Automated detection and classification of the proximal humerus fracture by using deep learning algorithm.

Area of Science:

• Radiology
• Artificial Intelligence in Medicine
• Orthopedic Surgery

Background:

• Proximal humerus fractures are common injuries requiring accurate diagnosis.
• Radiographic assessment is crucial for fracture detection and classification.
• The role of artificial intelligence (AI) in interpreting medical images is rapidly evolving.

Purpose of the Study:

• To evaluate an AI deep learning algorithm's ability to detect and classify proximal humerus fractures.
• To compare the AI's performance against general physicians, orthopedists, and shoulder specialists.

Main Methods:

• A deep convolutional neural network (CNN) was trained on 1,891 anteroposterior shoulder radiographs.
• The dataset included normal shoulders and four types of proximal humerus fractures.
• AI performance was measured using top-1 accuracy, AUC, sensitivity, specificity, and Youden index.

Main Results:

• The CNN achieved 96% top-1 accuracy in distinguishing fractures from normal shoulders (AUC 1.00).
• For fracture classification, the CNN demonstrated 65-86% top-1 accuracy and high AUC (0.90-0.98).
• AI performance was superior to general physicians and comparable to shoulder specialists, especially for complex fractures.

Conclusions:

• AI can accurately detect and classify proximal humerus fractures on plain radiographs.
• AI shows potential for assisting in orthopedic assessments, particularly for complex fracture patterns.
• Further clinical studies are needed to confirm AI's feasibility and impact on patient care.