**6. Future directions**

The widespread use of automatic tools for decomposing EMG signals is dependent on how easy is to detect MUAPs from the surface of the skin, on the identification of new applications, either online or offline, which might employ the results of the EMG decomposition, and also on the sharing and availability of developed tools to clinicians and researchers.

Ideally the sensors used for detecting MUAPs from the surface of the skin should be easy of applying and using. As a number of applications require the use of multiple sensors organized on an array the solution to this issue becomes more complex. Therefore, further research on this area, with the aim of developing sensors that allow for the reproduction of experiments is required.

Most applications found in the area of EMG decomposition are solely focused on the development of the automatic tool for decomposing the EMG signal, or in the classification and discrimination of MUAPs. Therefore, the identification of new useful applications are required in order to disseminate the relevance of the technique to other areas. For instance, the use of motor unit information could be employed in robotics, myofeedback, and human-machine interface development.

A major limitation of the results in the area of EMG decomposition is that the developed tools are not shared by researchers. The sharing of these tools together with EMG signal databases could be beneficial to the widespread use of the tools. Furthermore, this would allow researchers to objectively compare distinct solutions.

Considering the application of artefact removal from biomedical signals, the development of algorithms for reducing the influence of noise over signals in real time, without a priori knowledge about the origins and characteristics of the noise, are required. In this way the application of adaptive techniques such as Empirical Mode Decomposition should be further exploited.

It is also expected that the use of these filters can be part of the solution of the problem of EMG decomposition by directly decomposing the raw EMG signal into its motor unit action potential trains. If such filters are developed, EMG decomposition tools could be embedded in hardware speeding up the results from multichannel data.
