Golf Swing

**BACKGROUND** The golf swing can significantly contribute to lower spine degeneration and consequent lower spine pain due to the repetitive and intense forces it exerts on the lumbar spine1. Lower back pain makes up for about a quarter of all golf injuries and affects players of all ages and with varying skill level.2 Different styles of golf swings exist but, in general, they are characterized by extensive rotational movements and powerful downward movements that can place particularly high stress on the spine (primarily on the intervertebral disc and the facet joints). These complex loading patterns consist of a combination of compression, torsion, and shearing and are placed on the spine at a high frequency (professional golfers may perform hundreds of swings a day). Characteristic damage to the spine caused by repeated minor traumatic injuries might be the result3. In addition to targeted strengthening of muscles around the lumbar spine, understanding and consequently potentially modifying the swing biomechanics may help mitigate the risk of spine degeneration in golfers. The aim of this project is to determine the player-specific loading of the lumbar spine through a detailed biomechanical analysis of the player’s swing. Similar analyses have been successfully conducted for baseball but are yet missing in golf. **DESCRIPTION** Balgrist Sports Medicine is setting up a high-fidelity golf motion simulator equipped with advanced motion capture and biomechanical analysis capabilities. In addition to this, a cost-effective alternative using a mobile phone is planned for development. This master’s project aims to assess the feasibility and accuracy of a mobile phone-based motion capture system (Part A) and its potential for biomechanical analysis (Part B). _PART A: Pose estimation_ The first step o involves establishing a baseline by evaluating the most promising state-of-the-art methods for estimating 3D poses from monocular images, such as MMpose [5]. The performance of these methods will be preliminary and qualitatively compared on public golf swing datasets [6[ without ground truth data. Following this, a new pilot dataset of golf swings will be collected using the Balgrist golf simulator. Each swing will be recorded simultaneously with a monocular mobile phone camera and the simulator's multi-camera system. The multi-camera setup will serve as the ground truth, enabling an initial quantitative evaluation of the accuracy of monocular pose estimation. _PART B: Biomechanical analysis_ In part B, subject-specific information will be obtained from images and will be used to individualize an existing OpenSim model (e.g. fit the height of the model to the subject height, replicate the spinal column alignment of the subject,…)4. The pose estimation of a subject’s golf swing (PART A) will then be incorporated into the model (i.e. the swing kinematics are imposed). For model generation and analysis, the OpenSim application programming interface (API) with MATLAB will be used4. The golf swing will be simulated, leading to predictive values of joint reaction forces within the lower spine and muscle activities during task performance. **TASKS** _PART A: Pose estimation_ • Familiarization with the relevant existing literature • Implementing 2-3 state-of-the-art methods for 3D pose estimation • Collecting a preliminary dataset with ground-truth • Performing an accuracy evalution on this dataset _PART B: Biomechanical analysis_ • Familiarization with the relevant existing literature • Incorporation of subject-specific aspects (demographics, anatomy, alignment) into an existing OpenSim model based on radiological and/or rgb images. Streamline the procedure • Incorporation of the output of the pose estimation into the corresponding individualized OpenSim model. Streamline the procedure • Based on musculoskeletal simulations, analyze the loading pattern on the lower lumbar spine during a golf swing for the available cohort • How can/should the outcome be validated? How can one conclude if a certain loading pattern on the lower spine is detrimental on the long run? **REQUIREMENTS** _PART A: Pose estimation_ • Hands-on experience in training and implementing deep learning computer vision networks • Previous experience in pose estimation is an asset _PART B: Biomechanical analysis_ • Ability to work independently • Good MATLAB skills • Previous experience with musculoskeletal modeling is an asset **REFERENCES** 1. Bourgain M, Rouch P, Rouillon O, et al. Golf Swing Biomechanics: A Systematic Review and Methodological Recommendations for Kinematics. Sports 2022;10:91. 2. M. Lindsay D, A. Vandervoort A. Golf-Related Low Back Pain: A Review of Causative Factors and Prevention Strategies. Asian J Sports Med;5. Epub ahead of print November 1, 2014. DOI: 10.5812/asjsm.24289. 3. Walker CT, Uribe JS, Porter RW. Golf: a contact sport. Repetitive traumatic discopathy may be the driver of early lumbar degeneration in modern-era golfers. Journal of Neurosurgery: Spine 2019;31:914–7. 4. Schmid S, Connolly L, Moschini G, et al. Skin marker-based subject-specific spinal alignment modeling: A feasibility study. Journal of Biomechanics 2022;137:111102. 5. https://github.com/open-mmlab/mmpose/tree/main/projects/rtmpose3d 6. https://github.com/wmcnally/golfdb?tab=readme-ov-file

**BACKGROUND**

The golf swing can significantly contribute to lower spine degeneration and consequent lower spine pain due to the repetitive and intense forces it exerts on the lumbar spine1. Lower back pain makes up for about a quarter of all golf injuries and affects players of all ages and with varying skill level.2
Different styles of golf swings exist but, in general, they are characterized by extensive rotational movements and powerful downward movements that can place particularly high stress on the spine (primarily on the intervertebral disc and the facet joints). These complex loading patterns consist of a combination of compression, torsion, and shearing and are placed on the spine at a high frequency (professional golfers may perform hundreds of swings a day). Characteristic damage to the spine caused by repeated minor traumatic injuries might be the result3.
In addition to targeted strengthening of muscles around the lumbar spine, understanding and consequently potentially modifying the swing biomechanics may help mitigate the risk of spine degeneration in golfers. The aim of this project is to determine the player-specific loading of the lumbar spine through a detailed biomechanical analysis of the player’s swing. Similar analyses have been successfully conducted for baseball but are yet missing in golf.

