Cardio-Respiratory Monitoring in Archery Using a Smart Textile Based on Flexible Fiber Bragg Grating Sensors

In precision sports, the control of breathing and heart rate is crucial to help the body to remain stable in the shooting position. To improve stability, archers try to adopt similar breathing patterns and to have a low heartbeat during each shot. We proposed an easy-to-use and unobtrusive smart textile (ST) which is able to detect chest wall excursions due to breathing and heart beating. The sensing part is based on two FBGs housed into a soft polymer matrix to optimize the adherence to the chest wall and the system robustness. The ST was assessed on volunteers to figure out its performance in the estimation of respiratory frequency (fR) and heart rate (HR). Then, the system was tested on two archers during four shooting sessions. This is the first study to monitor cardio-respiratory activity on archers during shooting. The good performance of the ST is supported by the low mean absolute percentage error for fR and HR estimation (≤1.97% and ≤5.74%, respectively), calculated with respect to reference signals (flow sensor for fR, photopletismography sensor for HR). Moreover, results showed the capability of the ST to estimate fR and HR during different phases of shooting action. The promising results motivate future investigations to speculate about the influence of fR and HR on archers’ performance.


Introduction
Archery is a precision sport which requires consistency and stability of movements [1]. A mismatch of physical, physiological, and psychophysical factors can influence athletic performance and deteriorate archers' accuracy and precision [2][3][4][5].
In precision sports, both breathing and heart rate (HR) influence the athlete's performance [5][6][7]. The control of such physiological activities facilitates the performance of repetitive shots in the same, stable posture [8]. Lakie et al. demonstrated that high values of HR can cause sway movements, tremor, and shaking of the body when aiming at the target [8]. Mohammed et al. showed that HR

Principle of Work of the Custom Smart Textile
Two flexible sensors were used to develop a ST consisting of two elastic bands (600 mm × 40 mm × 2.1 mm, 10 kgf of maximum load and 100% of polyamide) worn around the thorax and the abdomen, respectively. The fit of each band was adjusted by VELCRO ® fastener to put the two FBGs in contact to the chest in correspondence of the xiphoid process and the umbilicus. The use of anatomical landmarks allows the FBGs to be worn on the same measurement points. Each sensor consists of an FBG (Bragg wavelengths λ B of 1541 nm and 1545 nm for the band around the thorax and the abdomen, grating length of 10 mm and reflectivity of 90%; At Grating Technologies), previously housed in a flexible polymer packaging (90 mm × 24 mm × 1 mm) made of Dragon Skin ® 20 (Smooth-On, Inc. USA) as shown in Figure 1. A detailed description of the manufacturing process and of the sensors' metrological properties are reported in [19].
FBGs work as stop band filters of wavelength because they back-reflect a small portion of light traveling along the fiber at a specific wavelength (i.e., λ B ). The FBG working principle is well described by the following equation: where η eff, is the effective refractive index of the fiber core and Λ, the grating period. The dependence of η eff and Λ from temperature and strain makes FBGs an optimal solution for the development of measurement systems able to sense these two parameters [23]. Regarding the application of interest, breathing and heart beating cause periodic displacements of the chest and, in turn, stretch the flexible sensors embedded in the ST, as schematically reported in Figure 1.

Population and Experimental Protocol
The ST for cardio-respiratory monitoring was assessed on nine healthy volunteers (four males and five females) whose age and anthropometric characteristics are shown in Table 1. Each participant was asked to perform a protocol consisting of three main phases: i) a short apnea useful to synchronize the reference instruments (i.e., flowmeter for fR, photopletismography sensor -PPG-for HR) and the FBG outputs; ii) 16 quiet breaths; iii) a final apnea as long as each volunteer can. The study was approved by the local ethical committee (protocol number 27/18).

Experimental Set-Up
The vital signs under investigation were monitored by the ST. The flexible sensors were positioned corresponding to the xiphoid process and the umbilicus (see Figure 2). Each FBG embedded into ST (FBG T for the band around the thorax and FBG A for the band around the abdomen, a box in Figure 2) was connected to an optical spectrum interrogator (si425, Micro Optics Inc., b box) which worked at the sampling frequency of 250 Hz. The reference signal for the respiratory activity was collected by a commercial flow sensor (SpiroQuant P, EnviteC, Alter Holzhafen, Wismar, Germany, c box in Figure 2) connected to a differential pressure sensor (163PC01D75, Honeywell, Minneapolis, MN, USA). The output of reference system used for the respiratory monitoring was collected by using a DAQ (NI USB-6009, National Instrument, Rockville, MD, USA, d box in Figure  2) and a custom virtual instrument developed in LabVIEW® environment, at the sampling frequency of 250 Hz. The reference system for the cardiac monitoring was a photopletismography sensor (e box Regarding the application of interest, breathing and heart beating cause periodic displacements of the chest and, in turn, stretch the flexible sensors embedded in the ST, as schematically reported in Figure 1.

