BONE-Net: A novel hybrid deep-learning model for effective osteoporosis detection

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Bone Health

Background:

• Osteoporosis is a widespread condition causing reduced bone density and increased fracture risk, particularly in aging populations.
• Early detection of osteoporosis is crucial for effective intervention, reducing morbidity, mortality, and healthcare expenses.
• Current diagnostic methods may lack the efficiency and accuracy needed for widespread early screening.

Purpose of the Study:

• To develop and validate an advanced deep learning model for enhanced accuracy and efficiency in osteoporosis detection.
• To leverage hybrid deep learning architectures for comprehensive analysis of knee X-ray images.
• To improve early diagnosis of osteoporosis for better patient management.

Main Methods:

• Integration of features from DenseNet169 and Vision Transformer (ViT) pre-trained models.
• Implementation of a custom Attention Model (AM) for capturing spatial and channel-specific image information.
• Classification of knee X-ray images into osteoporotic or normal categories using a fully connected neural network.

Main Results:

• The proposed deep learning model achieved a high accuracy of 0.8611 on unseen test data.
• Demonstrated superior performance compared to existing methods with a specificity of 0.9474 and precision of 0.9286.
• Effectively combined convolutional and transformer-based features for robust bone characterization.

Conclusions:

• The developed deep learning approach significantly enhances osteoporosis detection accuracy from knee X-rays.
• This methodology shows strong potential for supporting early diagnosis and timely intervention in osteoporosis management.
• The findings underscore the value of hybrid AI models in improving bone health diagnostics and patient outcomes.