Two -stage contrastive learning framework for vertebral compression fracture screening in frontal chest X-ray

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Vertebral compression fractures (VCFs) are common osteoporosis indicators, often underdiagnosed due to reliance on limited initial examinations.
• Current VCF diagnosis involves multiple steps, posing challenges for timely detection, especially without clear symptoms.

Purpose of the Study:

• To develop and evaluate an AI-driven framework for VCF screening using initial frontal chest X-rays (CXRs).
• To improve the accuracy and efficiency of VCF detection, aiding early osteoporosis diagnosis.

Main Methods:

• A two-stage AI framework was developed: vertebral segmentation and a TADC-Net for VCF recognition.
• The TADC-Net incorporates Triplet Aggregation for feature extraction and Dual Contrastive Loss to address inter-class similarity.

Main Results:

• The proposed method demonstrated superior quantitative and qualitative performance across three multi-center datasets.
• Comparative analysis showed the AI enhanced radiologist accuracy and sensitivity in identifying VCFs on CXRs.

Conclusions:

• The AI framework shows significant potential for early VCF screening on routine CXRs.
• This approach can assist radiologists, leading to improved VCF detection rates and earlier osteoporosis management.