Population and Experimental Protocol
The ST for cardio-respiratory monitoring was assessed on nine healthy volunteers (four males and five females) whose age and anthropometric characteristics are shown in Table 1. Each participant was asked to perform a protocol consisting of three main phases: i) a short apnea useful to synchronize the reference instruments (i.e., flowmeter for f R , photopletismography sensor -PPG-for HR) and the FBG outputs; ii) 16 quiet breaths; iii) a final apnea as long as each volunteer can. The study was approved by the local ethical committee (protocol number 27/18).

Experimental Set-Up
The vital signs under investigation were monitored by the ST. The flexible sensors were positioned corresponding to the xiphoid process and the umbilicus (see Figure 2). Each FBG embedded into ST (FBG T for the band around the thorax and FBG A for the band around the abdomen, a box in Figure 2) was connected to an optical spectrum interrogator (si425, Micro Optics Inc., b box) which worked at the sampling frequency of 250 Hz. The reference signal for the respiratory activity was collected by a commercial flow sensor (SpiroQuant P, EnviteC, Alter Holzhafen, Wismar, Germany, c box in Figure 2) connected to a differential pressure sensor (163PC01D75, Honeywell, Minneapolis, MN, USA). The output of reference system used for the respiratory monitoring was collected by using a DAQ (NI USB-6009, National Instrument, Rockville, MD, USA, d box in Figure 2) and a custom virtual instrument developed in LabVIEW®environment, at the sampling frequency of 250 Hz. The reference system for the cardiac monitoring was a photopletismography sensor (e box in Figure 2), placed on the index fingertip of the left hand, as in [24]. The PPG sensor was input into two other analogue ports of the same DAQ used for the respiratory monitoring and collected at the sampling frequency of 250 Hz. in Figure 2), placed on the index fingertip of the left hand, as in [24]. The PPG sensor was input into two other analogue ports of the same DAQ used for the respiratory monitoring and collected at the sampling frequency of 250 Hz.

Data Analysis and Results
This paragraph will be grouped into two subsections according to the stages of the protocol (i.e., quiet breathing and apnea). Each subsection describes the data analysis performed to estimate fR (during quiet breathing) and HR (during apnea), and the obtained results.

Respiratory Frequency Estimation During Quiet Breathing
For each volunteer, data were processed following four main steps: i) the outputs of the FBGs and the flow sensor were synchronized by selecting the first minimum points (i.e., starting points) after the starting apnea (see Figure 3); ii) the quiet breathing stage was selected by cutting all the synchronized signals from the mentioned starting points and considering all the 16 breaths performed during the protocol (see Figure 3); iii) a filtering stage consisting of a second order passband filter (lower cut-off frequency of 0.05 Hz and higher cut-off frequency of 0.5 Hz) was applied on both FBGs and flow sensor outputs; iv) a custom algorithm was used to select the maximum peaks of each signal (see Figure 4).
The respiratory periods (i.e., TR A and TR T for the band around the abdomen and the thorax, respectively, and TR FLOW for the flow sensor) were calculated as the time interval between two consecutive maximum peaks. Then, the fR values estimated by both the FBGs (i.e., fR A and fR T ) and the flow sensor (i.e., fR FLOW ) were calculated as the ratio between 60 and the related respiratory periods in order to express fR in acts per minute (i.e., apm).

Data Analysis and Results
This paragraph will be grouped into two subsections according to the stages of the protocol (i.e., quiet breathing and apnea). Each subsection describes the data analysis performed to estimate f R (during quiet breathing) and HR (during apnea), and the obtained results.

