Forearm sEMG data from young healthy humans during the execution of hand movements – Scientific Data

Participants

The participants were selected according to the inclusion protocol SIP-20221503 approved by the research committee and regulated by the research and postgraduate secretariat (Secretaría de Investigación y Posgrado del Instituto Politécnico Nacional). The inclusion criteria set healthy participants without any neuromusculoskeletal, cardiovascular, pulmonary, or neurological diseases. In addition, the participants can be subjects of both sexes, males or females, between 18 and 37 years old, from any race, religion, and ethnic orientation. For this study, the data from 28 participants within the ages stipulated in the protocol was considered to create the database, Table 1 shows a summary of the participants in this study.

Instrumentation

The acquisition stage implements the WyoFlex armband¹⁵. This device is a wearable sEMG system for remote biosignals acquisitions. Figure 1 describes the complete instrumentation to obtain the sEMG signals from the participants. In this case, the considered electronic instrumentation has three main stages, the acquisition stage with the sEMG sensors; the signal processing and amplifying stage; and the transmission stage. In the first stage, each sEMG acquisition system (WyoFlex armband) has four Gravity Analog sEMG sensors. Each sensor has two modules; a dry electrode board comprises the first module. The electrodes have a differential input, high common mode rejection ratio, low power consumption, and single power supply. The second module has electronic elements for amplifying and transmitting the obtained signals. Table 2 summarizes the characteristics of the transmitter board. A total of eight sensors were employed to measure the data from the volunteers, four for each armband and one armband for each forearm.

Based on previous works regarding sEMG acquisition^{21,22,23,24,25}, in the second stage, to perform the measurements, the dry electrode board was placed around the circumference of the forearm. The sampled rate of each sEMG sensor was selected as 1000 Hz (considering that the dominant range of the sEMG is from 10 to 500 Hz⁴), with a 12-bit analog-digital converter (ADC) implemented in the FireBeetle ESP32 microcontroller. The microcontroller supports two power supply methods: USB and 3.7 V external lithium battery. Then, four ADC channels converted the analog information to its digital counterpart to be transmitted using an User Datagram Protocol (UDP) to a personal computer, where a graphical user interface (GUI) based on the Node-RED tool performed the data visualization and storage²⁶. For more details about the electronic instrumentation of the WyoFlex device the readers are referred to the work developed by Manuela Gomez-Correa and David Cruz-Ortiz¹⁵.

Experimental protocol

In order to carry out the signal acquisition, this work considered the following four relevant aspects to design the experimental protocol, which defines how the signals should be acquired.

1. 1.

   Selected movements

2. 2.

   Acquired metadata

3. 3.

   Number of participants

4. 4.

   Homogeneity of the signals

First of all, it was determined the four basic hand movements: flexion, extension, ulnar deviation, and radial deviation. These gestures are executed mainly by the superficial muscles of the forearm, which are the main contributors to the signals acquired by the sEMG sensors that make up the WyoFlex armband, as shown in Fig. 1. Furthermore, six additional movements were selected according to the types of grip defined by Schlesinger in the study of hand dexterity for the upper limb taxonomy²⁰. Therefore, the chosen six main grasps were the power grip, precision grip, hook grip, lateral grip, spherical grip, and pinch grip.

Once the movements were defined, it was determined to perform not only the recording of sEMG signals in the subjects but also the acquisition of information such as gender, age, physical activity of the people, and anthropometric measurements of the upper limb. With this information, it is possible to carry out a classification analysis, for example, evaluating the possible implications of the armband placement distance, such as differences in the signals according to gender and the implications of the physical activity conditions in the relevant characteristics of this type of signals such as the amplitude or the frequency.

Likewise, different databases were evaluated for the choice of the number of participants. According to the literature, in most published datasets, the acquisition of sEMG signals contemplates from 5 to 18 users. Because of this, and in order to create a broader database for different analyses, it was determined that taking 28 participants was an appropriate number for the established acquisition protocol considering that each executes six trials and three cycles for each forearm.

The last aspect considered was the homogeneity of the signals. Then, a tutorial was designed to show step-by-step how the test should be executed. The tutorial shows a participant performing the same test in such a way that the user can follow it simultaneously. With the above, it was possible to generate a homogeneous acquisition, considering that this method allows better control of the speed and execution time of each hand gesture. Also, studies like the one developed by María V. Arteaga were contemplated for the protocol¹⁸. Their protocol considered six seconds for the execution of each movement, three seconds to make the gesture, and three seconds for rest. Additionally, to acquire multiple trials of the same test subject, it was obtained that the signals for both forearms were simultaneous, performing three cycles per person. With this information, it could be possible to perform laterality and fatigue analysis.

