1. ARTIFICIAL INTELLIGENCE APPLIED TO Q-ANGLE MEASUREMENT: PRELIMINARY RESULTS ON AN ALGORITHM BASED ON BOUNDING BOX.
- Author
-
ALBERTI, L., LOIACONO, D., FAVARO, A., BONDI, A., BERTOLINO, L., and BONANZINGA, T.
- Abstract
Objective: The measurement of Q-angle lacks standardization, with the potential risk of failing to detect clinically relevant findings and misjudgments on types of interventions needed in patellofemoral pathologies. The aim of this study was to implement a machine-learning model for the accurate and reliable measurement of Q-angle directly from radiographs. Materials and Methods: A total of 187 radiographs (of which approximately 50 belonging to pediatric patients) were manually annotated for the anterior superior iliac spine (ASIS), the center of the patella, and the tibial tuberosity and then enhanced by means of different image preprocessing techniques. Eighty-five percent of X-rays were used for training and 15% for testing and validation. Prediction performance was tested using the full-leg radiograph (WLR) and the bounding boxes (BB) models in terms of mean squared error compared to the ground truth (key points and Q-angles determined by the operator). Results: Overall, mean prediction errors were the smallest for the patella and ASIS and the highest for the tibial tuberosity. The BB model yielded smaller mean errors in the prediction of all points and Q-angle compared to WLR (except for tibial tuberosity, which was comparable) and showed the highest agreement with ground truth, with no bias for Q-angle. Conclusions: This proof-of-concept study supports the use of an AI-driven automatic algorithm to identify the key points for measuring Q angle directly from the patient's radiographs. Results demonstrate the highest reliability with the bounding box approach and the algorithm's ability to correctly identify key points across a heterogeneous patient population. [ABSTRACT FROM AUTHOR]
- Published
- 2025