Markerless vision-based functional movement screening movements evaluation with deep neural networks.

Area of Science:

• Biomechanics
• Computer Science
• Sports Medicine

Background:

• The Functional Movement Screen (FMS) is a valuable tool for assessing fundamental movement patterns and predicting sports injuries.
• Current FMS assessment relies heavily on clinical expertise, limiting its accessibility and objectivity.
• There is a need for automated and reliable methods for FMS evaluation.

Purpose of the Study:

• To develop and validate an automatic framework for Functional Movement Screen (FMS) assessment using a multi-view deep neural network (MVDNN).
• To enhance the objectivity and accessibility of FMS testing through automated analysis.
• To improve the accuracy of injury risk prediction based on movement patterns.

Main Methods:

• An automatic FMS movement assessment framework was developed utilizing a multi-view deep neural network (MVDNN).
• The framework incorporated automatic skeleton extraction and manual feature selection to capture 3D joint trajectory data from two viewpoints.
• Time-series modeling techniques were employed to learn from skeleton sequences, with fusion of multi-view motion features.

Main Results:

• The MVDNN framework demonstrated superior performance compared to existing state-of-the-art methods on a public FMS dataset.
• Achieved high performance metrics, including an average miF1 score of 0.857, maF1 score of 0.768, and Kappa score of 0.640 over ten runs.
• The fusion of two-view motion features provided complementary information, enhancing assessment accuracy.

Conclusions:

• The developed MVDNN framework offers an effective and automated solution for FMS movement assessment.
• This automated approach has the potential to increase the objectivity and reliability of FMS testing.
• The findings suggest a promising direction for improving sports injury prediction through advanced computational methods.