Based on the previous facts, the proposed experimental protocol is divided into five main stages. The first one is the participant selection. Then, the second stage is the survey of personal data and the sign of informed consent. The third step considers the location of the WyoFlex device. The execution test protocol by the participant and signal acquisition integrate the stage four. Finally, the last stage describes the data curation. To sum up all the steps in the experimental protocol, Fig. 2 provides an overview of the proposed experimental protocol to obtain the dataset. A detailed description of each section is provided below.

General scheme of the experimental protocol stages.

Stage 1

The first stage of the experimental protocol is the participant selection. Here, it is corroborated that each female or male who wants to participate in the test is in the 18–37 age range. Also, it is verified that the participant is a healthy person without any neuromusculoskeletal problem or cardiovascular, pulmonary, or neurological disease.

Stage 2

In order to collect personal data such as name, age, as well as anatomical measurements of the upper limb, we designed a survey. The survey was filled out after the informed consent was obtained from each participant. Table 3 presents the information acquired through the survey with the nomenclature used for each parameter. Then, at the beginning of the test protocol, anthropometric information was measured with a measuring tape. Some of the obtained results are summarized in Table 4. Then, two WyoFlex armbands (one per each forearm) were used for data recording of the sEMG signals.

Stage 3

In order to place the sEMG sensors, this study considered the recommendations provided in the guide entitled European recommendations for sEMG (SENIAM)²⁷. Therefore, each WyoFlex device was placed in the middle of the forearm based on the recommendations obtained from a literature review^{21,22,23,24,25}. However, it should be emphasized that even when the middle of the forearm was the suggested location, if the forearm circumference of the participant was smaller than the diameter of the WyoFlex armband, the device had to be displaced to an upper part of the forearm in order to guarantee the correct contact between the electrodes and the participant skin.

Four sEMG sensors were placed on the intended muscles of each forearm. In this case, Sensor 1, labeled as S1 was located over the posterior part of the forearm, which corresponds to the Exterior digitorium muscle and Extensor carpi ulnaris muscle. Correspondingly, Sensor 2 (S2) was placed over the external side of the forearm, that is, over the muscles Palmaris longus and Flexor carpi ulnaris. Then, Sensor 3 (S3) was placed over the Brachioradialis muscle and Flexor carpi radialis muscles. Finally, the last Sensor (S4) was located in the position corresponding to the Extensor carpi radialis longus and the Extensor carpi radialis brevis muscles. To sum up the sensor location of the WyoFlex device, Fig. 3 evidences the location of each sensor.

Location of the sEMG sensors in the forearm.

Stage 4

Once the four electrodes were placed in the correct position, it was verified that each electrode had enough contact with the participant’s skin. For that, it was corroborated that the baseline of the sensors was 1.5 V through the GUI. Then, all the participants were instructed to perform ten different hand movements: flexion, extension, ulnar deviation, radial deviation, hook grip, power grip, spherical grip, precision grip, lateral grip, and pinch grip to obtain its corresponding sEMG of each of the four sensors (S1, S2, S3, and S4). As an example, Fig. 4 shows the ten mentioned movements aside from an example of the corresponding sEMG obtained signals. Then, aiming that the participants execute the correct form of the required movements, a video tutorial showing the movements that the participants should perform during the experimental protocol was created.

Movements executed in the test protocol with its corresponding sEMG signals.

The video tutorial lasts eight minutes and 35 seconds distributed equally on three cycles (see the video tutorial structure in Fig. 5). In this case, each test allows the acquisition of 24 sEMG signals per participant (four sEMG signals for each band during three cycles). With the participant seated, preserving 90 degrees between all lower limb joints, the tutorial starts with an introduction section, where a general explanation of the test is provided to the participant (first 50 seconds of the tutorial). Then, the participant starts with the first cycle, which consumes around 200 seconds. Here, it should be emphasized that the participant has 15 seconds to execute each of the ten movements (see the movements section in Fig. 5).

Video tutorial structure.

The 15 seconds are distributed as in the movement composition section of Fig. 5. In the first five seconds, the movement that the patient should execute is shown in the video tutorial (movement indications in Fig. 5). Then, the following three seconds are considered to execute the movement. After that, the participant must maintain the position for three seconds. Finally, the participant has four seconds to rest and continue to the next action. Notice that, at the end of each cycle, the participant has five extra seconds to rest after continuing with the second and third cycles. At this point, all the information data vectors generated during the execution of the test protocol are sent to the GUI.

