Deep learning application of vertebral compression fracture detection using mask R-CNN.

Area of Science:

• Orthopedic imaging
• Artificial intelligence in medicine
• Radiology

Background:

• Vertebral compression fractures (VCFs) impact the thoracolumbar spine, often due to osteoporosis or trauma.
• Early VCF diagnosis is crucial to prevent patient complications.
• Plain radiographs are the standard for VCF assessment.

Purpose of the Study:

• To develop and evaluate a deep learning model for detecting VCFs.
• To assess the model's utility as a primary care tool in orthopedics.

Main Methods:

• A dataset of 487 lateral radiographs with 598 VCFs (L1-T11) was compiled.
• The Mask R-CNN model was trained for VCF detection and compared with Cascade Mask R-CNN, YOLOACT, and YOLOv5.
• Instance segmentation was used to locate fractures pixel-wise.

Main Results:

• Mask R-CNN achieved the highest mean average precision (0.58) for VCF detection.
• The model demonstrated high sensitivity, specificity, and accuracy, indicating precise fracture identification.
• Pixel-wise localization of fractures was achieved.

Conclusions:

• The developed Mask R-CNN model shows significant potential for VCF detection from radiographs.
• This AI tool can assist orthopedic primary care physicians in initial VCF diagnosis.
• Accurate and timely VCF detection can improve patient outcomes.