CAPÍTULO II: Marco Teórico
2.1 Antecedentes de la investigación
2.1.7 Neoinstitucionalismo y acción pública: Elementos base para
It has been emphasised in various publications that the gap between clinical use and academic research of myoelectric prosthetic control keeps expanding and results in a low acceptance rate of multi-functional prostheses from the users [28]. The vast majority of prosthetic hand users tend to choose simply aesthetically pleasing or amplitude based ones in the lack of an ideal control. An ideal prosthetic control is of intuitiveness, robustness, low computational complexity, real-time performance and limited burden of re-training [58]. Other desired properties of the HMI include a minimal number of electrodes, an easy user training process, a closed-loop control with sensory feedback and long-term usability. Efforts towards the ideal properties have been extensively seen in recent years [131, 162, 163]. Yet to date, there are still quite many limitations to be addressed in current myoelectric prosthetic control as follows.
Fig. 2.10 Motion Test environment and tasks [6]
Failure in Deep Muscle Activity Sensing
Myoelectric signals can be acquired in either an invasive or noninvasive way, which leads to the iEMG and sEMG. The iEMG is capable of providing an insight into deep muscular structure and changes of targeted compartments. However, it suffers from the invasiveness while the noninvasiveness remains one of the most desired properties for prosthetic hand users. On the other hand, the sEMG can only detect activities mostly from superficial muscles and hardly capture the electrical manifestation of deep muscle contractions while some dexterous finger movements are naturally related to deep muscle contraction [76]. As a result, the failure in deep muscle activity sensing impedes the dexterous hand motion recognition especially in a long-term use, where the weak physiological signals from deep compartments become even more unstable for sEMG based detection.
Intrinsic Randomness and Sensitivity to Physiological Changes
Irrespective of the high recognition accuracy achieved in an offline scheme and short-term testing scenarios in a laboratory environment, the performance of hand motion recognition deteriorates severely for long-term clinical use [162]. The premise of accurate recognition of hand motions is crafted by the consistency of sEMG patterns. However the randomness of myoelectric signals in long-term use will result in the large variation of signal manifestation,
which hinders the accurate motion recognition in a clinical cross-day environment [29]. Typically, the inherent variability of myoelectric manifestation and inevitable day-to-day physiological changes contribute to a heavy burden of re-calibration and re-training [33]. Besides, physiological changes like fatigue and sweat remain some of the most severe factors that adversely impact the hand motion recognition accuracy. Muscle fatigue generally occurs after long-term muscle activity and results in a decline of the muscular ability to generate the desired force through contraction.
Electrode Shift and Crosstalk
Isotonic contraction of muscles tends to cause relative movement between targeted muscles and skin surface where the electrode is attached and negatively impacts the predefiend detection site distribution. The electrode shift has been pointed out as the most significant dynamic factor that adversely influences the performance of myoelectric pattern recognition based control [164]. Hargrove et al. [114] enriched the training dataset with samples captured from potential displacement locations to remedy the performance degradation caused by electrode shift. But the enlarged pooled samples are commonly captured at the price of a heavy burden of training which is undesirable for users. Intuitively, iEMG is less affected by muscular crosstalk than sEMG, which allows a more independent detection of targeted muscles. But it has been validated that iEMG does not outperform sEMG in pattern recognition based myoelectric control [53]. What’s more, crosstalk has been identified as an influential factor on the performance of recognition [103], which can be mitigated by pattern recognition based control [53] yet whose performance heavily depends on the robustness of the adopted classification strategy.
Lack of Sensory Feedback
An ideal prosthetic control process is expected to be closed-loop with proper sensory feedback modules. However, a practical feedback is missing in most cases of prosthetic hand control and has been highlighted as the drawback of existing myoelectric control systems since last decade [20, 28]. Though functional electrical stimulation (FES) is a module that provides sensory feedback and has been utilised in various motor function rehabilitation targeted applications, the inherent properties of FES inevitably lead to electrical interference in combination with traditional myoelectric control. To date, limited clinically viable feedback approaches have been proposed for myoelectric prosthetic control. Thus a non-electrical manifestation is desired to support the feedback instead of using solely visual feedback.
Burden of User Training
Controversy remains whether adequate user training is desired in clinical applications to date. On the one hand, repeated re-training process is time-consuming and forms a heavy burden on the users. On the other hand, the user training intuitively improves the consistency of sEMG patterns exerted by users [165]. Inconsistencies between the academic and clinical applications partly reside in the user training phase. In a laboratory environment the subjects are normally seated in a comfortable position without surrounding disturbances while in clinical scenarios the amputated users are involved in interactions with multiple objects under various abrupt changes. Powell et al. [166] introduced a clinical protocol to train the users with pattern recognition related concept and a prosthesis-guided training process for classifier re-calibration. Despite the advantages of user training in improving the recognition performance, the training burden remains to be further reduced from the users’ perspective [33].