Notice that the participants should execute the test comfortably and look ahead to the monitor where the tutorial was played. If the participants report fatigue during the test or execute the movements in an incorrect form, the trial should be rejected.

Stage 5

The data of each signal is sent through a message (from the microcontroller to the GUI) containing a character A to identify the start of the vector information and four characters representing the ADC value of each sEMG sensor. Thus, each data vector is integrated with 17 characters (one for character A, and four for each sensor). Equation (1) describes the vector information structure.

where S_{j, ki} represents ADC data from the j-th sensor with j={1, 2, 3, 4}, the variable k = {I, D}, refers to the WyoFlex located in the left and right forearm, respectively; and the subscript i = {1, 2, 3,…, n} denotes the i-th number of sample in the recorded sEMG signals. All the data vectors are received in the GUI designed in the Node-RED environment. The acquired signals consider an offset in their amplitude (a digital value of 1862, according to the sensor manufacturer). This means that the signals can vary in a digital value range from 0 to 4095 due to the ADC module resolution. Notice that even when the sensor manufacturer recommends considering a digital value of offset 1862 (or approximate 1.5 V), the authors corroborate in experimental tests that an offset value of 1756 should be selected to generate sEMG signals with a baseline on zero. In this particular case, the proposed offset corresponds to a mean value of all the acquired data vectors.

Dataset elaboration

As the final step on the GUI, the user obtains a data vector, which contains the sEMG signals of each of the eight sensors as can be observed in the first part of the scheme given in Fig. 6. This data vector is stored by the GUI in a comma-separated values (CSV) file. Then, in order to obtain the dataset, a data segmentation algorithm is implemented to separate the data of each sensor and store it in vectors for the sEMG signals offline visualization.

The scheme shows a graphical representation of the steps in the data segmentation algorithm. In the scheme’s first column, the data vector is provided as input for the segmentation algorithm. Then, the four mentioned stages are forearm segmentation, cycle segmentation, movement segmentation, and vector homogenization.

Data segmentation algorithm

A segmentation algorithm is implemented in Python to obtain homogeneous data vectors of each sEMG signal corresponding to each hand movement. Here, the algorithm considers four stages: forearm segmentation, cycle segmentation, movement segmentation, and vector homogenization. Figure 6 shows a block diagram where each column corresponds to the four mentioned stages in the segmentation algorithm. The blue color in some blocks of Fig. 6 denotes some examples (shown below) of the signals obtained in that particular stage of the segmentation algorithm.

Stage 1: Forearm segmentation

The algorithm segmentation starts when the CSV file containing the eight sensors information separated by semicolons is loaded. Then, each data vector is divided into eight sub-vectors, that is, four for the left forearm and four for the right forearm (see Fig. 7a).

Example of the obtained signals in each of the first three stages of the segmentation algorithm considering the Data S3.

Stage 2: Cycle segmentation

The subsequent step is in charge of dividing the data vector into three cycles per sensor (see Fig. 7b). To this end, the segmentation algorithm considers the data information measured from the sensor S₁, which in this particular case is established as a reference for the segmentation. This sensor is selected due to the characteristic (maximum) amplitudes generated during the execution of the extension movement. Notice that, to divide the cycle it is also considered the time provided by the video tutorial in order to improve the information synchronization.

Stage 3: Movement segmentation

This step consists of obtaining the ten movements per cycle, that is, one vector for each movement (see Fig. 7c).

Stage 4: Vector homogenization

After the movement segmentation stage, the vector homogenization is executed. The main objective of this step is to generate six vectors containing 13000 data points (see Fig. 7d). To this end, the algorithm calculates the difference between the length of the motion vector and 13000 samples. Then, half of the difference at the beginning of the vector is deleted, and the other half of the difference is deleted at the end.

At this point of the data segmentation algorithm, the corresponding sEMG signal of each movement is obtained. To improve the readability of the effect of each of the described stages, Fig. 7 has been added. This figure shows the signals obtained after implementing the four stages of the segmentation algorithm following the sequence of the blue blocks in Fig. 6.

Once the segmentation algorithm is finished, the sEMG of each motion is stored also in a CSV file with specific labels according to the movement to which they correspond. Here, it should be emphasize that two types of CSV files are obtained as output of the algorithm segmentation. The first files contain the sEMG signals in digital value, whereas the second ones contain the signals in voltage value. Notice that, in both cases, the dataset considers signals with and without amplitude offset. In the section given below a detailed explanation about the information in the dataset is provided.