Fig 1.
The structure of the IZ detection program including: (I) Pre-Processing, (II) Segmentation, (III) Pruning, (IV) Region identification and (V) Innervation points detection.
EMG matrix preparation extracts an appropriate epoch of data for image conversion. Graph-Cut algorithm was used for image segmentation. Parameters (Slope, center/edge coordinates) in step IV were estimated to consider the interaction between regions in the image.
Fig 2.
The segmentation result on a simulated sEMG with five IZs and 20dB SNR.
(a) The image was generated from 60-ms epochs of linear-electrode arrays (spatial interpolation factor of 100 for visualization and the inter-electrode-distance of 5 mm) with the sampling frequency of 4096 Hz and (b) the result of image segmentation via kernel-based Graph-Cut. Propagated potentials from each IZ towards both directions (up and down in the Figure) can be seen.
Fig 3.
The compensation of the breakpoints of the image frames in the pruning step.
(a) the segmented image with discontinuous regions (some breakpoints are shown in rectangular area). (b) The compensated image with fewer breakpoints and intensified regions.
Fig 4.
The morphological processing (pruning) steps on the result of Fig 3 consist of (a) Dilating image with pair mode structure, (b) Opening region via disk structure, (c) Region erosion via line structure and (d) reconstructing region according to the original image.
In such procedures, the irrelevant structures i.e. non-propagating parts were suppressed while the detectability of the propagating regions close to the innervation zones was improved.
Fig 5.
An example of the propagating region identification procedure (stage IV of the proposed algorithm) and feature extraction.
The slope parameters found by center/edge coordinates are shown by triangles and pentagons, respectively. Bold triangles show the closest distance of edges. The center is defined as the center of each propagating region. The edges are the upper and lower boundaries of such regions. The slope is calculated based on the angle between a virtual line representing the propagation region (bold line) and the horizontal line.
Fig 6.
The feature extraction procedure including (a) assigning labels to the paired-regions (paired regions are shown with the same labels). Each region is represented by a straight line to calculate the intersection points. (b) Calculation of the offset of the paired-region and intersection of lines, offsets and intersection points are shown by hexagons and stars inside black circles respectively. (c) The estimation of the innervation zone by choosing appropriate hexagon according to its distance with corresponding intersection point. The detected innervation zones are shown in pentagon stars.
Fig 7.
Examples of the simulated sEMG signals with 20 Single Differential (SD) channels and 60-ms epochs.
The image frames A to D contained 2 IZs (-5 dB SNR), 3 IZs (0 dB), 4 IZs (5 dB) and 5 IZs (10 dB), respectively. The location of the simulated IZs is shown by circles. The developed program automatically identified the location of IZs as the crossing of the ‘v’ shape propagation lines (upper lines in blue and lower lines in red color). The CV of the identified propagation pattern was then estimated by the proposed algorithm.
Table 1.
The spatial distribution parameters of the simulated EMG frames (MEAN±SD).
Table 2.
The performance of the proposed IZ detection algorithm on the simulated dataset (MEAN±SD).
Table 3.
The overall performance of the innervation zone detection algorithm.
Table 4.
The absolute channel error (IED ratio) for the innervation zone detection (MEAN±SD).
Table 5.
The absolute and relative muscle fiber conduction velocity error in m/s, and percentage, respectively of the proposed algorithm (MEAN±SD).
Table 6.
The running time of the innervation zone detection algorithm on each 60-ms sEMG frame in MEAN±SD.
Table 7.
Additional analysis of the proposed IZ detection algorithms.
Fig 8.
Examples of the simulated sEMG signals with 20 Single Differential (SD) channels and 60-ms epochs.
The image frames A to D contained 2 IZs (-5 dB SNR), 3 IZs (0 dB), 4 IZs (5 dB) and 5 IZs (10 dB), respectively. The location of the simulated IZs is shown by circles. The developed program automatically identified the location of IZs as the crossing of the ‘v’ shape propagation lines (upper lines in blue and lower lines in red color). The CV of the identified propagation pattern was then estimated by the proposed algorithm.