Evaluation of the PhySens as a Wrist-Worn Wearable in Pitch Detection and Biomechanical Workload Estimation

Abstract

Background:

The volume and frequency of throwing activity are among the most significant risk factors for developing overuse injuries in youth athletes. Despite introducing systematic guidelines for ‘pitch counts,’ throwing injuries continue to rise. Using technology to create enhanced measures of workload exposure in this unique population of athletes may help generate more effective and personalized injury prevention strategies.

Hypothesis:

The wrist-worn sensor system (PhySens) will: 1) accurately detect and differentiate throwing activity from other baseball movements, and 2) accurately predict ball velocity, arm slot angle, and elbow valgus torque.

Study Design:

Descriptive laboratory study.

Level of Evidence:

Level 5.

Methods:

Youth pitchers (n = 10) performed a standardized protocol of pitching, field-throwing, and batting. Pitching velocity and biomechanical data were simultaneously captured by the PhySens and traditional 3-dimensional motion capture. The accuracy of the pitching detection algorithm (throw vs batting) was analyzed by comparing truth data with throwing events cataloged by the device. Ball velocity, elbow valgus torque, and arm slot angle predictions were assessed with Pearson correlation coefficients and Bland-Altman plots.

Results:

A total of 230 events (pitches and bat swings) were analyzed. Pitch detection was excellent, with a sensitivity of 99.4% and specificity 97.9%. Pearson correlations were significant and excellent across all predicted variables, with ball velocity r = 0.96, elbow valgus torque r = 0.95, and arm slot angle r = 0.87. The system demonstrated excellent estimations of ball velocity, elbow valgus torque, and arm slot angle.

Conclusion:

This novel single-sensor wrist worn device was highly accurate in detecting pitching events, predicting ball velocity, and estimating arm slot angle and elbow valgus torque.

Clinical Relevance:

Throwing volume is highly associated with overuse injuries in youth baseball players. Sensor-based measures of workload monitoring can address inherent limitations related to human error and underestimation of true throwing exposure.

Baseball is one of the most popular sports for youth athletes and participation rates are at an all-time high. ¹⁴ The uniqueness of the overhead throwing motion has been described as the most violent act the human body can produce. ¹⁹ Specifically, the rapid acceleration required to generate ball velocity will also create a high amount of repetitive stress to the tissues of the shoulder and elbow. Not surprisingly, shoulder and elbow injuries are common in overhead throwers, with up to 30% to 50% arm injuries occurring within a single season.^7,8,13 Despite efforts to reduce the overall exposure to these repetitive stresses, data indicate that this problem is only getting worse, with increased burden of overuse injuries in youth baseball athletes in the United States. ²³ Thus, prevention of these injuries is of significant importance for athletes, parents, and coaches.

The most widely studied and significant risk factor for these injuries, in the youth population, is the volume of throwing and frequency of pitching outings.^21,22 In an effort to reduce injuries, many youth baseball organizations have adopted Pitch Smart guidelines, instituting pitch count restrictions and rest day requirements for all pitchers. ¹⁵ Despite these efforts, the rate of youth throwing-related injuries continues to rise.^11,23 Several inherent weaknesses of using pitch counts as a primary means of workload monitoring may account for this. The Pitch Smart guidelines track workload by documenting ingame exertion only, whereas research has identified that a majority of the exposure for youth athletes occurs during practice and warm-ups, not during ingame scenarios.^9,26 Thus, simply tracking ingame exposure does not accurately capture the total throwing volume, underrepresenting true exposure statistics.

Additional considerations that are unique to the youth baseball environment may also account for the limited effectiveness of Pitch Smart guidelines. These are primarily human-related factors, such as difficulty with accurate record keeping, lack of parent, athlete, or coach education, willful disregard of rules relating to over-use, and ineffective communication between the various stakeholders. In today’s youth sport environment, it is common for kids to play on multiple teams, in different leagues, and with differing coaching staffs. Tracking exposure between multiple organizations is inherently difficult, with most of the responsibility being placed on the child or parent. In addition, the current model of using pitch count as a proxy for workload monitoring is overly simplistic and not equitable. Every athlete and every pitch are currently treated the same, whereas it is understood intuitively that not every throw is the same and not all players are the same. For example, a maximal velocity throw or one with suboptimal mechanics will likely result in more injurious forces than a throw with submaximal effort and proper form. Hence, simple statistics (such as pitch count) may not sufficiently account for the workload encountered by each individual athlete.

