How Well can Machine Learning Models Classify Individuals with and without Non Specific Chronic Neck Pain based on Cervical Movements during Protraction and Retraction?

Background The human cervical spine, vital for supporting head movements, is susceptible to degenerative changes, especially non-specific chronic neck pain (NSCNP). Cervical protraction and retraction, which are key components of cervical spine motion, have been studied to assess their role in NSCNP. However, the existing research lacks quantitative assessments and explores nonlinear relationships. Purpose This study explored the relationship between cervical movements during protraction and retraction and NSCNP using machine learning (ML) techniques for classification. Study design Cross sectional study. Methods This study included 277 non-NSCNP and 463 NSCNP office workers. Data were collected from the musculoskeletal screening tests. Two-dimensional video analysis was used to track markers during cervical protraction and retraction. The head tilt angle (HTA), craniovertebral angle (CVA), head excursion angle (HEA), and protraction/retraction distances were measured. Six ML algorithms (random forest, neural network, decision tree, gradient boosting, logistic regression, and support vector machine) were employed to classify individuals with and without the NSCNP. The model performance was evaluated using the area under the curve (AUC), accuracy, recall, precision, and F1 score. Results Random forest performed best, with a test AUC of 0.800, followed by decision trees (0.790), and gradient boosting (0.701). Logistic regression and support vector machine had the lowest performance. CVA during retraction, CVA and HEA during protraction were significant predictors of NSCNP in the random forest model, indicating the importance of cervical retraction and protraction kinematics. Conclusions ML models can enhance our understanding of NSCNP and the role of cervical movements. These findings offer potential targets for assessment and intervention in NSCNP cases, and suggest the clinical utility of random forests for classification. Further research is needed to explore these relationships in diverse populations and investigate the underlying mechanisms.