Electromyography (EMG), an electrical signal collected by electrodes during muscle contractions, represents the bioelectrical properties of skeletal muscles and demonstrates the physiological processes of muscle contraction. It has been widely used for evaluation of muscle function in the areas of biomechanics and kinesiology [1-2], muscle pathology , muscle fatigue , and prosthetic device control . The root mean square (rms) magnitude and median frequency are commonly used to describe the time-domain and frequency-domain information of the EMG signal, respectively .
EMG is a complex signal; it is the summation of individual motor unit (MU) action potential trains generated by irregular discharges of active MUs during muscle activation. It can be influenced by many factors, e.g., muscle cross talk  and interelectrode distance . During the past decade, many efforts have been directed at developing different algorithms to process EMG signals, including classification of EMG using artificial network , fuzzy logic , and pattern recognition (multichannel EMG ) and decomposition of EMG signals with the Bayesian method . However, in addition to the complexity of the required signal processing methods, use of EMG for noninvasively measuring deep muscles is difficult because the deep muscle EMG signal may be more attenuated and/or mixed with the superficial muscle EMG signal by the time it reaches the skin surface.
Researchers have been searching for alternative signals that can better assess muscle function, including mechanomyography (MMG), electroencephalography [13-14], myokinemetric (MK) signals , and magnetic resonance imaging . For example, MMG is the sound generated by a muscle during its contraction and is used as a measure of mechanical muscle changes during contraction . Recently, it has been widely analyzed along with EMG for different purposes [18-20], such as control of a prosthesis with 2 degrees of freedom . However, MMG can be affected by many factors, such as muscle temperature , skinfold thickness , and external mechanical noise . These factors, together with challenges in sensor attachment and low-frequency noise elimination, can affect the stability and reliability of the MMG signal, thus limiting its application in fatigue assessment and prosthesis control. The MK signal represents the dimensional changes of muscle while it bulges during contraction and is detected with a Hall sensor . Sensor attachment is also a challenge when collecting the MK signal.
Because it has the advantages of being stable, easy to use, nonionizing, and capable of recording activities from deep muscles without cross talk from adjacent muscles , ultrasonography has been used to detect muscle thickness changes [26-27], pennation angle [28-29], cross-sectional area [30-31], and muscle fascicle length [32-34] both in static and quasi-static conditions during the past decades. Since skeletal muscle architecture is closely correlated with its function , ultrasound parameters have been widely employed to characterize muscle activity [36-38], and the relationship between EMG and muscle architecture changes detected with ultrasound has been reported [39-40].
In a previous study, Zheng et al. used sonomyography (SMG) with B-mode ultrasound images to describe real-time muscle thickness changes during contraction . A system was developed to record and analyze ultrasound images, force, joint angle, and surface EMG simultaneously. The system has been successfully used for the analysis of muscle fatigue, and the investigators found that muscle thickness increased during the fatigue process . The correlation between EMG and SMG of muscles during isometric contraction has also been investigated .
Although the two-dimensional SMG signal from ultrasound images is capable of detecting continuous muscle thickness changes, A-mode ultrasound, with a more portable and compact transducer, should be a less expensive and more practical alternative for detecting muscle thickness changes during contraction. …