From keypoints to classifications: A SHAP-driven TabPFN framework for forward head posture screening
Year: 2026
Type: article
Study design: Observational Study
Abstract: Forward Head Posture (FHP) is a prevalent postural abnormality among adolescents, linked to musculoskeletal imbalances and long-term health risks. To investigate its prevalence and develop a research-stage screening framework, we conducted a large-scale population study involving 1,577 Chinese adolescents, recruited for the screening and assessment of FHP. Building on this cohort, we propose an end-to-end predictive framework that integrates human pose estimation with tabular machine learning. RTMPose was used to extract key anatomical landmarks from side-view images, followed by the computational derivation of seven predefined, anatomy-informed posture features. A TabPFN classifier trained on these features achieved an F1 score of 82.46% and an AUC of 0.837. Permutation-based SHAP analysis identified the horizontal angle between the right eye and shoulder as the strongest predictive contributor. A SHAP dependence plot also suggested a potential nonlinear dependence between features beyond that captured by pairwise correlation. These associations describe the behavior of the fitted model and do not establish causal biomechanical effects. These findings support the feasibility of a non-invasive and interpretable image-based framework for FHP screening. The framework remains a research-stage classifier, and prospective clinical validation and implementation evaluation are required before its use in school-based or public-health practice.
Source: PLoS ONE
Language: English
Subfield: Human Factors and Ergonomics
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $2,382
OpenAlex ID: W7236079967
