Workflow and positioning analysis are critical for planning and de-signing low turbulence displacement or laminar airflow OR ventila-tion systemsWorkflow and positioning analysis are critical for planning and de-signing low turbulence displacement or laminar airflow OR ventila-tion systems

This abstract describes a new method for automatic separation of muscle signals into underlying motor unit components. From non-invasive high-density surface electromyography (HDsEMG), a motor unit is estimated and subsequently removed from the dataset, iteratively revealing more motor units. Such decomposition of HDsEMG is useful in analysis of motor unit patterns, providing insights to diseases and injuries affecting motor function. It can also provide better functional understanding of the EMG signal, to help devise strategies for more natural control of prosthetic limbs. Direct translation of user intent from motor units, and by extension individual motor neurons, would provide a more intimate relation to prosthetic devices, ultimately reducing rejection rates. The large number of electrodes in HDsEMG allows for the use of Independent Component Analysis (ICA). The basics of ICA is that the recorded signals are a linear mix of a number of sources, and reversely, an un-mixing matrix can be found in a search which maximizes the independence of sources. For decomposition of muscle signals, a fast fixed-point algorithm called FastICA is used to extract a single source estimate at a time. The estimated source output contains many action potential firings. These action potentials are sorted into spike trains using a density-based clustering approach to ensure consistency of the firings in a motor unit. The motor unit itself is removed using a compressed spike-triggered averaging process. The average signal for each electrode, across the time-instances of each spike, is calculated. Compression of principal components is then applied as a noise-reduction step, after which the motor unit is subtracted from the dataset, and the process begins again (Figure 1). After a set number of iterations, a variance metric of the source residuals is used to determine the reliability of each estimate. Figure 1 illustrates the iterative process in decomposition of muscle signals. HDsEMG data is fed into the FastICA algorithm, which produces a source estimate. From this estimate, peak-detection and waveform clustering extracts motor unit spike trains. The spike trains are used to calculate the spike-triggered average of the motor unit, which is compressed by principal components, before it is subtracted from the dataset for each firing. Innovationsverktyg för medicinsk teknik MedTech Innovation Model Thomas Mejtoft1,4, Olof Lindahl2,1,4, Fredrik Öhberg2,1,4, Linda Pommer1,4, Karolina Jonzén2,4, Britt Andersson1,4, Anders Eklund1,2,4, Anders Wåhlin1,2,4, Fredrik Nikolajeff3,4, Nina Sundström2,1,4, Per Hallberg2,1,4 1Umeå universitet, 2Region Västerbotten, 3Luleå tekniska universitet, 4Centrum för medicinsk teknik och fysik (CMTF)