From landmark uncertainty to clinical confidence: screening for developmental dysplasia of the hip

Astratt

Clinical image annotation-based assessment methods often rely on anatomical landmarks to derive landmark-based measurements (angle or distance) that underpin screening, diagnosis, and treatment decisions. In screening pathways, landmark-based measurements typically use predefined thresholds to classify patients as healthy or unhealthy. Natural variability exists among clinicians in landmark placement and, consequently, in the derived measurement and screening decisions, reflecting genuine uncertainty in these assessments. We propose a method to estimate uncertainty associated with landmark-based measurements. By defining this measurement uncertainty, we derive a confidence estimate for the clinical screening task to provide context for the measured uncertainty. While the approach is applicable to screening in general, we demonstrate its utility in detecting developmental dysplasia of the hip in neonates using two standard clinical landmark-based measurements: the Graf α angle and the femoral head coverage percentage. We show a strong correlation between multiple clinicians’ confidence and model’s confidence. This indicates that the model and the clinicians identify the same cases that are difficult or borderline. To our knowledge, this work presents the first method to evaluate a model’s confidence when predicting clinically relevant landmark-based measurements. This confidence estimate can establish trust in automated models, facilitating wider clinical adoption, which, in turn, can widen access to screening.