YOLO-Based Hip Fracture Detection with Hyperparameter Tuning

Abstract

Hip fractures are a significant global health issue that can occur across all ages and genders. The condition is a complex disorder that limits human visual perception and causes a certain margin of error during diagnosis through X-ray inspection, with high misdiagnosis rates in worst-case scenarios reaching up to 30%. Despite the advancement in the current artificial intelligence (AI) model for hip fracture detection, small prediction errors still persist. In order to address these issues, the study proposes an AI model for hip fracture detection through an optimization strategy with hyperparameter tuning. The strategy combines grid search and genetic algorithm based on the state-of-the-art YOLO algorithm. Consequently, the results demonstrate that meticulous tuning of 33 hyperparameters significantly affects model performance. The tuning strategy contributes to a notable outcome, the optimal model achieves an average accuracy of 0.968 on a test set comprising 573 images from internal clinical sources. A significant improvement over the suboptimal model achieves an average accuracy of 0.910. Furthermore, the proposed method demonstrates the potential of hyperparameter tuning not only to improve diagnostic accuracy but also to enhance clinical decision-making in hip fracture diagnosis.

Ähnliche Arbeiten