A machine learning approach to concussive group classification using discrete outcome measures from a low-cost...

Area of Science:

• Neurology
• Sports Medicine
• Biomechanical Engineering

Background:

• Objective neuromotor assessment is crucial for concussion diagnosis.
• Current tools are often unidimensional and lack clinical feasibility.
• Developing practical, objective measures is a significant clinical need.

Purpose of the Study:

• To evaluate the classification accuracy of a machine learning model for concussion diagnosis.
• To utilize features from a clinically feasible movement-based assessment system (MPASS).
• To differentiate between athletes with and without concussion using objective data.

Main Methods:

• Employed the Mizzou Point-of-care Assessment System (MPASS) for data collection.
• Collected kinematic and kinetic data during static balance, gait, and reaction time tasks.
• Utilized an XGBoost machine learning model with five-fold cross-validation.

Main Results:

• The machine learning model achieved 82.5% accuracy, 75% sensitivity, and 90% specificity.
• High positive predictive value (88.2%) and negative predictive value (78.3%) were observed.
• Movement-based features from MPASS demonstrated strong classification performance.

Conclusions:

• Movement-based features from a low-cost system show potential for objective concussion diagnosis.
• MPASS offers a clinically feasible solution to enhance concussion assessment objectivity.
• This approach could significantly improve diagnostic decision-making in sports medicine.