Technological advances in inertial measurements units (IMU) have made it possible to derive more precise onfield measurement of throw counts and biomechanical stress encountered by the throwing arm. Although these systems do exist, issues with validity, user interface, and form factor may limit their integration and utility in the youth baseball population. ² Thus, it is hypothesized that a more comprehensive, technologically advanced, and individualized approach to workload monitoring in youth baseball players would lead to more effective and sustainable injury reduction.

Therefore, the primary purpose of this study was to assess the accuracy of a wrist-worn sensor system designed to automatically detect, differentiate, and monitor throwing activities (throw counts) from other sports and daily movements. The secondary purpose was to evaluate the accuracy of the sensor system for predicting pitch speed, arm slot angle, and elbow valgus torque.

Methods

Participants

A total of 10 healthy, male, right-handed, youth baseball pitchers (mean age, 11.6 ± 0.7 years; mean height, 156.0 ± 11.5 cm; mean weight, 40 ± 5.6 kg) participated in this study. To be eligible, all players had to be between the ages of 10 and 14 years old, identify their primary position as pitcher, and be free of any current shoulder or elbow pain. All study procedures were reviewed and approved by the Institutional Review Board before initiating this study. Informed consent from parent or guardian and assent from the child was obtained before data collection.

Testing Procedure

Before performing the pitching protocol, each participant underwent a structured warm-up consisting of a series of dynamic stretches and active mobility for the upper and lower body, followed by a typical series of J-band exercises for muscular activation including shoulder external rotation and internal rotation. Participants then performed a series of warm-up throws with gradually increasing intensity over a period of 3 to 5 minutes, until they felt adequately prepared to pitch.

Participants were fitted for traditional high-speed camera 3-dimensional (3-D) motion capture with 40 infrared reflective markers that enabled tracking of the limbs, torso and head (Figure 1). Movement data was recorded with 8 Motion Analysis infrared cameras encompassing a capture volume of 26 feet length × 7 feet width × 8 feet height (7.9 m × 2.1 m × 2.4 m).

Figure 1.

Illustration of marker setup for 3-D motion capture. *R.Hand marker used only for right-hand-dominant players; L.hand marker used only for left-arm-dominant players. 3-D, 3-dimensional; ASIS, anterior-superior iliac spine; Epi, epicondyle; Inf, inferior; L., left; Lat, lateral; R., right; Rad, radius; Med, medial; Scap, scapula; Uln, ulna.

Participants were also fitted with the experimental PhySens system consisting of a single sensor that was affixed to the player’s throwing arm at the wrist, just proximal to the ulnar styloid process, utilizing a custom wristband and housing system (Figure 2). The PhySens sensor platform was developed by Innovative Design Labs - a research and development firm specializing in advanced inertial motion and navigation estimation systems. It is a versatile platform designed for use in applications such as sports monitoring, gait analysis, daily activity monitoring, and research data collection. As part of this effort, the sensing platform was updated with state-of-the art high dynamic range accelerometers and gyroscopes. The updated accelerometer and gyroscope are capable of running with dynamics ranges of ±2g to ±16g and ±125 deg/s to ±4000 deg/s, respectively, while sampling at 480 Hz. The sensor also features a High-G accelerometer with full scale range up to ±400g sampling at 1 KHz.

Figure 2.

Photograph of motion capture session with PhySens sensor attached just proximal to the ulnar styloid process of the throwing arm.

Each participant performed a standardized throwing protocol (20 throws per participant) from a portable indoor pitching mound using their individually preferred wind-up (stretch vs full wind-up). All pitches were thrown using a 142 g (5 oz) regulation baseball to a strike zone target located at a regulation distance of 46 ft (14.02 m). The pitching protocol consisted of 5 warm-up throws (self-selected speed), 5 fastballs at self-selected, 75% intensity, 5 fastballs at self-selected, 100% intensity. The portable pitching mound was then removed, and the players performed an additional 5 throws utilizing a crow-hop, meant to mimic more standard field-based throwing mechanics. After completing this throwing protocol, each player then performed 5 swings with a baseball bat within the motion capture area.

