A Prototype of MyoWare (Electromyography Muscle Sensor) for Measuring People’s Muscle Strengths

: Human-Computer Interaction (HCI) becomes a solution to help humans connect with computers. Research and tools related to HCI have been developed by many researchers. HCI is able to help humans connect between humans and computers and humans with humans at a considerable distance. One of HCI model is applied to the MyoWare tool that can capture hand muscle movements using an electromyograph (EMG) sensor. This article describes how to assemble and identify the raw data generated from the MyoWare tool. Using MyoWare on the hand could produce EMG data output. MyoWare only used the EMG sensor and generated data in the form of Envelope EMG and Raw EMG which differed in scale and size. This required a extraction features process to make the data uniform. This study uses the Moment Invariant method to extract features and min-max to normalize each data generated on the MyoWare sensor. Testing was done by doing simple hand movements. The test results showed that the differences in gestures were recognized well even though they were performed in different positions


Introduction
The development of technology in the field of Human-Computer Interaction (HCI) has a tremendous impact, especially in the field of hand gesture recognition. Nowadays, research on hand gesture recognition such as virtual reality, robotics, computer games (animated), remote control, the introduction of sign language, and more has been done in various fields. The development of these technologies enables people to interact and build good communication between humans and machines [1].
In addition, the development of computerized hand gesture recognition is an important requirement in the industrial era 4.0, especially during the COVID-19 pandemic. The pandemic has made people more concerned about health. Hence, many researchers have conducted several studies and approaches to health surveillance using hand gesture recognition technology. At this time, hand gesture recognition technology is very closely related to daily activities. It is a useful solution to knowing and monitoring the state of human bodies.
There are various developments of HCI technology for hand gesture recognition, such as the use of smartwatches to determine the heart rate in the blood vessels on the wrist [2], The use of Myo Armband aims to assist deaf communication using a sign language [3], and hand and arm strength recognition tools are developed to monitor the condition of users and physiotherapy patients [1]. The various types of approaches in hand gesture recognition technology aim to help humans to connect and control their health through HCI. HCI technology in the health sector have been researched, especially in post-stroke patients who seek for treatment or therapy [4]. Previous research found that muscle tone in post-stroke patients increased and caused muscle spasticity. Muscle spasticity causes stiffness, pain, and difficulty moving, and it will affect the normal movement of poststroke patients [5]. In addition, increased muscle tone is a sign to not optimal therapy in post-stroke patients.
Based on the issues, a prototype device was assembled in the form of an electromyograph (EMG) sensor to help treat post-stroke patients [6]. The novelty of this research is how we can measure and predict the movement muscle on human arm to help stroke patients. The prototype of the device is used to improve the quality of spasticity treatment by finding the pattern of the EMG sensor performance [4]. However, at this stage, this study aimed to describe the tool development process and measure muscle strengths in a normal human arm. This study also aimed to create a tool that can be used in future research on post-stroke patients.

Materials and Methods
This paper aimed to find out the performance of the MyoWare raft tool and the EMG sensor data output . It will also inform the basis for further research on a technology for detecting muscle tone with easy use by therapists and patients' families. In this section, a device design of MyoWare muscle sensor kit and its evaluation were explained along with how it affects hand movements.

Related Work
A. F. Ruiz-olaya [7] predicted that gesticulation can be detected using pattern recognition through the biological signal from the EMG sensor. This study compares two experiment results of hand gesture recognition algorithm based on the EMG sensor. That algorithm used information-generated kinetics, and EMG data were better for classifications of five hand movements rather than the use of EMG data.
M. Sathiyanarayanan [1] demonstrated the use of the MYO armband for physiotherapeutic treatment. This study analyzed the EMG sensor on MYO diagnostics to detect abnormalities in hand movements. The test was conducted on 24 medical students using the SUS questionnaire. This study was an alternative way to support the interactive physiotherapy analysis and help better understand the entire myocardial system with early diagnosis. Muscle activation via electric potential or electromyography (EMG) has traditionally been measured for diagnosis of neuromuscular disorders in medical research [8]. However, with the advent of ever shrinking yet more powerful microcontrollers and integrated circuits, EMG circuits and sensors have contbuted to prosthetics, robotics, and other control systems [9][10][11] [12].

Hand Gesture Experiments
These experiments used some hand gestures following dynamic movements that received different treatment. This step aimed to identify patterns, the influence of the movement against the existing sensors on the MyoWare and feature extraction. Dynamic movement is a movement of the right and left (radial-deviation), up and down (flexionextension), and pronation-supination. The movement was used to measure speed, direction, and position contained in MyoWare as shown in Figure 3.

Proposed Method
The measurement results of hand movement showed nonuniformity sensor data both in scale and size. When nonuniform data were used directly for the recognition process, a lot of errors happened. Errors were caused by ambiguous movement, sensor failure, short measurement, or early measurement failures [13] The data showed inconsistent sizes and scales. Signal data were not easy to translate, and the extraction process was required to be a feature vector.  Figure 4 shows the chart patterns of muscle contraction obtained from the EMG sensor. The data were raw data that required treatment prior extraction into feature vectors. In mathematical terms, the raw data are shown in the formula (1). The sensor data produced were not consistent with the scale, thus requiring the process of normalization. This study used the Min-Max method for data normalization.

EMG sensor Formula
Where: = maximum value of = minimum value of This method can exert the balance between the values of the comparison data. No data were dominant over others. Data normalization was performed because the data obtained from EMG generated 2-dimensional patterns with different data values [14].

Moment Invariant Method
Extraction feature in this study was invariant moment, a scalar value that captures a significant feature. This value provided object's characteristics that uniquely represent the shape. Using the moment invariant method, this study aimed to describe the object with invariant, insensitive to a certain deformation. As a result, it could provide enough power to recognize objects from different classes [15]. The next process was to form feature vectors by calculating the mean, median, standard deviation, and skewness of each axis in the sensor data.
The next step was to measure whether the feature vectors already represented a hand gesture or not. The calculation of the formulas (8) and (9)  −1 < < +1 Where: = differential of data ( ) = feature vector  Figure 5 shows the feature vector of each sensor. This study found differences in the patterns of the movements. The feature vectors could identify different movements that could be robust to their positions.  Figure 6 shows that the data produced different patterns. The difference in movement affects the data patterns.

Conclusions
Based on the method proposed and variations in hand motion, MyoWare recognized movements whose positions and directions were the same, except movements (robust). This technology also recognized the same hand movements in different directions. The use of the moment invariant method as a feature extraction was still powerful enough for the technology to recognize robust hand movements. In the next development, MyoWare and the moment invariant method can be used to help treat post-stroke patients through hand movement recognition.
Institutional Review Board Statement: Not applicable.

Informed Consent Statement: Not applicable.
Data Availability Statement: Data is unavailable due to privacy or ethical restrictions.