An AI-based pipeline for osteoporosis/osteopenia prediction using hip radiographs

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Bone Health Diagnostics

Background:

• Osteoporosis and osteopenia pose significant public health challenges, often detected late.
• Current diagnostic methods like DXA scans can be inaccessible or costly.
• There is a need for accessible tools to identify at-risk individuals from routine imaging.

Purpose of the Study:

• To develop and validate an AI tool for diagnosing osteoporosis/osteopenia using hip radiographs.
• To classify femur images into risk categories for improved patient management.
• To enhance early detection and preventive strategies for bone density loss.

Main Methods:

• AI model trained on 3227 single femur images from 2691 hip radiographs.
• Preprocessing included image splitting, prosthesis identification, and femur cropping.
• Model validated on separate test (826 images) and evaluation (313 images) sets.

Main Results:

• High accuracy in preprocessing: 99.0% (femur classification), 99.3% (prosthesis detection), 99.2% (femur cropping).
• Prediction model achieved AUC of 86.6% (test set) and 81.0% (evaluation set) for osteoporosis/osteopenia detection.
• The AI tool demonstrated robust performance in identifying risk categories.

Conclusions:

• The AI pipeline shows significant potential for predicting osteoporosis/osteopenia from hip radiographs.
• This tool can aid in DXA scan triage and incidental findings of bone density loss.
• The AI can support clinical decision-making, particularly in resource-limited settings.