Is Artificial Intelligence Algorithms What Unlocks the Power of Inertial Measurement Units in Gross Motor Development Assessment?
Oral Presentation
Paper ID : 2147-SSRC
Authors
1Assistant Professor of Gorgan University of Agricultural Sciences & Natural Resources, Gorgan, Iran
2گروه طب ورزشی پژوهشگاه علوم ورزشی
3Faculty of New Sciences and Technologies, University of Tehran, Tehran, Iran
Abstract
Background: Researchers have informed the power of inertial sensors as a reliable fundamental movement skill (FMS) monitoring technology, however, regular uptake of this technology has not become common practice. Although there is an agreement between different evaluators in the overall score of FMS, there is disagreement in some sub-scales, such as throwing .The purpose of this study is to investigate the power of artificial intelligence algorithms to determine the accuracy of wearable inertial sensors against the scores of an expert examiner in the test of gross motor skills development (TGMD).
Method: 123 overhand throwing were performed by children aged 4 to 10 years (age = 7±1.84) (53% = boys). Three IMU sent signals of angular velocity, linear acceleration of the preferred hand, non-predominant ankle, and lumbar region of the children. Each performance was scored according to the criteria of the TGMD-3 by reviewing the video of the performed skills. The "k nearest neighbor" (KNN) algorithm was used for automatic data classification. The classification accuracy was 85%. In comparing the accuracy of criteria 2 and 3, the criteria related to leg movement was 93% accurate, but the accuracy of the distinguishing trunk and hip rotation was only 78%. After reviewing the videos, it was found that in some trials, the leg action was artificially "correct". In these performances, putting the opposite leg forward undergoes a separate process of hip and trunk rotation; as a result, it did not play an influential role in increasing the force of the projectile.
Conclusions: The nobility of this study is the use of automatic classification algorithms for evaluation. One of the advantages of this method was a more detailed analysis of the execution process without extracting temporal phases and kinematic outputs of complex signals. Automatic scoring has gone one step further than TGMD evaluation. It has identified the lack of use of power transfer due to upper body rotation to the projectile. These results show that the quantitative approach allows for a more detailed analysis of the overhand throwing process by highlighting differences that cannot be detected by traditional on-field assessment.
Method: 123 overhand throwing were performed by children aged 4 to 10 years (age = 7±1.84) (53% = boys). Three IMU sent signals of angular velocity, linear acceleration of the preferred hand, non-predominant ankle, and lumbar region of the children. Each performance was scored according to the criteria of the TGMD-3 by reviewing the video of the performed skills. The "k nearest neighbor" (KNN) algorithm was used for automatic data classification. The classification accuracy was 85%. In comparing the accuracy of criteria 2 and 3, the criteria related to leg movement was 93% accurate, but the accuracy of the distinguishing trunk and hip rotation was only 78%. After reviewing the videos, it was found that in some trials, the leg action was artificially "correct". In these performances, putting the opposite leg forward undergoes a separate process of hip and trunk rotation; as a result, it did not play an influential role in increasing the force of the projectile.
Conclusions: The nobility of this study is the use of automatic classification algorithms for evaluation. One of the advantages of this method was a more detailed analysis of the execution process without extracting temporal phases and kinematic outputs of complex signals. Automatic scoring has gone one step further than TGMD evaluation. It has identified the lack of use of power transfer due to upper body rotation to the projectile. These results show that the quantitative approach allows for a more detailed analysis of the overhand throwing process by highlighting differences that cannot be detected by traditional on-field assessment.
Keywords