Assessing the Potential of a Deep Learning Tool to Improve Fracture Detection by Radiologists and Emergency...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning (DL) models show promise in medical image analysis.
• Accurate fracture detection in extremity radiographs is crucial for patient care.
• Evaluating AI tools alongside human interpretation is essential.

Purpose of the Study:

• To assess the standalone performance of a DL tool for extremity fracture detection.
• To evaluate the impact of DL assistance on the diagnostic performance of radiologists and emergency physicians.

Main Methods:

• A DL tool was developed and validated on a large dataset of appendicular skeletal radiographs.
• Standalone performance was tested on 2626 de-identified extremity radiographs.
• A multi-reader study compared fracture detection with and without DL aid among 24 physicians.

Main Results:

• The DL tool achieved high standalone accuracy (0.986), sensitivity (0.987), and specificity (0.885).
• DL assistance significantly improved reader accuracy (0.047 increase) and sensitivity (0.865 to 0.955).
• Average reading time decreased by 7.1 seconds (27%), with greater benefits for emergency physicians and non-MSK radiologists.

Conclusions:

• The DL tool exhibits excellent standalone performance in detecting extremity fractures.
• AI-assisted interpretation enhances diagnostic accuracy and efficiency for physicians.
• This technology has the potential to improve the interpretation of extremity radiographs.