Osteoporosis diagnosis in knee X-rays by transfer learning based on convolution neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Osteoporosis significantly degrades bone quality, leading to fractures, particularly in postmenopausal women and the elderly.
• Current diagnostic methods for osteoporosis are costly, necessitating research into more cost-effective and accessible systems.
• Manual interpretation of X-rays for osteoporosis is challenging due to difficulties in feature extraction and classifier selection.

Purpose of the Study:

• To develop a cost-effective deep learning system for early osteoporosis detection using X-ray imaging.
• To create and validate a dataset of 381 knee X-rays for osteoporosis classification.
• To evaluate the efficacy of transfer learning with various Convolutional Neural Network (CNN) architectures for diagnosing osteoporosis.

Main Methods:

• Utilized transfer learning with deep learning CNNs (AlexNet, VggNet-16, ResNet, VggNet-19) for image analysis.
• Classified knee X-ray images into normal, osteopenia, and osteoporosis categories.
• Compared the performance of pretrained CNNs against non-pretrained networks on a validated dataset.

Main Results:

• Achieved a maximum accuracy of 91.1% using the pretrained AlexNet architecture.
• The pretrained AlexNet model demonstrated a lower error rate (0.09) and validation loss (0.54) compared to non-pretrained networks.
• Non-pretrained networks yielded an accuracy of 79% with a higher error rate (0.21).

Conclusions:

• Deep learning systems, particularly those employing transfer learning, show significant promise for the early detection of osteoporosis.
• The proposed CNN-based approach offers a viable, cost-effective tool to assist clinicians in diagnosing osteoporosis and mitigating fracture risks.