Data were captured simultaneously from each pitching, field-throwing, and batting event using the PhySens and traditional 3-D motion capture. Before beginning their wind-up, each player performed 5 repetitions of shoulder external and internal rotation with the arm elevated in approximately 90 degrees of abduction. This motion was performed to allow for temporal syncing of the PhySens and the high-speed camera 3-D motion capture data during analysis. The motion capture data were analyzed using Cortex software. Position data were filtered using a 12-Hz fourth-order Butterworth low-pass filter. Segment coordinate systems were defined following International Society of Biomechanics (ISB) recommendations. ²⁴ The PhySens sensor was controlled via a custom developed smartphone application. All data from the sensor were captured simultaneously with the 3-D motion system. Data from the sensor system were written to a micro-SD card and then uploaded for analysis. The biomechanical variables of interest were elbow valgus torque, and arm slot angle. The ball velocity of each throw was measured utilizing a Rapsodo Pitching Version 2.0 (Rapsodo LLC) baseball monitoring system.

The accuracy of biomechanical variable estimates was analyzed by comparing truth data from the motion capture system (throwing versus batting) with the throwing event cataloged by the device software. Pearson correlation coefficients and Bland-Altman plots were created to analyze the relationship and accuracy between PhySens prediction and Rapsodo for ball velocity, and motion capture for elbow valgus torque and arm slot angle metrics.

Results

A total of 10 players with a mean age of 11.6 years participated in this study. On average, participants began playing baseball at 4.9 ± 0.7 years old and began pitching at a mean age of 7.6 ± 1.07 years. All players reported no shoulder or elbow pain or injury at the time of data collection. In addition to fastball, participants reported the ability to pitch changeups (9/10 players), curveballs (7/10 players), and sliders (1/10 players) in games. The mean ball velocity for full effort fastball was 54.3 mph with a range from 42 to 67.7 mph.

A total of 230 of the 250 (92%) collected movements (pitches, bat swings) had valid sensor data and 3-D motion capture data. Of the failed collects, 19 were rendered unusable due to obscured or lost markers that affected the motion capture system. The other (1) failed collection was due to missing data from the sensor system. A variety of machine-learning techniques were explored to identify the most accurate system for movement discrimination (throw versus not a throw) and estimation of ball velocity, elbow valgus torque, and arm slot angle from the data collected by the PhySens system. The best performing algorithm for movement discrimination was the bagged decision tree, whereas Gaussian Process Regression performed best for all estimation factors. Performance estimates were generated with 5-fold cross validation for each statistic.

Throwing versus batting discrimination was excellent, with a sensitivity 99.4% and specificity 97.9% (throwing positive class). Pearson correlations were significant (P < .05) and excellent across all predicted variables with ball velocity r = 0.96, elbow valgus torque r = 0.95, and arm slot angle r = 0.87 (Figure 3). The system demonstrated excellent ability to estimate pitch speed (within 1.5 mph), arm slot angle (within 3.5 degrees), and elbow valgus torque (mean absolute error 7%) (Figure 4).

Figure 3.

Pearson correlation plots for PhySens prediction of (a) ball velocity, (b) elbow valgus torque, and (c) arm slot angle. BW-H, percentage of body weight multiplied by height; Mocap, motion capture.

Figure 4.

Bland-Altman plots demonstrating agreement between the PhySens and 3-D motion capture for prediction of (a) ball velocity, (b) elbow valgus torque, and (c) arm slot angle. Solid line represents the mean difference between measures. Dashed line indicates the upper and lower limits of the 95% CI. All measurements demonstrated good agreement. BW-H, percentage of bodyweight multiplied by height; CV, coefficient of variation; Mocap, motion capture; RPC, reproducibility coefficient.

Discussion

The sensor system investigated in this study demonstrated excellent accuracy in measuring several key metrics related to baseball workload, kinematics, and elbow stress. Increased incidence and severity of youth athlete throwing injuries highlights the need for a paradigm shift in injury prevention methodology in the youth baseball community. The current standard of workload monitoring, the Pitch Smart guidelines, were developed in 2014 in response to growing concern of the rising rate of arm injuries in youth baseball players. ¹⁵ These evidence-based guidelines outline age-specific pitch count limitations and rest requirements for youth pitchers. Although these guidelines have raised awareness of the importance of workload monitoring and some evidence supports that adherence may reduce arm injuries,^1,25 the continued rise of arm injuries in youth baseball players points to lack of efficacy. Several practical limitations such as a lack of stakeholder (parent, coach, player) education,^6,17 inability to enforce or monitor compliance, ¹⁰ and underreporting of pitching volumes, ²⁶ may be contributing to the lack of efficacy utilizing this methodology of injury mitigation. The pitch detection functionality of the PhySens system would allow for the automatic tracking of throwing exposures. The exposures could then be audited against existing Pitch Smart Guidelines, removing the limitations of education, enforcement, and willful underreporting, which currently complicate and limit the effectiveness in the current landscape of youth baseball.