Respiratory Frequency Estimation During Quiet Breathing
For each volunteer, data were processed following four main steps: i) the outputs of the FBGs and the flow sensor were synchronized by selecting the first minimum points (i.e., starting points) after the starting apnea (see Figure 3); ii) the quiet breathing stage was selected by cutting all the synchronized signals from the mentioned starting points and considering all the 16 breaths performed during the protocol (see Figure 3); iii) a filtering stage consisting of a second order pass-band filter (lower cut-off frequency of 0.05 Hz and higher cut-off frequency of 0.5 Hz) was applied on both FBGs and flow sensor outputs; iv) a custom algorithm was used to select the maximum peaks of each signal (see Figure 4).
The respiratory periods (i.e., T R A and T R T for the band around the abdomen and the thorax, respectively, and T R FLOW for the flow sensor) were calculated as the time interval between two consecutive maximum peaks. Then, the f R values estimated by both the FBGs (i.e., f R A and f R T ) and the flow sensor (i.e., f R FLOW ) were calculated as the ratio between 60 and the related respiratory periods in order to express f R in acts per minute (i.e., apm).   The ST was assessed in terms of both breath-by-breath and mean fR values. The Bland-Altman analysis was performed to describe the agreement between fR values estimated by the proposed system and the reference one [25]. This analysis considered all the fR values of the enrolled volunteers (i.e., a total of 135 fR values for each Bland-Altman analyses) for the calculation of the mean of difference (MOD) and the limits of agreement (LOA + = MOD + 1.96·SD and LOA -= MOD − 1.96·SD, where SD is the standard deviation of the differences between the data collected by the proposed  The ST was assessed in terms of both breath-by-breath and mean fR values. The Bland-Altman analysis was performed to describe the agreement between fR values estimated by the proposed system and the reference one [25]. This analysis considered all the fR values of the enrolled volunteers (i.e., a total of 135 fR values for each Bland-Altman analyses) for the calculation of the mean of difference (MOD) and the limits of agreement (LOA + = MOD + 1.96·SD and LOA -= MOD − 1.96·SD, where SD is the standard deviation of the differences between the data collected by the proposed The ST was assessed in terms of both breath-by-breath and mean f R values. The Bland-Altman analysis was performed to describe the agreement between f R values estimated by the proposed system and the reference one [25]. This analysis considered all the f R values of the enrolled volunteers (i.e., a total of 135 f R values for each Bland-Altman analyses) for the calculation of the mean of difference (MOD) and the limits of agreement (LOA + = MOD + 1.96·SD and LOA − = MOD − 1.96·SD, where SD is the standard deviation of the differences between the data collected by the proposed system and the reference one). The mentioned analysis was performed considering data provided by each flexible sensor. The mean absolute percentage error in the f R estimation (i.e., MAPE fR ) was used to compare mean f R values and was calculated as: where f R smart_textile and f R re f erence denote the values of the f R obtained by the proposed system and the reference one, respectively. Results of the breath-by-breath analysis are shown in Figure 5 and in Table 2. The good agreement between the f R values measured by the proposed system and the reference one is confirmed by the high value of the correlation coefficient (R 2 ) for both FBG T and FBG A and by the low value of both MOD and MAPE fR . system and the reference one). The mentioned analysis was performed considering data provided by each flexible sensor. The mean absolute percentage error in the fR estimation (i.e., MAPE fR ) was used to compare mean fR values and was calculated as: where textile smart R f _ and reference R f denote the values of the fR obtained by the proposed system and the reference one, respectively.
Results of the breath-by-breath analysis are shown in Figure 5 and in Table 2. The good agreement between the fR values measured by the proposed system and the reference one is confirmed by the high value of the correlation coefficient (R 2 ) for both FBG T and FBG A and by the low value of both MOD and MAPE fR .    The HR estimation during the apnea was performed according to the following four main phases: i) for each trial, the first minimum points after the holding of breath were selected on the FBGs output to define the starting point of the apnea stage for the FBGs and the PPG sensors; ii) the same time interval (i.e., 10 s) was chosen to estimate HR of all the volunteers during the apnea (see Figure 6); iii) a fourth-order Butterworth pass-band filter with a lower cut-off frequency of 0.6 Hz and a higher cut-off frequency of 20 Hz was applied on the signals. This band of frequency was chosen according to the frequency components of vibrations induced on the chest wall by the blood flow ejection into the vascular bed [26]; iv) a custom algorithm was used to select minimum peaks (blue markers in Figure 7) on each filtered FBG signal. The beat-by-beat cardiac period (T C ) from the FBG outputs was calculated considering the minimum peaks, as the time elapsed between two consecutive minimum peaks (T C T and T C A from the band around the thorax and the abdomen, respectively). Minimum peaks were chosen because they are easier to detect on the filtered FBG signals, automatically. Beat-by-beat cardiac periods were calculated considering the time interval between two consecutive maximum peaks on the filtered PPG signal (T C PPG ). The HR values (i.e., HR A , HR T , and HR PPG ) were calculated as the ratio between 60 and T C T , T C A , and T C PPG , in this way HR is expressed in bpm. The HR estimation during the apnea was performed according to the following four main phases: i) for each trial, the first minimum points after the holding of breath were selected on the FBGs output to define the starting point of the apnea stage for the FBGs and the PPG sensors; ii) the same time interval (i.e., 10 s) was chosen to estimate HR of all the volunteers during the apnea (see Figure 6); iii) a fourth-order Butterworth pass-band filter with a lower cut-off frequency of 0.6 Hz and a higher cut-off frequency of 20 Hz was applied on the signals. This band of frequency was chosen according to the frequency components of vibrations induced on the chest wall by the blood flow ejection into the vascular bed [26]; iv) a custom algorithm was used to select minimum peaks (blue markers in Figure 7) on each filtered FBG signal. The beat-by-beat cardiac period (TC) from the FBG outputs was calculated considering the minimum peaks, as the time elapsed between two consecutive minimum peaks (TC T and TC A from the band around the thorax and the abdomen, respectively). Minimum peaks were chosen because they are easier to detect on the filtered FBG signals, automatically. Beat-by-beat cardiac periods were calculated considering the time interval between two consecutive maximum peaks on the filtered PPG signal (TC PPG ). The HR values (i.e., HR A , HR T , and HR PPG ) were calculated as the ratio between 60 and TC T , TC A , and TC PPG , in this way HR is expressed in bpm.  The ST capability of monitoring HR was assessed in terms of beat-by-beat and mean HR values. The Bland-Altman analysis was performed by comparing HR values estimated by the proposed system and the reference one, considering all the volunteers for a total number of 149 beats. The mean absolute percentage error (i.e., MAPE HR ) was also calculated as follows: where HR smart_textile and HR re f erence , values estimated by the smart textile and the PPG sensors, respectively. Results of the beat-by-beat analysis are shown in Figure 8 and agreement between the f R values measured by the proposed system and the reference one is confirmed by the high value of the R 2 for both FBG T and FBG A and by the low value of both MOD and MAPE HR .  The ST capability of monitoring HR was assessed in terms of beat-by-beat and mean HR values. The Bland-Altman analysis was performed by comparing HR values estimated by the proposed system and the reference one, considering all the volunteers for a total number of 149 beats. The mean absolute percentage error (i.e., MAPE HR ) was also calculated as follows: , values estimated by the smart textile and the PPG sensors, respectively. Results of the beat-by-beat analysis are shown in Figure 8 and Table 3. The good agreement between the fR values measured by the proposed system and the reference one is confirmed by the high value of the R 2 for both FBG T and FBG A and by the low value of both MOD and MAPE HR .   Archers were invited to perform two shooting sessions. Each session consists of six arrows to be shot in five minutes at designed 70 m targets. The first session is a practice round. In this round the arrows are shot at the beginning and do not count as part of the score. Instead, the second one is a scoring round with an awardable maximum score of 60 points (10 points per arrow).

