Classification Model for Discriminating Trunk Fatigue During Running.

PURPOSE

Fatigue is often associated with increased injury risk. Many studies have focused on fatigue in the lower extremity muscles brought on by running, yet few have examined the relationship between fatigue of the core musculature and associated changes in running gait. To investigate the relationship between trunk fatigue and running dynamics, this study had two goals: (1) to use machine learning to determine which gait parameters are most associated with trunk fatigue; and (2) to develop a machine learning algorithm that uses those parameters to classify individuals with trunk fatigue. We hypothesized that we could effectively differentiate between the non-fatigued and fatigued states using machine learning models derived from running gait parameters.

METHODS

Seventy-two individuals performed a trunk fatigue protocol. Lower extremity running biomechanics were collected pre- and post- the trunk fatigue protocol using an instrumented treadmill and 10-camera motion capture system.The fatiguing protocol involved executing a series of trunk fatiguing exercises until established fatigue criteria were reached. Gait variables extracted from the non-fatigued and fatigued states served as model inputs to aid in the development of the machine learning model that would distinguish between non-fatigued and fatigued running.

RESULTS

The machine learning protocol determined three variables - stance time, maximum propulsive GRF and maximum braking GRF - to be the best discriminators between non-fatigued and fatigued running. The SVM with Bagging was the best performing model that discriminated between non-fatigued and fatigued running with an accuracy of 82%, precision of 77%, recall of 90%, and area under the receiver operating curve of 0.91.

CONCLUSION

The machine learning model was effective in classifying between non-fatigued and fatigued running using three gait parameters extracted from GRF waveforms. The ability to classify fatigue using these easy to measure GRF derived parameters enhances the ability for the model to be integrated into wearable technology and the clinical setting to aid in the detection of fatigue and potentially injury, as fatigue is often a precursor to injury.Clinical Relevance- This model has the potential to be implemented in a clinical setting to determine the onset of trunk fatigue through basic gait analysis, involving only the ground reaction forces. This model would be aimed toward injury prevention since fatigue is linked to increased risk of injury.