Black Box Prediction Methods in Sports Medicine Deserve a Red Card for Reckless Practice: A Change of Tactics is Needed to Advance Athlete Care

Bullock, Garrett S.; Hughes, Tom; Arundale, Amelia H; Ward, Patrick; Collins, Gary S.; Kluzek, Stefan

doi:10.1007/s40279-022-01655-6

Cited by 30 publications

(19 citation statements)

References 53 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The use of machine learning tools that utilise 2D video data inputs is on the rise in the discipline of sports biomechanics and clinical biomechanics. A clear statement of the limitations of a machine learning model is of distinct importance given that they can be used to infer an athlete’s training load from broadcast footage with non-valid estimations leading to potentially harmful downstream decision making [ 40 ]. Furthermore, the use of remotely obtained video footage to estimate personal information such as load imposes a risk to the privacy and autonomy of an athlete [ 18 , 41 ].…”

Section: Discussionmentioning

confidence: 99%

Estimating Ground Reaction Forces from Two-Dimensional Pose Data: A Biomechanics-Based Comparison of AlphaPose, BlazePose, and OpenPose

Mundt

Born

Goldacre

et al. 2022

Sensors

View full text Add to dashboard Cite

The adoption of computer vision pose estimation approaches, used to identify keypoint locations which are intended to reflect the necessary anatomical landmarks relied upon by biomechanists for musculoskeletal modelling, has gained increasing traction in recent years. This uptake has been further accelerated by keypoint use as inputs into machine learning models used to estimate biomechanical parameters such as ground reaction forces (GRFs) in the absence of instrumentation required for direct measurement. This study first aimed to investigate the keypoint detection rate of three open-source pose estimation models (AlphaPose, BlazePose, and OpenPose) across varying movements, camera views, and trial lengths. Second, this study aimed to assess the suitability and interchangeability of keypoints detected by each pose estimation model when used as inputs into machine learning models for the estimation of GRFs. The keypoint detection rate of BlazePose was distinctly lower than that of AlphaPose and OpenPose. All pose estimation models achieved a high keypoint detection rate at the centre of an image frame and a lower detection rate in the true sagittal plane camera field of view, compared with slightly anteriorly or posteriorly located quasi-sagittal plane camera views. The three-dimensional ground reaction force, instantaneous loading rate, and peak force for running could be estimated using the keypoints of all three pose estimation models. However, only AlphaPose and OpenPose keypoints could be used interchangeably with a machine learning model trained to estimate GRFs based on AlphaPose keypoints resulting in a high estimation accuracy when OpenPose keypoints were used as inputs and vice versa. The findings of this study highlight the need for further evaluation of computer vision-based pose estimation models for application in biomechanical human modelling, and the limitations of machine learning-based GRF estimation models that rely on 2D keypoints. This is of particular relevance given that machine learning models informing athlete monitoring guidelines are being developed for application related to athlete well-being.

show abstract

Section: Discussionmentioning

confidence: 99%

Estimating Ground Reaction Forces from Two-Dimensional Pose Data: A Biomechanics-Based Comparison of AlphaPose, BlazePose, and OpenPose

Mundt

Born

Goldacre

et al. 2022

Sensors

View full text Add to dashboard Cite

show abstract

“…In general, few research teams and practitioners have access to these types of data with a limited cohort of a sub-sample of a military population, thus we recommend the collaborations across institutions necessary to collect large cohorts of varying types of military populations. This is turn would allow for the validation of models between populations for practical prescription use case ( Bullock et al, 2022 ).…”

Section: Discussionmentioning

confidence: 99%

Predictive utility of commercial grade technologies for assessing musculoskeletal injury risk in US Marine Corps Officer candidates

Bird

Koltun

et al. 2023

Front. Physiol.

View full text Add to dashboard Cite

Recently, commercial grade technologies have provided black box algorithms potentially relating to musculoskeletal injury (MSKI) risk and functional movement deficits, in which may add value to a high-performance model. Thus, the purpose of this manuscript was to evaluate composite and component scores from commercial grade technologies associations to MSKI risk in Marine Officer Candidates. 689 candidates (Male candidates = 566, Female candidates = 123) performed counter movement jumps on SPARTA™ force plates and functional movements (squats, jumps, lunges) in DARI™ markerless motion capture at the start of Officer Candidates School (OCS). De-identified MSKI data was acquired from internal OCS reports for those who presented to the Physical Therapy department for MSKI treatment during the 10 weeks of training. Logistic regression analyses were conducted to validate the utility of the composite scores and supervised machine learning algorithms were deployed to create a population specific model on the normalized component variables in SPARTA™ and DARI™. Common MSKI risk factors (cMSKI) such as older age, slower run times, and females were associated with greater MSKI risk. Composite scores were significantly associated with MSKI, although the area under the curve (AUC) demonstrated poor discrimination (AUC = .55–.57). When supervised machine learning algorithms were trained on the normalized component variables and cMSKI variables, the overall training models performed well, but when the training models were tested on the testing data the models classified MSKI “by chance” (testing AUC avg = .55–.57) across all models. Composite scores and component population specific models were poor predictors of MSKI in candidates. While cMSKI, SPARTA™, and DARI™ models performed similarly, this study does not dismiss the use of commercial technologies but questions the utility of a singular screening task to predict MSKI over 10 weeks. Further investigations should evaluate occupation specific screening, serial measurements, and/or load exposure for creating MSKI risk models.

show abstract

“…We applaud Bullock and colleagues for their steadfast efforts to evaluate the methodological performance of various studies using traditional or cutting-edge modeling techniques and to support the transparency of creating, disclosing, and extrapolating multivariate prediction models for diagnosis and prognosis. 3,4,7 Additionally, we have learned about the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) statement with its explanations and elaborations. 6,15 We are grateful for the excellent suggestions of Bullock and his colleagues based on the TRIPOD guidelines.…”

Section: Authors' Responsementioning

confidence: 99%

Predicting the Objective and Subjective Clinical Outcomes of Anterior Cruciate Ligament Reconstruction: A Machine Learning Analysis of 432 Patients: Response

Ye¹,

Zhang²,

Wu³

et al. 2023

Am J Sports Med

View full text Add to dashboard Cite

Black Box Prediction Methods in Sports Medicine Deserve a Red Card for Reckless Practice: A Change of Tactics is Needed to Advance Athlete Care

Cited by 30 publications

References 53 publications

Estimating Ground Reaction Forces from Two-Dimensional Pose Data: A Biomechanics-Based Comparison of AlphaPose, BlazePose, and OpenPose

Estimating Ground Reaction Forces from Two-Dimensional Pose Data: A Biomechanics-Based Comparison of AlphaPose, BlazePose, and OpenPose

Predictive utility of commercial grade technologies for assessing musculoskeletal injury risk in US Marine Corps Officer candidates

Predicting the Objective and Subjective Clinical Outcomes of Anterior Cruciate Ligament Reconstruction: A Machine Learning Analysis of 432 Patients: Response

Contact Info

Product

Resources

About