Artificial intelligence (AI) vs. human in hip fracture detection

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Hip fractures are a common and serious injury, particularly in older adults.
• Accurate and timely diagnosis is crucial for effective treatment and patient outcomes.
• Current diagnostic methods rely on radiograph interpretation, which can be subject to human error and variability.

Purpose of the Study:

• To evaluate the diagnostic accuracy and sensitivity of the YOLOv4-tiny AI model for detecting and classifying hip fractures.
• To compare the performance of the AI model against human physicians in identifying hip fractures.

Main Methods:

• A retrospective study utilized 1000 hip and pelvic radiographs, divided into training and testing sets.
• A YOLOv4-tiny deep convolutional neural network AI model was trained on augmented images with labeled hip fracture types.
• The AI model's performance was evaluated on a separate testing set and compared with the assessments of human doctors.

Main Results:

• The YOLOv4-tiny AI model achieved a sensitivity of 96.2%, specificity of 94.6%, and overall accuracy of 95% in detecting hip fractures.
• The AI model's detection sensitivity was superior to general practitioners and first-year residents.
• The AI model's performance was equivalent to that of specialist doctors in classifying hip fractures.

Conclusions:

• The YOLOv4-tiny AI model demonstrates high accuracy and sensitivity in detecting and classifying hip fractures.
• The AI model's diagnostic performance is comparable to that of experienced radiologists and orthopedists.
• This AI tool holds significant potential for improving the efficiency and accuracy of hip fracture diagnosis.