Detection of Sacral Fractures on Radiographs Using Artificial Intelligence

Area of Science:

• Orthopaedic radiology
• Medical artificial intelligence
• Diagnostic imaging

Background:

• Sacral fractures are challenging to diagnose using standard radiographs.
• Advanced imaging like CT and MRI improve detection but are not always feasible.
• Artificial intelligence (AI) has demonstrated accuracy comparable to specialists in detecting other orthopaedic fractures.

Purpose of the Study:

• To investigate the efficacy of AI in detecting sacral fractures on radiographs.
• To develop and validate an AI model for accurate sacral fracture detection.
• To compare the AI model's performance against orthopaedic surgeons.

Main Methods:

• Trained eight convolutional neural network (CNN) models on 2,038 radiograph images labeled using CT scans.
• Evaluated model performance on a separate test set of 200 images.
• Compared the detection accuracy of the top-performing CNNs with that of four experienced orthopaedic surgeons.

Main Results:

• The top three CNN models (InceptionV3, Xception, Inception ResNetV2) achieved high areas under the curve (0.989, 0.987, 0.984, respectively).
• These AI models demonstrated significantly higher detection rates for sacral fractures compared to orthopaedic surgeons.
• The AI models showed superior ability in assessing fracture presence on radiographs.

Conclusions:

• AI models can significantly enhance the detection of sacral fractures on radiographs.
• AI may offer superior diagnostic accuracy for sacral fractures compared to human specialists.
• Further development in AI processing could improve diagnostic capabilities in challenging fracture cases.