Evaluating Canine Hip Analysis: NANet Model for Norberg Angle Prediction

The Norberg angle (NA) plays a crucial role in evaluating hip joint conformation in canines by quantifying femoral headsubluxation within the hip joint. It is a key metric for assessing hip joint quality and diagnosing canine hip dysplasia (CHD),a multifactorial orthopedic disease influenced by both genetic and environmental factors. While contemporary tools offerautomated NA quantification, they still rely on substantial manual labeling and verification by trained veterinarians. Rather thanfocusing on workflow efficiency, the objective of this study is to develop a fully automated system capable of predicting theNA directly from radiographs, thereby reducing dependence on manual annotation and enabling consistent and reproducibleassessment. To address the limited availability of diverse, high-quality annotated radiographs, our previous work incorporated diffusion-basedgenerative models to augment the training dataset from 1,168 images to 2,693 images. Training with this expanded datasetresulted in substantial performance gains, with an average improvement of 35.3% across evaluation metrics. Our proposedNANet model outperformed state-of-the-art alternatives, demonstrating superior predictive accuracy. In particular, the predictedNorberg angle fell within 5 degrees of the reference measurement in 81.30% of test cases, approaching or surpassing theconsistency observed among human annotators.