Experimental Set-Up
During each session of shooting, archers worn the flexible sensors on the same positions investigated during tests on volunteers (see Figure 9). The optical spectrum interrogator (si425, Micro Optics Inc. Hackettstown, NJ, USA) for the acquisition of the FBGs output was placed at 3 m of distance from the archer. The sampling frequency was 250 Hz. Since the assessment of the ST was performed on volunteers (Section 3), no reference instruments were used during the shooting action to not impair the archer movements.

Data Analysis and Results
Changes of FBG T and FBG A output of the two practice shooting sessions are plotted in Figure 10.

Data Analysis and Results
Changes of FBG T and FBG A output of the two practice shooting sessions are plotted in Figure 10. The analysis was performed by selecting FBG T output because the respiratory acts and the shooting phases are clearly discernible while FBG A output has a more irregular trend (see Figure 10). FBG T output changes were analyzed following three main phases: i) the six shooting actions were selected, as shown in Figure 10; ii) for each shooting action, the signal related to the aiming phase was filtered and its minimum peaks were detected to calculate the HR values ( Figure 11); iii) the signal related to the breathing activity which precedes the shooting actions was filtered and its maximum peaks were detected to estimate the fR values. Results in terms of fR and HR for all the four trials are summarized in Table 5. The analysis was performed by selecting FBG T output because the respiratory acts and the shooting phases are clearly discernible while FBG A output has a more irregular trend (see Figure 10). FBG T output changes were analyzed following three main phases: i) the six shooting actions were selected, as shown in Figure 10; ii) for each shooting action, the signal related to the aiming phase was filtered and its minimum peaks were detected to calculate the HR values ( Figure 11); iii) the signal related to the breathing activity which precedes the shooting actions was filtered and its maximum peaks were detected to estimate the f R values. Results in terms of f R and HR for all the four trials are summarized in Table 5. Sensors 2019, 19, x FOR PEER REVIEW 11 of 14 Figure 11. Output of the sensor changes during both breathing activity and shooting action.

