(B-72) SMART GLOVE THROW DETECTION ALGORITHM FOR BASEBALL AND SOFTBALL PITCHERS

Introduction:: Baseball and softball pitchers are at high risk for pitching arm injuries. Overuse, which may be quantified by throwing workload metrics (e.g., number of pitches, frequency of pitching), may be the predominant injury mechanism [1,2]. Guidelines [1] have been developed to reduce these injuries and require counting official in-game pitches, but they neglect to consider other hard throws that occur in games or practices that likely contribute to overuse injuries [3].

Recent studies have used an inertial measurement unit (IMU) sensor on the pitching arm to measure throwing workload of baseball players [4-6]. While those studies were encouraging, poor compliance of usage was reported [5-6]. Here, a “Smart Glove” method was considered, where the IMU was attached with tape to the pitcher’s glove, with the long-term goal of increasing compliance during routine use by instrumenting a piece of gear that is already used nearly every time a player throws a ball.

The hypothesis was that a general throw detection algorithm could be developed to analyze participant-specific IMU acceleration data from a Smart Glove to accurately count pitches. The aims were to 1) conduct experiments with an IMU attached to each pitcher’s glove in a motion analysis lab, 2) use manual data analysis to identify key participant-specific parameters from the IMU data, 3) develop a general throw detection algorithm with those participant-specific input parameters, and 4) assess if pitch count from the throw detection algorithm fell within a target range of 96% - 104% of the true manual pitch count.

Materials and Methods:: Five baseball pitchers 10-16 years old and four softball pitchers 16-20 years old participated. Acceleration data were collected by an IMU (Mbientlab, USA) taped to the pitcher’s glove as participants threw pitches from an artificial mound (baseball) or flat ground (softball) into a net with a simulated strike zone 25 feet away. Using the Mbientlab iPhone app, the IMU was programmed to record up to 16 g’s of acceleration at 200 Hz. Manual data analysis revealed that the Smart Glove detected a repeatable, but participant-specific, acceleration pulse for baseball and softball pitches. Each pitch consistently generated two peaks in acceleration; the first peak occurred when the glove hand separated from the trunk and the second peak occurred when the glove hand returned to the trunk (Fig. 1). Typically, a pitch corresponded to at least two, and sometimes as many as four, peaks in acceleration occurring between 0.1 and 0.5 seconds of each other. This trend was consistent for all baseball and softball pitchers. Thus, the algorithm was developed using three key, participant-specific parameters: the minimum height a peak must exceed and the minimum and maximum pulse widths from the first peak to the second peak (Fig. 1). After analyzing the data with a custom MATLAB (MathWorks, USA) code of the algorithm, precision was calculated as true positives / (true positives + false positives), sensitivity was calculated as true positives / (true positives + false negatives), and a t-test assessed if calculated pitch counts differed between sports (p< 0.05 significance).

Results, Conclusions, and Discussions:: Over the course of this study, 105 combined pitches were thrown by the participants. The algorithm correctly counted 103 total pitches which was 98.1% of the true pitch count and within the target range of 96% to 104%. Of the 105 pitches thrown, there were two false negatives and five false positives, resulting in a total precision of 95.4% and a total sensitivity of 98.1%. Also, the algorithm correctly counted 98.6% and 97.1% of the true pitch counts for baseball and softball participants, respectively. There was no statistical significance between those results to suggest that the sport of the participant resulted in a higher or lower precision or sensitivity (p = 0.285 and p = 0.673, respectively).

The results supported the hypothesis that the Smart Glove and throw detection algorithm could count pitches within a target range of 96%-104% of the pitch count. Recent studies have shown that when embedding an IMU in an arm sleeve that is worn on the pitcher’s throwing arm, algorithms could count pitchers with 98-100% of the true pitch count [4]. Thus, the results obtained in this study using a smart glove are comparable to other studies that used an IMU on the pitching arm to track pitches [5].

There were several limitations in this study. First, the study was performed in a lab setting which may have impacted pitching biomechanics and did not account for other motions (including throws from other positions) that may occur in a game or practice setting. Second, the algorithm only used the IMU acceleration data and neglected the angular velocity data – use of angular velocity data might be needed to distinguish pitches from other types of throws in future studies. Third, it was assumed that a single algorithm could be used to quantify throwing workload for both baseball and softball participants. To address those limitations, future studies should test the Smart Glove in game and practice settings while using the angular velocity data to improve throw counts for both baseball and softball pitchers.

Acknowledgements (Optional): : Funding was provided by the Donald E. Bently Center for Engineering Innovation (SK).

References (Optional): : [1] MLB Advanced Media, https://www.mlb.com/pitch-smart/risk-factors, 2022. [2] Kerut et al, J La State Med Soc. 160(2): 95-98, 2008. [3] Fleisig et al. Sports Health. 4(5):419-424, 2012. [4] Schweiger et al, Southeast Conf IEEE, 2020. [5] Wahl et al, J Pediatr Orthop. 40(7): 609-615, 2020. [6] Rose et al, J Biomech Engng. 143: 03452 1-7, 2021.