Page 217 - Handbook of Biomechatronics

214 Georgios A. Bertos and Evangelos G. Papadopoulos

poor clinical performance, there are many practical challenges which
impede clinical repeatability and robustness such as: electrode shift from
prolonged wearing of prosthetic socket,variationinforce levels,and var-
iation in EMG due to recording environment (skin impedance changes
over time, motion artifacts, noise, etc.). As accuracy falls below 85%, a
multi-DoF prosthesis can become frustrating and usability is poor
(Scheme and Englehart, 2011).
In order for surface EMGs and pattern recognition’s practical problems
to be minimized, frequent training sessions have been proposed. These fre-
quent sessions are not welcome by the amputees which they expect their
prosthesis to perform like the natural hand. This makes it even more difficult
for pattern recognition controllers to be used clinically.
The main advantage of surface EMG, its noninvasiveness comes with a
lot of limitations which might make the application of pattern recognition
not practical clinically. The solution to all these problems is the use of IMESs
which will reduce or eliminate all the above problems. Use of IMESs will
enable simpler pattern recognition classifier performances since they will
be close to the “source of truth,” that is, the muscles generating the move-
ments instead of the surface of the forearm.
A pattern classifier able to adapt to changes has been shown to decrease
the error % of the pattern recognition (Sensinger et al., 2009). Unsupervised
(user’s intended class is known) pattern recognition for myoelectric control
resulted in at least 26% reduction in error and supervised pattern recognition
led to smaller reduction in error due to higher uncertainty of correct class
(Sensinger et al., 2009).
In addition, Hargrove et al. (2017) have found that pattern recognition as
a method for controlling a multi-DoF prosthesis is superior when used for
TMR subjects than direct EMG control.
The Osseointegration group in collaboration with Integrum from
Sweden has developed and has made available an open source platform
called Biopatrec. Biopatrec uses pattern recognition algorithms on acquired
from the subject in real-time multi-site EMG signals for the control of a
multi-DoF prosthetic arm or hand (Ortiz-Catalan et al., 2013, 2014b).
On the other hand, a competitor of Biopatrec, which is CoApt, LLC,
a Chicago-based startup connected to researchers from Shirley Ability
Lab—formerly RIC, is commercializing a pattern recognition multi-DoF
myoelectric controller with 3-DoFs: one DoF at the hand and the remaining
to be at the elbow and wrist (Parker et al., 2006).