Convolutional Neural Network for Second Metacarpal Radiographic Osteoporosis Screening

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Osteoporosis and osteopenia are prevalent conditions leading to fragility fractures.
• Accurate and accessible screening methods are crucial for early detection and intervention.

Purpose of the Study:

• To evaluate a computer learning system's ability to classify osteoporosis using hand radiographs.
• To assess the efficacy of using the second metacarpal cortical percentage as a predictor for osteoporosis.

Main Methods:

• Utilized a dataset of 4,000 posteroanterior hand radiographs.
• Employed convolutional neural networks (CNNs) for image standardization (laterality and vertical alignment correction) and segmentation.
• Developed a full pipeline integrating these components to predict osteoporosis based on the second metacarpal cortical percentage.

Main Results:

• The full pipeline achieved an overall accuracy of 93.9% in classifying osteoporosis.
• CNNs demonstrated high accuracy in laterality (99.62%) and segmentation (94.8%) tasks.
• The system achieved a sensitivity of 82.4% and specificity of 95.7% on the testing dataset.

Conclusions:

• A series of CNNs can accurately identify osteoporosis from non-osteoporosis in hand radiographs.
• The developed CNN system can perform image adjustments for laterality and vertical alignment.
• This AI approach shows promise as an adjunct to dual-energy x-ray absorptiometry (DXA) scans for osteoporosis screening.