Automated Detection of Pediatric Slipped Capital Femoral Epiphysis: A Deep Learning Approach Using Anatomically...

Area of Science:

• Orthopedics
• Radiology
• Artificial Intelligence

Background:

• Slipped capital femoral epiphysis (SCFE) is a common adolescent hip disorder.
• Subtle radiographic findings often lead to delayed diagnosis and complications.
• Traditional diagnostic methods have limitations in accuracy and consistency.

Purpose of the Study:

• To develop and validate an attention-guided deep learning model for automated SCFE detection.
• To improve diagnostic accuracy and reduce delays in identifying SCFE on pediatric pelvic radiographs.

Main Methods:

• A two-stage deep learning model was created using U-Net++ and EfficientNet B1.
• The model focused on an anatomically defined region of interest (ROF) for classification.
• Performance was evaluated using AUC, accuracy, sensitivity, and specificity on a test set.

Main Results:

• The model achieved an AUC of 0.893, with 91.4% accuracy, 93.3% sensitivity, and 90.0% specificity.
• Gradient-weighted Class Activation Mapping (Grad-CAM) confirmed the model's focus on relevant anatomical regions.
• The model demonstrated high diagnostic performance in identifying SCFE.

Conclusions:

• An attention-guided deep learning model effectively detects SCFE with high accuracy.
• The model's interpretability enhances its clinical utility as a decision-support tool.
• This automated system has the potential to minimize diagnostic delays and improve SCFE detection consistency.