Advanced Deep Learning Techniques Applied to Automated Femoral Neck Fracture Detection and Classification

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Femoral neck fractures are common injuries requiring accurate diagnosis.
• Current diagnostic methods can be limited, necessitating improved imaging analysis.
• Deep learning offers potential for enhanced fracture detection and classification.

Purpose of the Study:

• To develop and evaluate a deep learning model for diagnosing and classifying femoral neck fractures.
• To assess the impact of advanced data augmentation techniques on model performance.

Main Methods:

• A retrospective study utilized 1063 hip radiographs from 550 patients.
• A deep neural network, including a CNN, was trained on real and augmented images.
• Advanced augmentation included Generative Adversarial Networks (GANs) and Digitally Reconstructed Radiographs (DRRs).

Main Results:

• The two-class model (fracture vs. no fracture) achieved an AUC of 0.92.
• The three-class model (Garden I/II, Garden III/IV, normal) achieved an AUC of 0.96.
• Advanced augmentation significantly improved AUC from 0.80 to 0.91 (DRR) and 0.87 (GAN).

Conclusions:

• Deep learning models, enhanced by GANs and DRRs, can accurately diagnose and classify femoral neck fractures.
• Advanced data augmentation is crucial for improving the performance of AI in fracture detection.
• This approach holds promise for a more precise and efficient diagnostic tool.