**DESCRIPTION**

Balgrist Sports Medicine is setting up a high-fidelity golf motion simulator equipped with advanced motion capture and biomechanical analysis capabilities. In addition to this, a cost-effective alternative using a mobile phone is planned for development. This master’s project aims to assess the feasibility and accuracy of a mobile phone-based motion capture system (Part A) and its potential for biomechanical analysis (Part B).

_PART A: Pose estimation_
The first step o involves establishing a baseline by evaluating the most promising state-of-the-art methods for estimating 3D poses from monocular images, such as MMpose [5]. The performance of these methods will be preliminary and qualitatively compared on public golf swing datasets [6[ without ground truth data. Following this, a new pilot dataset of golf swings will be collected using the Balgrist golf simulator. Each swing will be recorded simultaneously with a monocular mobile phone camera and the simulator's multi-camera system. The multi-camera setup will serve as the ground truth, enabling an initial quantitative evaluation of the accuracy of monocular pose estimation.

_PART B: Biomechanical analysis_
In part B, subject-specific information will be obtained from images and will be used to individualize an existing OpenSim model (e.g. fit the height of the model to the subject height, replicate the spinal column alignment of the subject,…)4. The pose estimation of a subject’s golf swing (PART A) will then be incorporated into the model (i.e. the swing kinematics are imposed). For model generation and analysis, the OpenSim application programming interface (API) with MATLAB will be used4.
The golf swing will be simulated, leading to predictive values of joint reaction forces within the lower spine and muscle activities during task performance.

**TASKS**

_PART A: Pose estimation_
• Familiarization with the relevant existing literature
• Implementing 2-3 state-of-the-art methods for 3D pose estimation
• Collecting a preliminary dataset with ground-truth
• Performing an accuracy evalution on this dataset

_PART B: Biomechanical analysis_
• Familiarization with the relevant existing literature
• Incorporation of subject-specific aspects (demographics, anatomy, alignment) into an existing OpenSim model based on radiological and/or rgb images. Streamline the procedure
• Incorporation of the output of the pose estimation into the corresponding individualized OpenSim model. Streamline the procedure
• Based on musculoskeletal simulations, analyze the loading pattern on the lower lumbar spine during a golf swing for the available cohort
• How can/should the outcome be validated? How can one conclude if a certain loading pattern on the lower spine is detrimental on the long run?

**REQUIREMENTS**

_PART A: Pose estimation_
• Hands-on experience in training and implementing deep learning computer vision networks
• Previous experience in pose estimation is an asset

_PART B: Biomechanical analysis_
• Ability to work independently
• Good MATLAB skills
• Previous experience with musculoskeletal modeling is an asset


**REFERENCES**

1. Bourgain M, Rouch P, Rouillon O, et al. Golf Swing Biomechanics: A Systematic Review and Methodological Recommendations for Kinematics. Sports 2022;10:91.
2. M. Lindsay D, A. Vandervoort A. Golf-Related Low Back Pain: A Review of Causative Factors and Prevention Strategies. Asian J Sports Med;5. Epub ahead of print November 1, 2014. DOI: 10.5812/asjsm.24289.
3. Walker CT, Uribe JS, Porter RW. Golf: a contact sport. Repetitive traumatic discopathy may be the driver of early lumbar degeneration in modern-era golfers. Journal of Neurosurgery: Spine 2019;31:914–7.
4. Schmid S, Connolly L, Moschini G, et al. Skin marker-based subject-specific spinal alignment modeling: A feasibility study. Journal of Biomechanics 2022;137:111102.
5. https://github.com/open-mmlab/mmpose/tree/main/projects/rtmpose3d
6. https://github.com/wmcnally/golfdb?tab=readme-ov-file

_PART A: Pose estimation_ The goal of part A, is to evaluate the performance of state-of-the-art 3D pose estimation approaches for from monocular videos of golf swings, captured with a mobile phone. _PART B: Biomechanical analysis_ The aim of PART B is to predict joint loading within the lower spine during a golf swing for a cohort of golf players. Eventually, the tools developed in PART A and PART B shall allow the implementation of the analysis as a smartphone application allowing to record a golf swing, perform real-time pose estimation (PART A), and determine the spinal load during the movement (PART B). With correct guidelines, it might become possible to predict potentially detrimental biomechanical impact on the lower back with the potential to suggest a technique change or a reduction of training intensity (with respect to swing repetitions).

_PART A: Pose estimation_
The goal of part A, is to evaluate the performance of state-of-the-art 3D pose estimation approaches for from monocular videos of golf swings, captured with a mobile phone.

_PART B: Biomechanical analysis_
The aim of PART B is to predict joint loading within the lower spine during a golf swing for a cohort of golf players.

Eventually, the tools developed in PART A and PART B shall allow the implementation of the analysis as a smartphone application allowing to record a golf swing, perform real-time pose estimation (PART A), and determine the spinal load during the movement (PART B). With correct guidelines, it might become possible to predict potentially detrimental biomechanical impact on the lower back with the potential to suggest a technique change or a reduction of training intensity (with respect to swing repetitions).

Prof. Dr. Philipp Fürnstahl: philipp.fuernstahl@balgrist.ch PD Dr. Jonas Widmer: jonas.widmer@balgrist.ch

Prof. Dr. Philipp Fürnstahl: philipp.fuernstahl@balgrist.ch

PD Dr. Jonas Widmer: jonas.widmer@balgrist.ch