Osteoporosis prediction from hand X-ray images using segmentation-for-classification and self-supervised learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Osteoporosis is a widespread metabolic bone disease often undiagnosed due to limited access to bone mineral density (BMD) testing like Dual-energy X-ray absorptiometry (DXA).
• Research is exploring alternative, more accessible screening methods using peripheral skeletal sites.
• Hand and wrist X-rays offer a cost-effective and widely available imaging option for potential osteoporosis assessment.

Purpose of the Study:

• To develop and evaluate a method for predicting osteoporosis using hand and wrist X-ray images.
• To address the challenge of limited access to DXA scans for osteoporosis diagnosis.
• To establish the association between peripheral X-ray imaging and DXA-based osteoporosis diagnoses.

Main Methods:

• Utilized an image segmentation model with probabilistic U-Net decoders to segment ulna, radius, and metacarpal bones, capturing predictive uncertainty.
• Formulated the segmentation task as an optimal transport (OT) problem to handle medical image variability.
• Employed a self-supervised learning (SSL) strategy for pretraining on unlabeled data, followed by supervised fine-tuning for osteoporotic classification.

Main Results:

• The proposed method was evaluated on X-ray images from 192 individuals with confirmed DXA diagnoses.
• The framework successfully combined uncertainty-aware segmentation and self-supervised feature learning.
• Demonstrated a promising vision-based strategy for early osteoporosis detection using peripheral X-rays.

Conclusions:

• The developed framework offers a viable, vision-based approach for early osteoporosis detection.
• Peripheral X-ray imaging, combined with advanced AI techniques, can enhance osteoporosis screening accessibility.
• This method holds potential for improving early diagnosis and management of osteoporosis.