How Can a Deep Learning Algorithm Improve Fracture Detection on X-rays in the Emergency Room?

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Emergency imaging, particularly X-rays for trauma, is increasing.
• Deep learning (DL) offers potential for improving fracture detection by radiologists and emergency room (ER) physicians.

Purpose of the Study:

• To evaluate the performance of a DL algorithm for detecting traumatic limb fractures on conventional X-rays in an ER setting.
• To assess the algorithm's utility without requiring local data training.

Main Methods:

• A DL algorithm for appendicular skeleton fracture detection was tested on 125 patients with limb trauma at a specific ER.
• X-rays were analyzed by the DL algorithm and compared against radiologist annotations.

Main Results:

• The algorithm achieved a sensitivity of 96% (24/25 fractures detected) and a specificity of 86% (14/100 non-fractures misclassified).
• The negative predictive value was high at 98.85%.

Conclusions:

• DL algorithms demonstrate potential as valuable diagnostic tools for fracture detection in emergency departments.
• These algorithms could also serve as aids in training junior radiologists for fracture identification.