In addition to improving current practices, the PhySens system holds the promise of improving the capacity of measurement and developing more precise methods of workload monitoring in this specialized population. Ingame pitching exposure does not account for a large proportion of throwing workload experienced by youth athletes. Zaremski et al ²⁶ demonstrated that >40% of pitches performed in the context of a game (warm-up, bullpen, etc) were not accounted for when using traditional pitch count monitoring in high school baseball players. In addition, Freehill et al ⁹ found youth baseball pitchers had high levels of throwing exposure when they were pitching or not pitching (ie, playing field positions). This evidence clearly supports the need to expand throwing exposure accounting beyond those currently utilized in the current system. In addition, younger, elementary or middle school aged athletes may experience high volumes of throwing activity outside of a traditional baseball game or practice environment, such as during school recess or free play activities. It is the authors’ experience that these unaccounted workloads contribute to a young baseball player’s throwing workload, and should be considered when calculating a player’s overall workload. Thus, although previous studies have utilized an IMU sensor, similar to the PhySens, strapped to the upper arm (midhumeral) to quantify throwing volumes,^9,18 we assert that further innovation in form-factor is necessary to allow these devices to be worn on a day-to-day basis, to truly capture upper extremity throwing workloads experienced by this highly active group of young athletes. Developing the PhySens device as a wrist-worn wearable, similar to a watch or commercially available activity tracker, does not require a compression sleeve or other strapping mechanism to be worn on the upper arm, allowing these young athletes to not only wear it during sporting events, but also during the normal course of daily activities. This will allow for a more accurate assessment of their everyday throwing activities. Although further research is necessary, the high level of accuracy (98.6%) of the pitch detection algorithm of the PhySens system offers promise of measurement fidelity and the ability to obtain a more intimate understanding of individual workload outside of a traditional baseball environment.

Ball velocity is one of the most important metrics in baseball. The ball speed prediction for the PhySens device demonstrated good accuracy being within ±1.5 mph of the reference value from the Rapsodo system. Although the use of radar guns has been associated with increased risk of throwing related injury, ¹⁶ routine knowledge of a young athletes ball velocity may be advantageous due to the established relationship between increased ball velocity and shoulder or elbow injury.^3,4,12 Ball velocity is associated closely with increased age and height and, thus, a natural increase in ball velocity is expected as a course of normal physiological development in youth athletes. ²⁰ Thus, this period of athletic improvement is also associated with a period of increased risk of injury. In addition, reduction in ball velocity may be indicative of player fatigue or injury. ⁵ Consistent longitudinal monitoring of velocity changes may allow for early identification of at-risk players, leading to more informed and personalized injury risk mitigation strategies. Monitoring workload and velocity may lead to more informed rehabilitation, as clinicians could utilize these data to accurately monitor athlete’s progressing through an interval throwing program.

Injury prevention efforts for youth baseball players may be improved upon by utilizing more precise measures of workload estimation. In the current Pitch Smart guidelines, every athlete and every pitch are treated the same, whereas intuitively it is understood that not every throw and not every player are the same. Although highly accurate in throwing detection, the PhySens sensor also demonstrated high levels of accuracy in the estimation of elbow valgus torque, arm slot angle, and ball velocity. Although further research will be necessary to determine what combination factors increase the risk for injury, the authors hypothesize that injury risk stratification can be improved upon by utilizing a combination of metrics encompassing volume, arm stress, technique, and velocity. The combination of these factors may offer a more individualized and robust approach to workload monitoring.

This study has major limitations. This was a small group of 10 homogeneous adolescent athletes and a very small number of movements were studied. A more robust validation study with players of varying skill level, including female athletes, more diverse age range, and left-handed players, is necessary to further establish validity of this system. The pitch discrimination index was referenced on distinguishing a throwing event from a bat swing. Although the pitch detection prediction algorithm performed with high accuracy for baseball-specific activities, future studies should include a larger array of discriminatory movements.

Conclusion

The novel single-sensor wrist-worn device was highly accurate in detecting pitching events, predicting ball velocity, and estimating arm slot angle and elbow valgus torque.