A recurring trend among numerous publications is that there exists an increasing gap between academic research and clinical use. This gap at present only continues to widen while prosthesis rejection remains a pressing issue. Frequently prosthesis users may choose an aesthetically pleasing prosthesis over functionality or in the cases when functionality is desirable then simple direct control systems may be chosen. Ideally a prosthesis should enable intuitive and natural control, which is robust to long term and dynamic use, providing real time performance for low computational complexity and low user burden during use. Other traits that would be desirable would be a reduced complexity of implemented hardware design while still detecting a good degree of muscle activity, easier user training, and closed loop feedback that could better facil- itate the rehabilitative training process and to better inform the user when interacting with their environment.
Although numerous efforts have been undertaken as to implement these ideal prop- erties in recent years, there still remains numerous challenges and limitations along the road to developing an ideal prosthesis implementation.
2.5.1
Inherent Sensing Challenges Posed During Long Term Use
Although many researchers have demonstrated the capability of sEMG based sensing to produce high quality results during offline performances and within short labora- tory sessions, these performances still experience challenges when extending to longer period use and inter day use. Typical pattern recognition approaches tend to focus on the assumption that the produced sEMG signals will remain either consistent during use. However, clinical and daily implementations of prosthesis typically differ greatly in the range of physiological variables experienced and environmental variables which produce inconsistent signals which are difficult to predict within a laboratory environ- ment. As a consequence of this inherent instability and variation across daily use of a prosthesis many pattern recognition based approaches would require retraining or re-calibration that imposes heavy burden upon the user. Although it may be possi- ble to prepare for individual variables, such as electrode shift, typically inter day use
of a powered prosthesis would encounter a combination of impeding factors that fre- quently will be worsened by aspects that introduce muscle produced crosstalk which poses increased randomness in a given system.
2.5.2
Limitations of Single Modality Sensing
The current state of the art shows sEMG based sensing to be very promising in terms of accuracy within laboratory environments and adequate for simple prosthesis con- trol. However, within clinical environments there exists challenges towards interpret- ing muscle activity over long periods of time due to factors such as fatigue producing crosstalk from neighboring muscle groups. As sEMG based sensing attempts to sense the manifestation of electrical energy as it transmits across the upper layers of fibres within the human body there exists the inherent bias of sEMG based sensing towards surface level muscle activity. While it is very possible to accurately infer motion intent from surface level EMG signals there exists numerous dexterous finger movements that are closer related to deep muscle activity. As a result of this bias towards surface level muscle activity the ability to recognize dexterous hand motions may become lim- ited, especially during long term use where physiological changes within the signal may contribute to unstable sEMG based detection whereas physical muscle activity may appear more constant for other methodologies.
2.5.3
Haptics - Closing the Loop
As discussed, an ideal prosthesis can enable not just intuitive control and interaction within an environment but to further enable an accurate sense of self with the pros- thesis and during interaction. Through poor control schemes and lack of adequate feedback during use the issue of prosthesis rejection becomes more prominent. Unfor- tunately, although there exists multiple viable methodologies for provision of haptic feedback these methodologies are seldom implemented within active prosthesis con- trol schemes. While novel haptic feedback approaches such as thermal or stretch based feedback can inspire realistic sensations of touch they still maintain caveats such as limited safe usage during long term use and more importantly present implementa- tions of these feedback methods are bulky and cumbersome therefore limiting their viability within a upper limb prosthesis. Furthermore, it has been recognized that a
feedback approach must be capable of seemingly providing instantaneous feedback whether based on prosthesis activation of whether from sensed interaction via force sensors. Vibrotactile feedback proposes promising compact solutions for implemen- tation within a prosthesis yet the ramp up time of this methodology may cause false positives or confusing feedback during dynamic use. Electrotactile feedback appears as a promising route of feedback due to instantaneous response with no ramp up time and present implementations can be made in a non bulky, wearable, manner. To date, while many promising haptic feedback solutions to enable better prosthesis control have been proposed there still exists limited research into clinical use.
2.5.4
Burden of Rehabilitation and User Training
The concept of adequate rehabilitative results are distinctly varied based on the in- tended outcomes. However, there exists the complication of defining a general ap- proach to an ideal outcome. While exhaustive rehabilitation and training may provide desirable results the degree of repetitive and cumbersome processes involved make such exhaustive practices undesirable to users especially given the potential burden imparted on them. It is within this challenge of recognizing that user training can promote the formation of consistent sEMG patterns in users while also balancing the extent and implementation of user training. Many laboratory environment implemen- tations of sEMG based sensing focus on training the user for consistent single strength sEMG patterns whereas this challenge is further complicated when considering that daily needs of a user may require dynamic contraction strengths.