Machine learning prediction of combat basic training injury from 3D body shape images

Area of Science:

• Sports Medicine
• Military Health
• Biotechnology

Background:

• Athletes and military personnel face high risks of disabling injuries from intense physical activity.
• Predicting injury susceptibility is crucial, particularly in the military, to prevent recruit discharge.

Purpose of the Study:

• To investigate the use of 3D body imaging and anthropometric measurements to predict injury risk in military recruits.
• To compare the predictive performance of different machine learning models for injury-related discharge.

Main Methods:

• Utilized 3D body imaging scans from 17,680 US Army basic training recruits (ages 17-21).
• Extracted 161 anthropometric measurements to build predictive models.
• Employed logistic regression, random forest, and artificial neural network (ANN) models, comparing performance via ROC curve AUC.

Main Results:

• The artificial neural network (ANN) model demonstrated superior predictive accuracy (AUC = 0.70) compared to logistic regression (AUC = 0.67) and random forest (AUC = 0.65).
• The ANN model's performance was statistically significant, with a confidence interval of [0.68, 0.72].

Conclusions:

• Three-dimensional body shape profiles derived from scans can predict dischargeable physical injuries in military personnel.
• Integrating the ANN model into 3D scanners offers real-time risk prediction, enabling proactive injury prevention strategies.