Front Hum Neurosci. 2014 May 23;8:335.

Tensorial extensions of independent component analysis for multisubject FMRI analysis.
Beckmann CF, Smith SM.
Neuroimage. 2005 Mar;25(1):294-311. Epub 2005 Jan 8.

Electromyography Data Processing Impacts Muscle Synergies during Gait for Unimpaired Children and Children with Cerebral Palsy.
Shuman BR, Schwartz MH, Steele KM.

The number and choice of muscles impact the results of muscle synergy analyses.
Steele KM, Tresch MC, Perreault EJ.
Front Comput Neurosci. 2013

Int J Neural Syst. 2017 Aug;27(5):1750007. doi: 10.1142/S0129065717500071. Epub 2016 Sep 23.
On the Methodological Implications of Extracting Muscle Synergies from Human Locomotion.

Santuz A1, Ekizos A1, Janshen L1, Baltzopoulos V2, Arampatzis A3.

Matrix factorization algorithms for the identification of muscle synergies: evaluation on simulated and experimental data sets.

Tresch MC1, Cheung VC, d'Avella A.

Methodological Choices in Muscle Synergy Analysis Impact Differentiation of Physiological Characteristics Following Stroke.

Banks CL1,2,3, Pai MM4, McGuirk TE1, Fregly BJ2,4, Patten C1,3.