Discussion and Conclusion
This work is focused on cardio-respiratory monitoring in archery using a custom ST based on flexible FBGs.
The feasibility of the system was assessed on nine volunteers during both quiet breathing and apnea. The system showed promising results in terms of both fR and HR estimation, as shown in Tables 2 and 3. The position of FBGs does not influence the system performance in the fR estimation, while FBG T allows a more accurate HR monitoring than FBG A (see Table 2 and 3).
In a previous work [19], we already characterized the proposed sensing element. The flexible FBG sensors showed sensitivity to strain of 0.125 nm⋅mε −1 , sensitivity to temperature changes of 0.012 nm⋅°C −1 and negligible influence of relative humidity on its response. We were the first group that used Dragon skin® 20 as polymer matrix to improve the sensor robustness and skin adherence for the cardio-respiratory monitoring. The main novelties of this work were the assessment of the custom ST on volunteers for the monitoring of both respiratory and cardiac activities and the ST application on archers during shooting sessions.
In the literature, FBGs encapsulated into flexible materials for cardio-respiratory monitoring were mainly proposed for clinical applications (e.g., during magnetic resonance exams). These

Discussion and Conclusions
This work is focused on cardio-respiratory monitoring in archery using a custom ST based on flexible FBGs.
The feasibility of the system was assessed on nine volunteers during both quiet breathing and apnea. The system showed promising results in terms of both f R and HR estimation, as shown in Tables 2 and 3. The position of FBGs does not influence the system performance in the f R estimation, while FBG T allows a more accurate HR monitoring than FBG A (see Tables 2 and 3).
In a previous work [19], we already characterized the proposed sensing element. The flexible FBG sensors showed sensitivity to strain of 0.125 nm·mε −1 , sensitivity to temperature changes of 0.012 nm· • C −1 and negligible influence of relative humidity on its response. We were the first group that used Dragon skin®20 as polymer matrix to improve the sensor robustness and skin adherence for the cardio-respiratory monitoring. The main novelties of this work were the assessment of the custom ST on volunteers for the monitoring of both respiratory and cardiac activities and the ST application on archers during shooting sessions.
In the literature, FBGs encapsulated into flexible materials for cardio-respiratory monitoring were mainly proposed for clinical applications (e.g., during magnetic resonance exams). These technological solutions consist of FBG sensors encapsulated into polydimethylsiloxane, PDMS [27], polyvinyl chloride, PVC [28], and fiberglass [21]. In [27], an FBG was encapsulated into PDMS matrix (dimensions: 85 mm × 85 mm × 5 mm). Tests were carried out by positioning the sensing element on the back of two males and two females during MR examination. Results showed maximum relative errors of 4.41% and 5.86% for respiratory and cardiac monitoring, respectively. In [28], PVC was used as flexible matrix to house an FBG for simultaneous respiratory and cardiac monitoring. System showed good performances in the estimation of f R and HR. In [21], an FBG was housed into fiberglass matrix (dimensions: 30 mm × 10 mm × 0.8 mm). The accuracy of the proposed sensor was characterized by relative error <4.64% for f R and <4.87% for HR. In the present study, tests performed on volunteers showed promising results in both the estimation of f R and HR (i.e., MAPE fR ≤ 1.97% and MAPE HR ≤ 5.74%).
Focusing on the shooting sessions, results showed that the proposed ST can detect all the six shooting actions and, in turn, monitor f R during the breathing activity and HR during the aiming phase. In the literature, commercially available devices able to monitor only one of these parameters (f R and HR) were used on archers. Breathing activity and patterns were studied in [9] by using Zephyr Bio-Harness devices (Model PSM Research version 1.5, single transmitter and receiver, Annapolis, MD, USA). In [2,5,7] HR values were monitored by using Polar FT4, three silver-silver chloride chest electrodes, and Fitbit Charge HR (Fitbit, Inc. Boston, MA, USA), respectively.
These studies showed that both f R and HR are important since they influence archers' performance. Therefore, the main improvement of the proposed system is the possibility to monitor both f R and HR with high accuracy during the different phases of the shooting action using the same sensing element. Since the sensing element can be connected to the optical interrogator by means of long, flexible and lightweight fiber optic, the proposed system does not impair the shooting action. This feature encourages further assessment of the proposed system in sports science applications (e.g., during walking running on treadmill and during cycling).
In future works, a high number of archers will be enrolled to investigate how f R and HR influence the shooting performance, and how HR can be estimated in the presence of breathing. These findings will aid the optimization of training strategies according to the experience of each archer and the maximization of their shooting action and scores.